RECENT ADVANCES in BIOLOGY, MEDICAL PHYSICS, MEDICAL ... · Prof. Wasfy B Mikhael (IEEE Fellow, University of Central Florida Orlando,USA) Prof. Sunil Das (IEEE Fellow, University

RECENT ADVANCES in BIOLOGY, MEDICAL PHYSICS, MEDICAL

CHEMISTRY, BIOCHEMISTRY and BIOMEDICAL ENGINEERING

Proceedings of the 2013 International Conference on Biology, Medical Physics, Medical Chemistry, Biochemistry and Biomedical Engineering

(BIOMED 2013)

Venice, Italy September 28‐30, 2013

RECENT ADVANCES in BIOLOGY, MEDICAL PHYSICS, MEDICAL CHEMISTRY, BIOCHEMISTRY and BIOMEDICAL ENGINEERING

Proceedings of the 2013 International Conference on Biology, Medical Physics, Medical Chemistry, Biochemistry and Biomedical Engineering (BIOMED 2013)


Copyright © 2013, by the editors

All the copyright of the present book belongs to the editors. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior written permission of the editors.

All papers of the present volume were peer reviewed by no less than two independent reviewers. Acceptance was granted when both reviewers' recommendations were positive.

ISBN: 978‐1‐61804‐213‐2

RECENT ADVANCES in BIOLOGY, MEDICAL PHYSICS, MEDICAL

CHEMISTRY, BIOCHEMISTRY and BIOMEDICAL ENGINEERING


(BIOMED 2013)


Organizing Committee General Chairs (EDITORS)

Professor Charles A. Long Professor Emeritus University of Wisconsin Stevens Point, Wisconsin, USA

Photios A. Anninos Professor Emeritus Democritus University of Thrace. Alexandroupolis, Greece

Senior Program Chair

Professor Philippe Dondon ENSEIRB Rue A Schweitzer 33400 Talence France

Program Chairs

Professor Filippo Neri Dipartimento di Informatica e Sistemistica University of Naples "Federico II" Naples, Italy

Prof. Constantin Udriste, University Politehnica of Bucharest, Bucharest, Romania

Professor Sandra Sendra Instituto de Inv. para la Gestión Integrada de Zonas Costeras (IGIC) Universidad Politécnica de Valencia Spain

Tutorials Chair

Professor Pradip Majumdar Department of Mechanical Engineering Northern Illinois University Dekalb, Illinois, USA

Special Session Chair

Professor Pavel Varacha Tomas Bata University in Zlin Faculty of Applied Informatics Department of Informatics and Artificial Intelligence Zlin, Czech Republic

Workshops Chair

Professor Ryszard S. Choras Institute of Telecommunications University of Technology & Life Sciences Bydgoszcz, Poland

Local Organizing Chair

Assistant Professor Klimis Ntalianis, Tech. Educ. Inst. of Athens (TEI), Athens, Greece

Publication Chair

Professor Gen Qi Xu Department of Mathematics Tianjin University Tianjin, China

Publicity Committee

Professor Reinhard Neck Department of Economics Klagenfurt University Klagenfurt, Austria

Professor Myriam Lazard Institut Superieur d' Ingenierie de la Conception Saint Die, France

International Liaisons

Professor Ka‐Lok Ng Department of Bioinformatics Asia University Taichung, Taiwan

Professor Olga Martin Applied Sciences Faculty Politehnica University of Bucharest Romania

Professor Vincenzo Niola Departement of Mechanical Engineering for Energetics University of Naples "Federico II" Naples, Italy

Professor Eduardo Mario Dias Electrical Energy and Automation Engineering Department Escola Politecnica da Universidade de Sao Paulo Brazil

Steering Committee Professor Aida Bulucea, University of Craiova, Romania Professor Zoran Bojkovic, Univ. of Belgrade, Serbia Professor Metin Demiralp, Istanbul Technical University, Turkey Professor Imre Rudas, Obuda University, Budapest, Hungary

Program Committee Prof. Lotfi Zadeh (IEEE Fellow,University of Berkeley, USA) Prof. Leon Chua (IEEE Fellow,University of Berkeley, USA) Prof. Michio Sugeno (RIKEN Brain Science Institute (RIKEN BSI), Japan) Prof. Dimitri Bertsekas (IEEE Fellow, MIT, USA) Prof. Demetri Terzopoulos (IEEE Fellow, ACM Fellow, UCLA, USA) Prof. Georgios B. Giannakis (IEEE Fellow, University of Minnesota, USA) Prof. George Vachtsevanos (Georgia Institute of Technology, USA) Prof. Abraham Bers (IEEE Fellow, MIT, USA) Prof. David Staelin (IEEE Fellow, MIT, USA) Prof. Brian Barsky (IEEE Fellow, University of Berkeley, USA) Prof. Aggelos Katsaggelos (IEEE Fellow, Northwestern University, USA) Prof. Josef Sifakis (Turing Award 2007, CNRS/Verimag, France) Prof. Hisashi Kobayashi (Princeton University, USA) Prof. Kinshuk (Fellow IEEE, Massey Univ. New Zeland), Prof. Leonid Kazovsky (Stanford University, USA) Prof. Narsingh Deo (IEEE Fellow, ACM Fellow, University of Central Florida, USA) Prof. Kamisetty Rao (Fellow IEEE, Univ. of Texas at Arlington,USA) Prof. Anastassios Venetsanopoulos (Fellow IEEE, University of Toronto, Canada) Prof. Steven Collicott (Purdue University, West Lafayette, IN, USA) Prof. Nikolaos Paragios (Ecole Centrale Paris, France) Prof. Nikolaos G. Bourbakis (IEEE Fellow, Wright State University, USA) Prof. Stamatios Kartalopoulos (IEEE Fellow, University of Oklahoma, USA) Prof. Irwin Sandberg (IEEE Fellow, University of Texas at Austin, USA), Prof. Michael Sebek (IEEE Fellow, Czech Technical University in Prague, Czech Republic) Prof. Hashem Akbari (University of California, Berkeley, USA) Prof. Yuriy S. Shmaliy, (IEEE Fellow, The University of Guanajuato, Mexico) Prof. Lei Xu (IEEE Fellow, Chinese University of Hong Kong, Hong Kong) Prof. Paul E. Dimotakis (California Institute of Technology Pasadena, USA) Prof. M. Pelikan (UMSL, USA) Prof. Patrick Wang (MIT, USA) Prof. Wasfy B Mikhael (IEEE Fellow, University of Central Florida Orlando,USA) Prof. Sunil Das (IEEE Fellow, University of Ottawa, Canada) Prof. Panos Pardalos (University of Florida, USA) Prof. Nikolaos D. Katopodes (University of Michigan, USA) Prof. Bimal K. Bose (Life Fellow of IEEE, University of Tennessee, Knoxville, USA) Prof. Janusz Kacprzyk (IEEE Fellow, Polish Academy of Sciences, Poland) Prof. Sidney Burrus (IEEE Fellow, Rice University, USA) Prof. Biswa N. Datta (IEEE Fellow, Northern Illinois University, USA) Prof. Mihai Putinar (University of California at Santa Barbara, USA) Prof. Wlodzislaw Duch (Nicolaus Copernicus University, Poland) Prof. Tadeusz Kaczorek (IEEE Fellow, Warsaw University of Tehcnology, Poland) Prof. Michael N. Katehakis (Rutgers, The State University of New Jersey, USA) Prof. Pan Agathoklis (Univ. of Victoria, Canada) Prof. P. Demokritou (Harvard University, USA) Prof. P. Razelos (Columbia University, USA) Dr. Subhas C. Misra (Harvard University, USA) Prof. Martin van den Toorn (Delft University of Technology, The Netherlands) Prof. Malcolm J. Crocker (Distinguished University Prof., Auburn University,USA) Prof. S. Dafermos (Brown University, USA) Prof. Urszula Ledzewicz, Southern Illinois University , USA. Prof. Dimitri Kazakos, Dean, (Texas Southern University, USA) Prof. Ronald Yager (Iona College, USA) Prof. Athanassios Manikas (Imperial College, London, UK)

Prof. Keith L. Clark (Imperial College, London, UK) Prof. Argyris Varonides (Univ. of Scranton, USA) Prof. S. Furfari (Direction Generale Energie et Transports, Brussels, EU) Prof. Constantin Udriste, University Politehnica of Bucharest , ROMANIA Dr. Michelle Luke (Univ. Berkeley, USA) Prof. Patrice Brault (Univ. Paris‐sud, France) Dr. Christos E. Vasios (MIT, USA) Prof. Jim Cunningham (Imperial College London, UK) Prof. Philippe Ben‐Abdallah (Ecole Polytechnique de l'Universite de Nantes, France) Prof. Photios Anninos (Medical School of Thrace, Greece) Prof. Ichiro Hagiwara, (Tokyo Institute of Technology, Japan) Prof. Metin Demiralp ( Istanbul Technical University / Turkish Academy of Sciences, Istanbul, Turkey) Prof. Andris Buikis (Latvian Academy of Science. Latvia) Prof. Akshai Aggarwal (University of Windsor, Canada) Prof. George Vachtsevanos (Georgia Institute of Technology, USA) Prof. Ulrich Albrecht (Auburn University, USA) Prof. Imre J. Rudas (Obuda University, Hungary) Prof. Alexey L Sadovski (IEEE Fellow, Texas A&M University, USA) Prof. Amedeo Andreotti (University of Naples, Italy) Prof. Ryszard S. Choras (University of Technology and Life Sciences Bydgoszcz, Poland) Prof. Remi Leandre (Universite de Bourgogne, Dijon, France) Prof. Moustapha Diaby (University of Connecticut, USA) Prof. Brian McCartin (New York University, USA) Prof. Elias C. Aifantis (Aristotle Univ. of Thessaloniki, Greece) Prof. Anastasios Lyrintzis (Purdue University, USA) Prof. Charles Long (Prof. Emeritus University of Wisconsin, USA) Prof. Marvin Goldstein (NASA Glenn Research Center, USA) Prof. Costin Cepisca (University POLITEHNICA of Bucharest, Romania) Prof. Kleanthis Psarris (University of Texas at San Antonio, USA) Prof. Ron Goldman (Rice University, USA) Prof. Ioannis A. Kakadiaris (University of Houston, USA) Prof. Richard Tapia (Rice University, USA) Prof. F.‐K. Benra (University of Duisburg‐Essen, Germany) Prof. Milivoje M. Kostic (Northern Illinois University, USA) Prof. Helmut Jaberg (University of Technology Graz, Austria) Prof. Ardeshir Anjomani (The University of Texas at Arlington, USA) Prof. Heinz Ulbrich (Technical University Munich, Germany) Prof. Reinhard Leithner (Technical University Braunschweig, Germany) Prof. Elbrous M. Jafarov (Istanbul Technical University, Turkey) Prof. M. Ehsani (Texas A&M University, USA) Prof. Sesh Commuri (University of Oklahoma, USA) Prof. Nicolas Galanis (Universite de Sherbrooke, Canada) Prof. S. H. Sohrab (Northwestern University, USA) Prof. Rui J. P. de Figueiredo (University of California, USA) Prof. Valeri Mladenov (Technical University of Sofia, Bulgaria) Prof. Hiroshi Sakaki (Meisei University, Tokyo, Japan) Prof. Zoran S. Bojkovic (Technical University of Belgrade, Serbia) Prof. K. D. Klaes, (Head of the EPS Support Science Team in the MET Division at EUMETSAT, France) Prof. Emira Maljevic (Technical University of Belgrade, Serbia) Prof. Kazuhiko Tsuda (University of Tsukuba, Tokyo, Japan) Prof. Milan Stork (University of West Bohemia , Czech Republic) Prof. C. G. Helmis (University of Athens, Greece) Prof. Lajos Barna (Budapest University of Technology and Economics, Hungary) Prof. Nobuoki Mano (Meisei University, Tokyo, Japan)

Prof. Nobuo Nakajima (The University of Electro‐Communications, Tokyo, Japan) Prof. Victor‐Emil Neagoe (Polytechnic University of Bucharest, Romania) Prof. E. Protonotarios (National Technical University of Athens, Greece) Prof. P. Vanderstraeten (Brussels Institute for Environmental Management, Belgium) Prof. Annaliese Bischoff (University of Massachusetts, Amherst, USA) Prof. Virgil Tiponut (Politehnica University of Timisoara, Romania) Prof. Andrei Kolyshkin (Riga Technical University, Latvia) Prof. Fumiaki Imado (Shinshu University, Japan) Prof. Sotirios G. Ziavras (New Jersey Institute of Technology, USA) Prof. Constantin Volosencu (Politehnica University of Timisoara, Romania) Prof. Marc A. Rosen (University of Ontario Institute of Technology, Canada) Prof. Alexander Zemliak (Puebla Autonomous University, Mexico) Prof. Thomas M. Gatton (National University, San Diego, USA) Prof. Leonardo Pagnotta (University of Calabria, Italy) Prof. Yan Wu (Georgia Southern University, USA) Prof. Daniel N. Riahi (University of Texas‐Pan American, USA) Prof. Alexander Grebennikov (Autonomous University of Puebla, Mexico) Prof. Bennie F. L. Ward (Baylor University, TX, USA) Prof. Guennadi A. Kouzaev (Norwegian University of Science and Technology, Norway) Prof. Eugene Kindler (University of Ostrava, Czech Republic) Prof. Geoff Skinner (The University of Newcastle, Australia) Prof. Hamido Fujita (Iwate Prefectural University(IPU), Japan) Prof. Francesco Muzi (University of L'Aquila, Italy) Prof. Les M. Sztandera (Philadelphia University, USA) Prof. Claudio Rossi (University of Siena, Italy) Prof. Christopher J. Koroneos (Aristotle University of Thessaloniki, Greece) Prof. Sergey B. Leonov (Joint Institute for High Temperature Russian Academy of Science, Russia) Prof. Arpad A. Fay (University of Miskolc, Hungary) Prof. Lili He (San Jose State University, USA) Prof. M. Nasseh Tabrizi (East Carolina University, USA) Prof. Alaa Eldin Fahmy (University Of Calgary, Canada) Prof. Ion Carstea (University of Craiova, Romania) Prof. Gh. Pascovici (University of Koeln, Germany) Prof. Pier Paolo Delsanto (Politecnico of Torino, Italy) Prof. Radu Munteanu (Rector of the Technical University of Cluj‐Napoca, Romania) Prof. Ioan Dumitrache (Politehnica University of Bucharest, Romania) Prof. Corneliu Lazar (Technical University Gh.Asachi Iasi, Romania) Prof. Nicola Pitrone (Universita degli Studi Catania, Italia) Prof. Miquel Salgot (University of Barcelona, Spain) Prof. Amaury A. Caballero (Florida International University, USA) Prof. Maria I. Garcia‐Planas (Universitat Politecnica de Catalunya, Spain) Prof. Petar Popivanov (Bulgarian Academy of Sciences, Bulgaria) Prof. Alexander Gegov (University of Portsmouth, UK) Prof. Lin Feng (Nanyang Technological University, Singapore) Prof. Colin Fyfe (University of the West of Scotland, UK) Prof. Zhaohui Luo (Univ of London, UK) Prof. Mikhail Itskov (RWTH Aachen University, Germany) Prof. George G. Tsypkin (Russian Academy of Sciences, Russia) Prof. Wolfgang Wenzel (Institute for Nanotechnology, Germany) Prof. Weilian Su (Naval Postgraduate School, USA) Prof. Phillip G. Bradford (The University of Alabama, USA) Prof. Ray Hefferlin (Southern Adventist University, TN, USA) Prof. Gabriella Bognar (University of Miskolc, Hungary)

Prof. Hamid Abachi (Monash University, Australia) Prof. Karlheinz Spindler (Fachhochschule Wiesbaden, Germany) Prof. Josef Boercsoek (Universitat Kassel, Germany) Prof. Eyad H. Abed (University of Maryland, Maryland, USA) Prof. F. Castanie (TeSA, Toulouse, France) Prof. Robert K. L. Gay (Nanyang Technological University, Singapore) Prof. Andrzej Ordys (Kingston University, UK) Prof. Harris Catrakis (Univ of California Irvine, USA) Prof. T Bott (The University of Birmingham, UK) Prof. Petr Filip (Institute of Hydrodynamics, Prague, Czech Republic) Prof. T.‐W. Lee (Arizona State University, AZ, USA) Prof. Le Yi Wang (Wayne State University, Detroit, USA) Prof. George Stavrakakis (Technical University of Crete, Greece) Prof. John K. Galiotos (Houston Community College, USA) Prof. M. Petrakis (National Observatory of Athens, Greece) Prof. Philippe Dondon (ENSEIRB, Talence, France) Prof. Dalibor Biolek (Brno University of Technology, Czech Republic) Prof. Oleksander Markovskyy (National Technical University of Ukraine, Ukraine) Prof. Suresh P. Sethi (University of Texas at Dallas, USA) Prof. Hartmut Hillmer(University of Kassel, Germany) Prof. Bram Van Putten (Wageningen University, The Netherlands) Prof. Alexander Iomin (Technion ‐ Israel Institute of Technology, Israel) Prof. Roberto San Jose (Technical University of Madrid, Spain) Prof. Minvydas Ragulskis (Kaunas University of Technology, Lithuania) Prof. Arun Kulkarni (The University of Texas at Tyler, USA) Prof. Joydeep Mitra (New Mexico State University, USA) Prof. Vincenzo Niola (University of Naples Federico II, Italy) Prof. Ion Chryssoverghi (National Technical University of Athens, Greece) Prof. Dr. Aydin Akan (Istanbul University, Turkey) Prof. Sarka Necasova (Academy of Sciences, Prague, Czech Republic) Prof. C. D. Memos (National Technical University of Athens, Greece) Prof. S. Y. Chen, (Zhejiang University of Technology, China and University of Hamburg, Germany) Prof. Duc Nguyen (Old Dominion University, Norfolk, USA) Prof. Tuan Pham (James Cook University, Townsville, Australia) Prof. Jiri Klima (Technical Faculty of CZU in Prague, Czech Republic) Prof. Rossella Cancelliere (University of Torino, Italy) Prof. L.Kohout (Florida State University, Tallahassee, Florida, USA) Prof. D' Attelis (Univ. Buenos Ayres, Argentina) Prof. Dr‐Eng. Christian Bouquegneau (Faculty Polytechnique de Mons, Belgium) Prof. Wladyslaw Mielczarski (Technical University of Lodz, Poland) Prof. Ibrahim Hassan (Concordia University, Montreal, Quebec, Canada) Prof. Stavros J.Baloyannis (Medical School, Aristotle University of Thessaloniki, Greece) Prof. James F. Frenzel (University of Idaho, USA) Prof. Mirko Novak (Czech Technical University in Prague,Czech Republic) Prof. Zdenek Votruba (Czech Technical University in Prague,Czech Republic) Prof. Vilem Srovnal,(Technical University of Ostrava, Czech Republic) Prof. J. M. Giron‐Sierra (Universidad Complutense de Madrid, Spain) Prof. Zeljko Panian (University of Zagreb, Croatia) Prof. Walter Dosch (University of Luebeck, Germany) Prof. Rudolf Freund (Vienna University of Technology, Austria) Prof. Erich Schmidt (Vienna University of Technology, Austria) Prof. Alessandro Genco (University of Palermo, Italy) Prof. Martin Lopez Morales (Technical University of Monterey, Mexico) Prof. Ralph W. Oberste‐Vorth (Marshall University, USA)

Prof. Vladimir Damgov (Bulgarian Academy of Sciences, Bulgaria) Prof. Menelaos Karanasos (Brunel University, UK) Prof. P.Borne (Ecole Central de Lille, France)

Additional Reviewers Lukas Zach Valeriu Prepelita Ioannis Gonos Shahram Javadi Metin Demiralp Valeri Mladenov Dimitris Iracleous Nikos Doukas Filippo Neri Nikos Karadimas Aida Bulucea Keffala Mohamed Rochdi Mihaiela Iliescu George Tsekouras Nikos Bardis Milan Stork Vassiliki T. Kontargyri

Table of Contents

Keynote Lecture 1: Ant Decision Systems for Combinatorial Optimization with Binary Constraints

15

Nicolas Zufferey

Keynote Lecture 2: A New Framework for the Robust Design of Analog Blocks Using Conic Uncertainty Budgeting

16

Claudio Talarico

Keynote Lecture 3: On Mutual Relations Between Bioinspired Algorithms, Deterministic Chaos and Complexity

17

Ivan Zelinka

Image Representation Method Based on Complex Wavelet Transform and Phase Congruency, with Automatic Threshold Selection

19

Arathi T., Latha Parameswaran

A Novel Approach for Protein Folding Using GA Feature Selection and Cellular Learning Automata in Sugarscape Model

30

Elahe Hosseinkhani, Saeed Setayeshi, Mohammad Teshnehlab

Reconstruction of High‐Resolution Computed Tomography Image in Sinogram Space 39

Osama A. Omer

Finite Element Analysis of the Lower Extrtemity ‐ Hinge Knee Behavior under Dynamic Load

44

L. Zach, S. Konvickova, P. Ruzicka

Control of Upper Limb Active Prosthesis Using Surface Electromyography 47

Muhammad Asim Waris, Mohsin Jamil, Syed Omer Gilani, Yasar Ayaz

Development of a System for Measurement on Asymmetric Sitting Posture 52

Ji‐Yong Jung, Yonggwon Won, In‐Sik Park, Tae‐Kyu Kwon, Jung‐Ja Kim

Carbonic Anhydrase as CO2 Capturing Agent: Its Classes and Catalytic Mechanisms 57

Bashistha Kumar Kanth, Seung Pil Pack

Feasibility of the C60 Fullerene Antioxidant Properties: Study with Density Functional Theory Computer Modeling

62

V. A. Chistyakov, Yu. O. Smirnova, I. Alperovich


13

Voice Pathologies Classification Using GMM And SVM Classifiers 65

Amara Fethi, Fezari Mohamed

The Use of Starch Matrix‐Banana Fiber Composites for Biodegradable Maxillofacial Bone Plates

70

Lamis R. Darwish, Mohamed Tarek El‐Wakad, Mahmoud Farag, Mohamed Emara

Authors Index 77


14

Keynote Lecture 1

Ant Decision Systems for Combinatorial Optimization with Binary Constraints

Professor Nicolas Zufferey HEC ‐ University of Geneva, Switzerland E‐mail: nicolas.zufferey‐[email protected]

Abstract: In this paper is considered a problem (P) which consists in minimizing an objective function f while satisfying a set of binary constraints. Function f consists in minimizing the number of constraints violations. Problem (P) is NP‐hard and has many applications in various fields (e.g., graph coloring, frequency assignment, satellite range scheduling). On the contrary to exact methods, metaheuristics are appropriate algorithms to tackle medium and large sized instances of (P). A specific type of ant metaheuristics is designed to tackle (P), where in contrast with state‐of‐the‐art ant algorithms, an ant is a decision helper and not a constructive procedure. Brief Biography of the Speaker: Swiss citizen, Nicolas Zufferey is Professor of Operations Management at the University of Geneva. He holds a PhD in Operations Research from EPFL. His research and publications relate to the heuristics, operations research, optimization, logistics management and quantitative management methods. The full paper of this lecture can be found on page 260 of the Proceedings of the 2013 International Conference on Applied Mathematics and Computational Methods, as well as in the CD‐ROM proceedings.


15

Keynote Lecture 2

A New Framework for the Robust Design of Analog Blocks Using Conic Uncertainty Budgeting

Professor Claudio Talarico Department of Electrical and Computer Engineering

Gonzaga University Spokane, WA, USA

E‐mail: [email protected]

Abstract: In nanoscale technologies process variability makes it extremely difficult to predict the behavior of manufactured integrated circuits (IC). The problem is especially exacerbated in analog IC where long design cycles, multiple manufacturing iterations, and low performance yields causes only few design to have the volume required to be economically viable. This paper presents a new framework that accounts for process variability by mapping the analog design problem into a robust optimization problem using a conic uncertainty model that dynamically adjust the level of conservativeness of the solutions through the introduction of the notion of budget of uncertainty. Given a yield requirement, the framework implements uncertainty budgeting by linking the yield with the size of the uncertainty set associated to the process variations depending on the design point of interest. Dynamically adjusting the size of the uncertainty set the framework is able to find a larger number of feasible solutions compared to other robust optimization frameworks based on the well known ellipsoidal uncertainty (EU) model. To validate the framework, we applied it to the design of a 90nm CMOS differential pair amplifier and compared the results with those obtained using the EU approach. Experimental results indicate that the proposed Conic Uncertainty with Dynamic Budgeting (CUDB) approach attain up to 18% more designs meeting target yield. Brief Biography of the Speaker: Claudio Talarico is Associate Professor of Electrical and Computer Engineering at Gonzaga University. He holds a PhD degree in electrical engineering from University of Hawaii where he conducted research in the area of Embedded System‐on‐Chip. Before joining Gonzaga University, he worked at Eastern Washington University, University of Arizona, University of Hawaii, and in industry where he held both engineering and management positions in the area of VLSI integrated circuits. The companies he worked for include Infineon Technologies, in Sophia Antipolis, France, IKOS Systems in Cupertino, CA and Marconi Communications, in Genova, Italy. The full paper of this lecture can be found on page 49 of the Proceedings of the 2013 International Conference on Electronics, Signal Processing and Communication Systems, as well as in the CD‐ROM proceedings.


16

Keynote Lecture 3

On Mutual Relations Between Bioinspired Algorithms, Deterministic Chaos and Complexity

Professor Ivan Zelinka Technical University of Ostrava

Czech Republic E‐mail: [email protected]

Abstract: This lecture is focused on mutual intersection of three interesting fields of research i.e. bioinspired algorithms, deterministic chaos and complexity, introducing a novel approach joining bioinspired dynamics, complex networks and CML systems exhibiting chaotic behavior. The first part will discuss a novel method on how dynamics of bioinspired algorithms can be visualized in the form of complex networks. An analogy between individuals in the populations in an arbitrary bioinspired algorithm and the vertices in a complex network will be discussed as well as the relationship between the communications of individuals in a population and the edges in a complex network. The second part will discuss the possibility of how to visualize the dynamics of a complex network by means of coupled map lattices and to control by means of chaos control techniques. The last part will discuss some possibilities on CML systems control, especially by means of bioinspired algorithms. The spirit of this keynote speech is to create a closed loop in the following schematic: bioinspired dynamics ‐‐> complex network ‐‐> CML system ‐‐> control CML ‐‐> control bioinspired dynamics. Real‐time simulations as well as animations and pictures demonstrating the presented ideas will be presented through this lecture. Brief Biography of the Speaker: Ivan Zelinka is currently working at the Technical University of Ostrava (VSB‐TU), Faculty of Electrical Engineering and Computer Science. He graduated consequently at Technical University in Brno (1995 ‐ MSc.), UTB in Zlin (2001 ‐ Ph.D.) and again at Technical University in Brno (2004 ‐ assoc. prof.) and VSB‐TU (2010 ‐ professor). Before academic career he was an employed like TELECOM technician, computer specialist (HW+SW) and Commercial Bank (computer and LAN supervisor). During his career at UTB he proposed and opened 7 different lectures. He also has been invited for lectures at 7 universities in different EU countries plus role of the keynote speaker at the Global Conference on Power, Control and Optimization in Bali, Indonesia (2009), Interdisciplinary Symposium on Complex Systems (2011), Halkidiki, Greece and IWCFTA 2012, Dalian China. He is and was responsible supervisor of 3 grant of fundamental research of Czech grant agency GAČR, co‐supervisor of grant FRVŠ ‐ Laboratory of parallel computing. He was also working on numerous grants and two EU project like member of team (FP5 ‐ RESTORM) and supervisor (FP7 ‐ PROMOEVO) of the Czech team. Currently he is professor at the Department of Computer Science and in total he has been supervisor of more than 30 MSc. and 20 Bc. diploma thesis. Ivan Zelinka is also supervisor of doctoral students including students from the abroad. He was awarded by Siemens Award for


17

his Ph.D. thesis, as well as by journal Software news for his book about artificial intelligence. Ivan Zelinka is a member of British Computer Socciety, Editor in chief of Springer book series: Emergence, Complexity and Computation, Editorial board of Saint Petersburg State University Studies in Mathematics, Machine Intelligence Research Labs (MIR Labs ‐ http://www.mirlabs.org/czech.php), IEEE (committee of Czech section of Computational Intelligence), a few international program committees of various conferences and international journals (Associate Editor of IJAC, Editorial Council of Security Revue, http://www.securityrevue.com/editorial‐council/). He is author of journal articles as well as of books in Czech and English language.


18

Image representation method based on Complex Wavelet Transform and Phase Congruency, with

Automatic Threshold Selection

Arathi T PhD Scholar, Department of Computer Science &

Engineering Amrita Vishwa Vidyapeetham

Coimbatore, India [email protected]

Latha Parameswaran Professor, Department of Computer Science & Engineering

Amrita Vishwa Vidyapeetham Coimbatore, India

[email protected]

Abstract— Image representation is an active area of research with increasing applications in military and defense. Image representation aims at representing an image with lesser number of coefficients than the actual image, without affecting the image quality. It is the first step in image compression. Once the image is represented by using some set of coefficients, it is further encoded using various compression algorithms. This paper proposes an adaptive method for image representation, which uses Complex Wavelet transform and the concept of phase congruency, where the number of coefficients used for image representation depends on the information content in the input image. The efficiency of the proposed method has been assessed by comparing the number of coefficients used to represent the image using the proposed method with that used when Complex Wavelet transform is used for image representation. The resultant image quality is determined by computing the PSNR values and Normalized Cross Correlation. Experiments carried out show highly promising results, in terms of the reduction in the number of coefficients used for image representation and the quality of the resultant image.

Keywords— Image Representation; Complex Wavelet Transform; Coefficient of Variation; Phase Congruency; Peak Signal to Noise Ratio

I. INTRODUCTION Transforming images from the spatial domain to the

frequency domain has been found to be the general trend followed in various representation methods. The key reason for this is the way frequency domain representation of the image makes the image coefficients uncorrelated with each other, making their analysis easier. The commonly used tool for converting the image data from spatial to spectral domain is the Fourier transform [1]. Due to the non-local property of Fourier transform, they were replaced by wavelet functions, which due to their unique property of being local [2], was found to be better suited for data representation. Wavelets were found to be highly efficient in approximating data with sharp discontinuities [2]. The advent of wavelets opened a new path for the development of image representation algorithms. The availability of a large variety of wavelets

allowed the image to be analyzed at multiple resolutions, thereby allowing the redundancy to be removed from each resolution level. However, even though Discrete Wavelet Transform (DWT) proved promising, it has inherent limitations. Since the DWTs are critically sampled, it is not shift invariant and lacks directional selectivity. This led to the development of a variant of the conventional wavelet called the Complex Wavelet Transform (CWT) [3]. This makes the transform shift invariant, but the directional selectivity is still a problem in CWT.

Phase Congruency is a feature operator which is invariant to illumination and scale. It assumes an image to be highly rich in information and very little redundancy. This property makes sure that the proposed technique doesn’t treat any major information in the image as redundant and remove it [4].

This paper combines CWT and the concept of Phase Congruency and proposes a new technique for image representation. Even though image representation techniques aim at reducing the number of coefficients used for representing the image [5], it also results in loss of information in the image. Hence, a trade-off must be obtained between the reduction in the number of coefficients obtained and the quality of the resultant image. A good representation algorithm should aim to obtain a fairly good trade-off between the two. The percentage of reduction in the coefficients and the resultant image quality [6, 7] are also dependent on the type of the input image. The proposed technique adaptively changes the threshold value for redundancy removal, depending on the input image. Experimental results show that the proposed method achieves very good levels of coefficient reduction and at the same time does not compromise much on image quality, as the information loss is kept to the bare minimum.

II. COMPLEX WAVELET TRANSFORM The Discrete Wavelet Transform has the following drawbacks [3]:

• DWT coefficients oscillate at zero crossings; • It is not shift invariant;


19

• Aliasing occurs due to downsampling during analysis, which is not always cancelled during synthesis;

• DWT lacks directional selectivity;

CWT employs quadrature wavelets, which provides a magnitude and phase representation, shift invariance and no aliasing as well. CWT can be expressed as:

)()()( tjtt irc ψψψ +=

where, )(trψ is real and even and )(tj iψ is imaginary and

odd. If )(trψ and )(tiψ are chosen to be out of phase with

each other by 090 (Hibert transform pair), then )(tcψ is an analytic signal and supported on only one-half of the frequency axis. The complex wavelet coefficients can be expressed as:

),(),(),( njjdnjdnjd irc += with magnitude

22 ),(),(),( njdnjdnjd irc +=

and phase angle given by

=∠

),(),(

arctan),(njdnjd

njdr

ic

CWT enables to analyse and represent both real-valued and complex-valued signals, just like Fourier transform [18].

A. DT-CWT Filterbank The filterbank structure of the 1-D, DT-CWT resembles

the conventional DWT, with twice the complexity [8]. It can be thought of as two conventional DWT trees operating in parallel. One is the real tree and the other is the imaginary tree. The conjugate filters used in analysis are of the form:

xx jgh + , where xh is the set of filters { }10 ,hh and xg is

the set of filters{ }10 , gg . Figure 1 shows that the filters

0h and 1h are the real-valued lowpass and highpass filters

respectively for the real tree. 0g and 1g are the real-valued lowpass and highpass filters respectively for the imaginary tree.

Figure 1: DT-CWT structure with two separable DWT

The synthesis filter pairs form orthogonal pairs with their respective counterparts of the analysis tree. For the 2D DT-CWT, the filter structure has four trees for analysis and synthesis as shown in Figure 2. The pair of conjugate filters is applied to two dimensions (x and y) and is expressed as: ( )( ) ( ) ( )yxyxyxyxyyxx hgghjgghhjghjgh ++−=++

Figure 2: Filter bank structure for 2-D DT-CWT

The 2-D DT-CWT is 4-times expensive than the standard

2-D DWT, since it has 4 different trees. Trees a and b are the real pair and trees c and d form the imaginary pair in the

analysis stage. Tree pairs (~a ,

~b ) and (

~c ,

~d ) are the real and

imaginary parts respectively in the synthesis stage, corresponding to the analysis pairs [9].

III. PHASE CONGRUENCY It has been traditionally a practice in image processing to

think about features in terms of derivatives. This is because, features in images are mostly thought of as edges, which are points of discontinuities. As a result, gradient based operators are mostly used to detect the features in images. The gradient based feature estimation techniques, such as those developed by Sobel [11], Marr and Hildreth [12] and Canny [13, 14], face two major drawbacks. Firstly, the gradient operators are sensitive to illumination variations. i.e; they cannot be relied on, when working with images of varying lighting and contrast. The second shortcoming of gradient operators is that, localization of features depends on the scale of analysis. Hence, the localization becomes innacurate when analysed at varying scales. This leads to the need of a feature operator that is invariant to illumination and scale. Phase congruency model of feature detection [4] assumes an image to be high in information and low in redundancy. Thus, instead of searching for points of sharp changes in intensity, this model searches for patterns, where the phase components of the Fourier transform of the image are in order (maximally in-phase). It is a frequency-based model and instead of spatial processing of data, it processes an image using the phase and amplitude components of the individual frequency components.


20

Consider a 1-D slice through an image. Such a signal f(x) can be reconstructed from its Fourier transform by:

∫∞

∞−

+= ωφω ωω dxTaxf )cos()( ,

where for each frequencyω , ωa is the amplitude of the cosine

wave and ωφω +xT is the phase offset of that wave. The term ‘T’ is related to the size of the image window. The phase congruency model in the discrete form is expressed as:

∑∞

=

+++

=0

])12[sin()12(

1)(n

p xnn

xf φ ,

where, p gives the decay in the amplitude with frequency and φ is the phase offset. Phase Congruency is the ratio of

local energy to amplitude. ∑

=

nn xAxE

xPC)(

)()( , where,

An(x) is the amplitude and E(x) is the local energy and 1)(0 ≤≤ xPC [10].

IV. PROPOSED TECHNIQUE FOR IMAGE REPRESENTATION In [9, 10] the authors have discussed an image

representation technique using CWT and Wavelet transform. This proposed technique de-correlates the input image information by transforming the image into the frequency domain using CWT. The proposed algorithm makes use of the concept of phase congruency, to determine the amount of redundant information that needs to be removed from the input image. In our earlier paper [15], we had proposed a technique for image compression, which uses Slantlet transform and phase congruency, where the threshold had to be provided by the user, which determined how much of the coefficients would be removed for a fairly good representation. However, choosing threshold values empirically can’t always be reliable, as the amount of redundancy is image dependent.

We have introduced a threshold selection method in this paper, which selects the threshold value automatically from the input image, based on its information content. Here the threshold value used is the Coefficient of Variation [16, 17] of the transformed image. The value thus obtained is found to be an optimum threshold, which reduces the number of coefficients used for representation by almost 60-70%, at the same time maintaining the image quality.

The algorithm is applied on the image as a whole. The following steps are carried out on the input image.

A. Decomposition of theiInput image using CWT Complex Wavelet transform is applied to each column of

the input image. The CWT filter coefficients used in this experimentation is obtained from [3]. Let I(x,y) be the input block to be compressed. The CWT of each column of the block is carried out, resulting in a corresponding coefficient block in the transform domain, denoted as TI(u,v). Let the

transform domain coefficients of the transformed block be represented as C(u,v)TI.

B. Phase Congruency map for the decomposed Image The next step in the proposed compression technique is to

create the phase congruency map for the transformed image block, TI(u,v). Each transform domain coefficient will thus have a phase congruency value corresponding to the position of the coefficient in the transformed image. Let the phase congruency map for the transformed image block be denoted as PCTI and each phase congruency value be represented as PC(u,v)TI.

C. Selecting the threshold value for selection of coefficients Figures and Tables Once the phase congruency map has been generated, we

need to select the threshold value for selecting the coefficients. We select the threshold value by computing the Coefficient of Variation [16] for the transformed image. The value for the threshold is obtained as:

( )

mn

vuCT

mn

iTI

SH

∑== 1

,

D. Removing the redundant CWT coefficients Figures and Tables The phase congruency map acts as the basis for removing

the redundant CWT coefficients. The compression algorithm chooses only those CWT coefficients from the transformed image block, which has edge strength greater than the obtained threshold TSH. The edge strength is represented by the normalized phase congruency value from the phase congruency map. The decision rule can be expressed as:

SHTI

SHTITITIC

TvuPCifTvuPCifvuCvuC<=

>=),( ,0

),( ,),(),(

where, TI

C vuC ),( are the CWT coefficients of the compressed image.

E. Obtaining the final mage block To get the resultant image block, the inverse Complex

Wavelet transform of the coefficients thus selected is taken. The final result is expressed as:

),((),( vuCISTyxF CIC = ,

where, ),( yxFIC is the final image block. It can be observed that the number of CWT coefficients that were used to reconstruct back the final image block is much lesser than the actual number of CWT coefficients that was used to represent the input image block. Experimental results also prove that this technique helps to achieve almost 60-70% reduction in the number of coefficients used for representing the image, without trade-off in the visual quality of the image.


21

F. Experimental Results and Analysis The proposed CWT and phase congruency based image

representation technique was applied on a set of grayscale images of size 512 x 512 of various categories such as standard test images, natural images, user created images and medical images. The number of coefficients used for representation, PSNR and Normalized Cross Correlation (NCC) values are computed to determine the degree of efficiency achieved in representing the image and the amount of information loss respectively. Experimental results pertaining to six images of various categories are discussed here. Figure 3 shows the input images used and Figure 4 shows the output images of the proposed method. It can be observed that the visual quality of the image has not degraded due to this technique.

Figure 3: Input Images

Figure 4: Resultant images using the proposed method

For analyzing the performance of the proposed method, percentage reduction in the number of coefficients when compared to conventional CWT, PSNR and NCC values are measured. In each case the number of CWT coefficients required to represent the image and the number of coefficients required to represent the same image, after the proposed method have been used for analysis. Table 1 shows the details regarding the number of coefficients in CWT based image

representation, the resultant coefficients after the proposed technique, percentage reduction in the number of coefficients used for representation in the proposed method, PSNR and NCC. It also shows the automatically generated threshold for each input image. It also shows the automatically generated threshold for each input image.

TABLE I. NO: OF COEFFICIENTS USED FOR REPRESENTATION IN CONVENTIONAL CWT AND THE PROPOSED METHOD, ITS PERCENTAGE

EQUIVALENT, PSNR AND NCC VALUES.

Observations Lena Peppers Natural User

Created CT MRI

CWT 301401 301401 301401 301401 301401 301401 Proposed Method

86482 79086 84029 109956 116339 125178

No: of coefficients

in %

28.69% 26.24% 27.88% 36.48% 38.60% 41.53%

Automatic Threshold

Value

0.1611 0.1590 0.1818 0.1912 0.0117 0.1053

PSNR 37.2480 34.7208 45.5925 35.9257 45.9276 41.4324 NCC 0.9954 0.9947 0.9961 0.9953 0.9967 0.9960

From Table 1 it is clear that, using the proposed method, the input image can be represented with almost only 30-40% of coefficients, as with conventional CWT. There is no much compromise on the image quality. Visual comparison of the original and the resultant image shows that there is no visual change between the two, ascertaining excellent visual quality. The values of PSNR are high, which shows that there is no much loss of information in the proposed method. Also, the NCC value is very close to 1, which tells that the resultant image obtained is 99% same as the original image.

CONCLUSION Complex wavelet transform, removes the

shortcomings of the conventional DWT. Hence, it has the advantages of shift invariance, directionality and it also avoids aliasing to a large extent. CWT can thus be thought of as a powerful tool in multiresolution analysis. The phase congruency map is generated from the CWT coefficients of the input image and is used as the decision rule to find out the coefficients that are used for image representation. The threshold to be used for the selection of coefficients is obtained automatically from the input image. Coefficients of variation is used for selection of the threshold, as it considers the information content in the image and selects an appropriate threshold value, which doesn’t remove any relevant information in the image. Exhaustive experiments conducted on grayscale images exhibit promising results. The experimental analysis of the results thus obtained shows that using the proposed method, a high degree of reduction in the number of coefficient for representing an image can be obtained, when compared to conventional Complex Wavelet transform.


22

REFERENCES

[1] Eric W. Weisstein, “Fourier Transforms” From Mathworld-a Wolfram Web Resource. http://mathworld.wolfram.com/FourierTransforms.html.

[2] M. Sifuzzaman1, M.R. Islam1 and M.Z. Ali, “Application of Wavelet Transform and its Advantages Compared to Fourier Transform”, Journal of Physica Sciences, Vol. 13, Pages 121-134, 2009.

[3] I. W. Selesnick, R. G. Baraniuk, and N. G. Kingsbury, "The dualtree complex wavelet transform,” IEEE Signal Processing Magazine, vol. 22, no. 6, pp. 123–151, November 2005.

[4] P. D. Kovesi. A dimensionless measure of edge significance from phase congruency pages calculated via wavelets. International First New Zealand Conference on Image and Vision Computing, Auckland, August 1993.

[5] Wei Hong, John Wright, Kun Huang, Yi Ma, Multi-Scale Hybrid Linear Models for Lossy Image Representation. IEEE Transactions on Image Processing, 2006.

[6] Deepak S. Turaga, Yingwei Chen, Jorge Caviedes, “No reference PSNR estimation for compressed pictures, Signal Processing: Image Communication, Elsevier, Vol. 19, Pages: 173-184, 2004.

[7] Tania Stathaki, “Image Fusion: Algorithms and Applications”, Academic Press, 2008 edition.

[8] Felix Fernandes, ‘Directional, Shift-insensitive, Complex Wavelet Transforms with Controllable Redundancy’, PhD Thesis, Rice University, 2002.

[9] Dr. Salih Husain Ali & Aymen Dawood Salman, Image Compression Based on 2D Dual Tree Complex Wavelet Transform (2D DT-CWT), Enggineering & Technical Journal, Vol. 28, No.7, 2010.

[10] M.B. Pardo, C.T. van der Reijden, “ Embedded lossy image compression based on wavelet transform”, Video/Image Processing and Multimedia Communications 4th EURASIP- IEEE Region 8 International Symposium on VIPromCom, November 2002.

[11] J. F. Canny. A computational approach to edge detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 8(6):112–131, 1986.

[12] R. Deriche. Using Canny’s criteria to derive an optimal edge detector recursively implemented. The International Journal of Computer Vision, 1:167–187, April 1987.

[13] D. L. Donoho. De-noising by soft thresholding. Technical Report 409, Department of Statistics. Stanford University, 1992.

[14] D. J. Field. Relations between the statistics of natural images and the response properties of cortical cells. Journal of the Optical Society of America A, 4(12):2379–2394, December 1987.

[15] Arathi T, Latha Parameswaran, Slantlet Transform and Phase congruency based Image Compression, AICWIC’13, Proceedings published by International Journal of Computer Applications, IJCA, January 2013.

[16] Anthony Tanbakuchi, Introductory Statistics Lectures-Measures of Variation, 2009.

[17] S.E Ahmed, A Pooling Methodology for Coefficient of Variation, The Indian Journal of Statistics, Volume 57, Series B, pages 57-75, 1995.

[18] N G Kingsbury: “Complex wavelets for shift invariant analysis and filtering of signals”, Journal of Applied and Computational Harmonic Analysis, vol 10, no 3, May 2001, pp. 234-253.


23

http://mathworld.wolfram.com/FourierTransforms.html

A novel approach for protein folding using GA feature selection and Cellular Learning Automata in Sugarscape Model

Elahe Hosseinkhani

Department of Computer Engineering

Science and Research Branch University, IAU.

Tehran, Iran. [email protected]

Saeed Setayeshi Department of Nuclear

Engineering University of Technology

(Tehran Polytechnic) Tehran, Iran.

[email protected]

Mohammad Teshnehlab Electrical and Computer Eng.

Dept. KNT University of Technology. Tehran, Iran.

[email protected]

Abstract: -This paper presents a novel approach to extracting features from motif content and amino acid sequence properties for protein secondary structures prediction. First, we formulate a protein sequence as a fixed-dimensional vector using the motif content and amino acid sequence properties. Then, the Genetic Algorithm (GA) is used to extract a subset of biological and functional sequence features. Finally we utilize the learning cellular automata in sugar space for predicting protein folding.

Key-Words: -Genetic algorithm; Motif content; Protein secondary structure; Protein folding; Learning cellular automata; Sugar space;

I. Introduction Proteins are complex organic compounds that consist of amino acids joined by peptide bonds. Amino acids are the basic building blocks of proteins. Amino acids play central roles both as building blocks of proteins and as intermediates in metabolism. The 20 amino acids that are found within proteins convey a vast array of chemical versatility. The precise amino acid content, and the sequence of those amino acids, of a specific protein, is determined by the sequence of the bases in the gene that encodes that protein. The chemical properties of the amino acids

of proteins determine the biological activity of the protein.

Proteins play a key role in almost all biological processes. They take part in, for example, maintaining the structural integrity of the cell, transport and storage of small molecules, catalysis, regulation, signaling and the immune system. Linear protein molecules fold up into specific three-dimensional structures, and their functional properties depend intricately upon their structures. The protein structure is the result of the so-called protein folding process in which the initially unfolded chain of amino


30

acids is transformed into its final structure. Under suitable conditions, this structure is uniquely determined by the sequence. There are four different structure types of proteins, namely Primary, Secondary, Tertiary and Quaternary structures. Primary structure refers to the amino acid sequence of a protein. It provides the foundation of all the other types of structures. Secondary structure refers to the arrangement of connections within the amino acid groups to form local structures. Helix, strand and coil are some examples of structures that form the local structure. Tertiary structure is the three dimensional folding of secondary structures of a polypeptide chain. Quaternary structure is formed from interactions of several independent polypeptide chains (Whitford, 2005).

Protein folding is one of the most important problems in the field of bioinformatics and one of the most demanding tasks of protein engineering in post genome era. Because of laboratory tests are time consuming and obtain determining proteins’ structures via experiments is difficult, in recent years, theoretical computing has become a feasible approach to which scientists have been turning for help to predict the structures of proteins from their amino acid sequences.

The difficulty of this problem is due to the roughness of the energy landscape with a multitude of local energy minima separated by high barriers. Conventional Monte Carlo and molecular dynamics simulations tend to become trapped in local minima and are hence incapable of exploring the global energy surface Even in simplified lattice model, the problem of finding the ground state of the protein is NP complete (Li, 2007).

Many computational approaches have been used and tested to for this problem, including Simulated annealing (Li, 2007), genetic algorithms (Argos, 1997), logic-based machine learning (Stephen Muggleton, 1992), neural networks (SANDER, 1993).

In this paper, first, we formulate a protein sequence as a fixed dimensional vector via the motif content and amino acid sequence properties. Then, the GA technique is used to extract a subset of biological and functional sequence features for using as rules and environmental response in learning cellular automata with sugarscape. Since protein folding has been happen in water, we use sugarscape model.

II. Methods

A. Using GA to extracting features from protein sequences

In this section, we use GA for extracting features from motif content and amino acids sequence properties for protein structure prediction.

B. Vectorization of protein sequences By analyzing the properties of such group of similar sequences, it is possible to derive a signature for a protein family or domain, which distinguishes its members from all other unrelated proteins. The first step of our proposed approach is to convert each protein sequence into a vector of fixed dimensionality based on the motif content and amino acids sequence properties. The set of motifs to be used can be chosen from the existing motif database RostSanderDataset (Sander, 1993) and Protein Data Bank (PDB) which is a database of protein families and domains.


31

Therefore, we consider helices, sheets and coil with length of 3 and 4from exiting database as motif. Since the RostSanderDataset database contains 370 entries, the number of features will therefore be 370. Each element of the vectors represents the presence or absence of a motif in the protein sequences. That is, the corresponding feature value will be 1 if a motif is present. Otherwise, it will be 0.

Amino acids are divided based on their remaining properties into polar or not polar, hydrophobic or hydrophilic, low vanderwaals volume or high, such as shown in table 1.

Table 1) Amino acids properties1

Property Low(0) High(1) Hydrophobic

A,D,N,C,E,Q,G,P,S,T,V

R,H,I,L,K,M,F,W,Y

Vanderwaals volume

R,D,N,E,Q,G,H,K,P,S,T,W,Y

A,C,I,L,M,F,V

Polar A,C,E,H,I,L,M,F,P,W,Y,V

R,,T,N,D,Q,G,K,S

Exiting motifs encoded by these properties, for example hydrophobic code of ARH and ARHR is 100 and1001, its vanderwaals code is 100 and 1000 and polar code is 010,0101 . For each protein sequence evaluate these codes. According to these properties, feature value will be filled. Therefore the number of features will be 370 + 3*8 + 3* 16 = 442.

C. Genetic Algorithms The genetic algorithm (GA) is a search heuristic that mimics the process of natural evolution. This heuristic is routinely used to generate useful solutions to optimization and

1 (Alexandre G. de Brevern, Serge A. Hazout, 2000) and (John Wiley and Sons, 2002)

search problems .The main element of a population of individual are represented by feature encoding chromosomes Genetic algorithms belong to the larger class of evolutionary algorithms (EA), which generate solutions to optimization problems using techniques inspired by natural evolution, such as inheritance, mutation, selection, and crossover.

D. 7BChromosome representation In this paper, the chromosome is encoded into a bit string. A chromosome represents the selected features. Let m be the total number of features, and chromosome is represented by a binary vector of dimension m. If the ith bit of the vector is equal to 1, the corresponding ith feature is selected; otherwise, the corresponding ith feature will not be selected.

E. 8BFitness function for chromosome evaluation

The goal of feature selection is to use fewer features to achieve the same or better performance compared with that obtained using the complete feature set. Hence, chromosome evaluation contains the following two objectives: (1) minimizing the number of features; (2) maximizing the classification accuracy. Obviously, there are some trade-offs between the accuracy and the number of features, among which the accuracy is our major concern. In the literature (Deb, K., Reddy, A. R., 2004) there have been many methods proposed for combining the above two terms. In this paper, a simple weighting method using linear aggregation of the two objectives is adopted. Given a chromosome g, the fitness function can be defined as:

f(g)= f1(g)+wf2(g) (1)


32

Where w is the weighting coefficient, f1(g) is the recognition rate obtained using features presented in g, and f2(g) is the number of features removed from the original feature set. A small value is given to w because we mainly focus on the recognition rates. In this paper, w is set at 4*10-5. It can be seen that the chromosomes with higher accuracies will outweigh those with lower accuracies; no matter how many features they contain (Xing-Ming Zhaoa,b, Yiu-Ming Cheungc, De-Shuang Huanga, 2005).

F. Genetic operators In this work, the tournament selection is adopted to select two parent chromosomes from the current population. Then, the uniform crossover method is applied to the two parent binary string vectors to produce two offspring’s, and the mutation operation mutates the offspring’s If the mutated chromosome is superior to both parent chromosomes, it replaces the similar one. If it is in between the two parents, it replaces the inferior one; otherwise, the most inferior one in the population is replaced. The procedure of selection, crossover and mutation is repeated until a termination criterion is satisfied.

G. 10BCellular Learning Automata in sugarscape

CLA is obtained by combining cellular automata and learning automata. It is a mathematical model for dynamical complex systems that consists of a large number of simple learning components. Any number of learning automaton can reside in a specific cell. Reinforcement signal for every automaton is computed according to CLA rule and actions of other learning automata residing in neighbor cells. This model has learning capability of learning automata and

collective behavior and locality of cellular automata. A d dimensional CLA is a quintuple CLA=(Zd ,Ø,A,N,F) that :

• Zd is a d-dimensional grid of cells.

• Ø is a finite set of states that each cell can possess.

• A is set of learning automata that each of them are assigned to a specific cell.

• N = {X1 ,...,Xm } is finite subset of Zd that is called neighborhood vector.

• F: Øm →β is local rule of CLA. β is a set of valid reinforcement signals that can be applied to learning automata.

Like cellular automata, CLA operate subject to a rule. The rule of CLA and the actions selected by neighboring learning automata of any particular learning automaton determine the reinforcement signal to the learning automaton residing in that cell. In CLA, the neighboring learning automata of any particular learning automaton constitute its local environment, which is nonstationary because it varies as action probability vector of neighboring learning automata vary (B.Jafarpour, 2007).

Learning automata is an abstract model which randomly selects one action out of its finite set of actions and performs it on a random environment. Environment then evaluates the selected action and responses to the automata with a reinforcement signal. Based on selected action, and received signal, the automata updates its internal state and selects its next action. The environment will then respond to the input by either giving a reward, or a penalty, based on the penalty probability ci associated with αi . This response serves as the input to the automata. Based upon the response from the


33

environment and the current information accumulated so far, the learning automata decide on its next action and the process repeats. The intention is that the learning automata gradually converge toward an ultimate goal. (H.Mostafaei, 2010)

Figure 1) Interaction between environment and automata

α = {α1, ... , αr} is the set of r actions offered by the environment that the LA must choose from.

β = {0, 1} is the set of inputs from the environment where ‘0’ represents a reward and ‘1’ a penalty.

P = [p1(n), ..., pr(n)] is the action probability vector where pi represents the probability of choosing action αi at the nth time instant.

If (β=1&& αi is chosen)

then Pi (n+1) = Pi(n) + α[1- Pi(n)]


then Pj (n+1) = (1-a)Pj(n) i ≠ j ∀j


then Pi (n+1) = (1-b)Pi(n)


then Pj (n+1) = b/(r-1) + (1-b)Pj(n) i ≠ j ∀j

According to above equation if a and b be equal the learning algorithm will be known as linear reward penalty. If (b<<a) the learning algorithm will known as linear reward epsilon penalty and if b=0 the learning algorithm will be a linear reward

inaction. In this paper we use linear reward epsilon penalty LA.

H. Sugarscape model The basic here are: a) agent, b) rules, c) landscape, d) sugar (resource)

A) Agent: In refers to each existing part in this space including people, population or institutions that simulate human behaviors.

B) Rules:There are some rules for life and survival of such agents in space. They can be changed according to needs in sugarscape. There are 2 groups of rules in sugarscape:

1) Rules governing agent, 2) Rules governing space

Since different rules cause different behaviors the results of simulation is always changing. Therefore a reliable simulation requires unified rules for all agents as well as right sequence of agents.

C) Landscape:There’s no fixed topology in sugarscape for landscape but it can be defined as a one dimension network. In other word, the landscape here can be considered as a puzzle in each part of which there may be sugar, agent, none, or both.

D) Sugar (assets or resource):Sugar is a general resource which agents eat for their survival sugar resources indicate assets. In first state, the sugar distribution in landscape occurs hazardously in spatial are as with a likely distribution among certain limit. Sugar can be renewed with certain rate in order to achieve maximum capacity. Agents here start working in random their primary situation, assets, and all their internal area. A subgroup of internal states always remains unchanged with in agent life; where as other subgroup depends on time. In


34

addition some of these states are spatial and different for each agent and some others are common. The spatial time independent states are primary assets, maximum life time, vision, metabolism rate, and …. The overall independent states include: time needed for increasing vision, poverty limit (=o), spatial time dependant states such as agent situation in landscape, real asset in sugar units, and … The agent executes rules simultaneously in searching for sugar. Thorough movement of population is an emergent result of simple spatial activities by agents. In computerized simulation, the model includes cellular automata i.e. it’s a fixed topology that never changes.

Sugarscape Model= Cellular Automata+ Agents+ Sugar+ Rules

This model can be considered as a two dimension cellular automata, each point of which possesses (x) features. A sugar level and a sugar capacity are considered for each point and the maximum sugar capacity is one. ( Rahman, Setayeshi, 2007).

III. PROPOSED ALGORITHM In this section a new approach for Protein Folding Prediction Problem, using cellular learning automata is proposed. First, a solution representation method should be chosen for this purpose. In this paper, permutation with repetition is used. We used an array with a length of n (length of protein sequence) to represent solutions that named Cell. Each cell has a learning automaton. Fig.2 shows the construction of used CLA in our algorithm.

Figure 2) the construction of CLA in proposed algorithm

Where LA is the learning automata residing in each cells. We consider two or eight neighborhoods for each cell. Thus for cell i, neighbors are cell (i-1) , cell (i-2) , cell (i-3) , cell (i-4) ,cell (i+1) , cell (i+2) , cell (i+3) and cell (i+4) . If the protein sequence is r1r2 . . . rn, then for the i-th residue, the following pairs and triples are considered particularly important for helical regions: (ri, ri+1), (ri, ri+3),(ri, ri+4), (ri, ri+1, ri+4), (ri, ri+3, ri+4). Note that residues three and four apart are considered, as they lie on the same face of an α-helix. Similarly, the pair (ri, ri+2) contains residues on the same face of a β-strand. Pairs and triplets of particular amino acids are then deemed as compatible or incompatible with helices and strands based on various rules that try to ensure that these residues present a face that allows tight packing of hydrophobic cores. Factors used to determine these rules that used in LA, include features that selected by GA.

GA and CLA_Sugarscape model algorithm

{

t=0;

initial papulation of individual p(0)

repeat

{

t=t+1;

select p(t) from p(t-1);

perform crossover on p(t);

mutate p(t);

evaluate fitness of p(t);

} until (termination criteria);

return feature set;


35

Initial Sugarscape

Initial CLA

repeat

{

update each cell;

serve environment response;

for each cell if convert coil to α-helix or β-sheet , free water ;

for each cell if convert α-helix or β-sheet to coil, consume water ;

} until (exiting water and convergence happen);

return protein structure;

}

IV. Experiments and discussions In this section we present the results on using CLA in sugarscape for RostSanderDataset. Values are given for the percentage accuracy on test set structure sequences and evaluate the percentage accuracy test. Results of using CLA on sugarscape on each of protein sequences of the dataset are given in Tables 2.

The model is tested for protein sequences 43 and on average over 68% prediction was accurate. However, in some cases more than 90% predicted correctly have been diagnosed.


36

Table 2) proposed approach test

Annotation of a given protein sequence Length true

predication Percent

1CSE-ICOMPLEX(SERINEPROTEINASE-INHIBITOR)03-JU 63 41 65.08

1FDL-HCOMPLEX(ANTIBODY-ANTIGEN)27-AU 218 140 64.22

1FDX-1ELECTRONTRANSPORT01-AU 54 47 87.04

1FKF-1ISOMERASE07-MA 107 63 58.88

1FXI-AELECTRONTRANSFER(IRON-SULFURPROTEIN)28-AU 96 64 66.67

1HIP-1ELECTRONTRANSFER(IRON-SULFURPROTEIN)01-AP 85 58 68.24

1IL8-ACYTOKINE08-MA 71 43 60.56

1L58-1HYDROLASE(O-GLYCOSYL)06-MA 164 131 79.88

1MCP-LIMMUNOGLOBULIN09-JU 220 121 55.00

1OVO-APROTEINASEINHIBITOR(KAZAL)18-JA 56 35 62.50

1PAZ-1ELECTRONTRANSFER(CUPROPROTEIN)28-JU 120 73 60.83

1PRC-CPHOTOSYNTHETICREACTIONCENTER04-FE 332 290 87.35

1PRC-LPHOTOSYNTHETICREACTIONCENTER04-FE 273 241 88.28

1PRC-MPHOTOSYNTHETICREACTIONCENTER04-FE 323 293 90.71

1PRC-HPHOTOSYNTHETICREACTIONCENTER04-FE 258 188 72.87

1PYP-1ACIDANHYDRIDEHYDROLASE03-FE 280 215 76.79

1R09-2RHINOVIRUSCOATPROTEIN04-MA 255 150 58.82

1RBP-1RETINOLTRANSPORT02-AP 174 93 53.45

1RHD-1TRANSFERASE(THIOSULFATECYANIDESULFUR)23-NO 293 234 79.86

1S01-1HYDROLASE(SERINEPROTEINASE)21-AU 275 194 70.55

1TNF-ALYMPHOKINE25-AU 152 82 53.95

1UBQ-1CHROMOSOMALPROTEIN02-JA 76 47 61.84

1WSY-BLYASE(CARBON-OXYGEN)19-SE 385 289 75.06

2AAT-1TRANSFERASE(AMINOTRANSFERASE)30-MA 396 326 82.32

2CAB-1HYDRO-LYASE05-OC 256 149 58.20

2CYP-1OXIDOREDUCTASE(H2O2(A))27-AU 293 218 74.40

2FNR-1OXIDOREDUCTASE(NADP+(A)FERREDOXIN(A))21-JU 296 178 60.14

2FXB-1ELECTRONTRANSPORT08-FE 81 60 74.07

2GBP-1PERIPLASMICBINDINGPROTEIN23-FE 309 209 67.64

2GLS-ALIGASE(AMIDESYNTHETASE)19-MA 468 342 73.08

2I1B-1CYTOKINE02-J 153 82 53.59

2LTN-ALECTIN26-JU 181 98 54.14

2LTN-BLECTIN26-JU 47 31 65.96

2PAB-ATRANSPORT(THYROXINERETINOL)INSERUM16-SE 114 66 57.89 2PCY-1ELECTRONTRANSPORTPROTEIN(CUPROPROTEIN)03-NO 99 63 63.64

2PHH-1OXIDOREDUCTASE19-JU 391 259 66.24

2SNS-1HYDROLASE(PHOSPHORICDIESTER)14-MA 141 81 57.45


37

2STV-1VIRUS08-JU 184 105 57.07

2TGP-ICOMPLEX(PROTEINASE/INHIBITOR)27-SE 58 44 75.86

2TMV-PVIRUS15-SE 154 130 84.42

2TSC-ATRANSFERASE(METHYLTRANSFERASE)03-JU 264 177 67.05

3B5C-1ELECTRONTRANSPORT16-JA 85 50 58.82

3CLN-1CALCIUMBINDINGPROTEIN11-MA 143 116 81.12

3GAP-AGENEREGULATORYPROTEIN15-AP 208 162 77.88

References

[1] Rahman, Setayeshi. (2007). Social

Behavior Analysis in Artificial Life. Journal of Autonomous Agents and Multi-Agent Systems.

[2] Alexandre G. de Brevern, Serge A. Hazout. (2000). Hybrid Protein Model (HPM) : a method to compact protein 3D-structure information and physicochemical properties. IEEE , 49-54.

[3] Argos, T. D. (1997). Applying experimental data to protein fold prediction with the genetic algorithm. Protein Engineering , 877-893.

[4] B.Jafarpour, M. M. (2007). A Hybrid Method for Optimization (Discrete PSO + CLA). International Conference on Intelligent and Advanced Systems, IEEE .

[5] Deb, K., Reddy, A. R. (2004). Large-scale scheduling of casting sequences using a customized genetic algorithm. Lecture Notes in Computer Science , 141–152.

[6] H.Mostafaei, M. M. (2010). A Learning Automata Based Area Coverage Algorithmfor Wireless Sensor Networks. JOURNAL OF ELECTRONIC SCIENCE AND TECHNOLOGY , 200-205.

[7] John Wiley and Sons. (2002). Computational Methods for Protein Folding: Advances in Chemical Physics.

[8] Li, X. (2007). Protein Folding Based on Simulated Annealing Algorithm. IEEE , 418.

[9] SANDER, B. R. (1993). Improved prediction of protein secondary structure by use of sequence profiles and neural networks. Proc. Natl. Acad. Sci. , 7558-7562.

[10] Sander, R. a. (1993). Prediction of protein secondary structure at better than 70% accuracy. Mol. Bio. , 584-599.

[11] Stephen Muggleton, R. D. (1992). Protein secondary structure prediction using logic-based machine learning. Protein Engineering , 647-657.

[12] Whitford, D. W. (2005). Proteins Structure and Function. . England.

[13] Xing-Ming Zhaoa,b, Yiu-Ming Cheungc, De-Shuang Huanga. (2005). A novel approach to extracting features from motif content and protein composition for protein sequence classification. Elsevier , 1019-1028.


38

Reconstruction of High-Resolution ComputedTomography Image in Sinogram Space

Osama A. OmerAswan faculty of Engineering, Aswan University, Aswan, Egypt

Email: [email protected]

Abstract—An important part of any computed tomography(CT) system is the reconstruction method, which transforms themeasured data into images. Reconstruction methods for CT canbe either analytical or iterative. The analytical methods can beexact, by exact projector inversion, or non-exact based on Backprojection (BP). The BP methods are attractive because of thiersimplicity and low computational cost. But they produce sub-optimal images with respect to artifacts, resolution, and noise.This paper deals with improve of the image quality of BP by usingsuper-resolution technique. Super-resolution can be beneficial inimproving the image quality of many medical imaging systemswithout the need for significant hardware alternation. In thispaper, we propose to reconstruct a high-resolution image fromthe measured signals in Sinogram space instead of reconstructinglow-resolution images and then post-process these images to gethigher resolution image.

I. INTRODUCTION

In the areas of medical diagnostics and non-destructivetesting, it is of great interest to be able to capture images ofthe interior of objects. One common technique to accomplishthis feat is known as Computed Tomography (CT), whichinvented in 1972 [1]. A CT scanner uses digitally sampledX-ray images acquired in multiple directions to calculate cross-sectional images of the X-ray attenuation of an object.

An important part of any CT system is the reconstructionmethod, which transforms the measured data into images.Reconstruction methods for CT can be either analytical oriterative. Analytical methods can be either exact and non-exact.Exact methods are based on exact inversion of the projectorin the continuous domain. Although efficient exact methodsexist they are currently not found in clinical use. Instead, man-ufacturers of clinical CT systems employ non-exact methods,based on Backprojection (BP) methods. Due to approximationsin the derivation of these methods, reconstruction results arecontaminated by artifacts. In return, non-exact methods arecomputationally less demanding, simpler to implement, andoffer a better dose utilization than exact methods [1], [2].

On the other hand, high-resolution images reveal moreinformation than low-resolution images, which therefore easedisease diagnosis and detection. Early, fast, and accuratedetection of imaging biomarkers of the onset and progressionof diseases is of great importance to the medical communitysince early detection and intervention often results in optimaltreatment and recovery. However, earlier biomarkers of diseaseonset are often critically smaller or weaker in contrast com-pared to their corresponding features in the advanced stagesof disease [3].

One way to increase the images resolution is to physicallyreduce the pixel size and therefore increase the number ofpixels per unit area. However, a reduction of pixel size causesdegradation in the image quality. Instead of altering the sensormanufacturing technology, digital image processing methodsto obtain an HR image from low-resolution (LR) observationshave been investigated by many researchers [3]–[10]. Otherresearchers use interpolation techniques to enhance the qualityof the medical images [11], however, the interpolation willnot add new information to the under-sampled signals. Inpractice, it is common to take multiple scans of the samesubject and average them to improve the signal-to-noise ratio(SNR) of the final image [12]. Also, such an approach makesno improvement in image resolution. Super-Resolution (SR)algorithms are an interesting way to increase the resolutionof images. They are based on the fact that, by combiningvarious low resolution (LR) and highly correlated images, it ispossible to obtain a high resolution image (HR) by using theinformation from different images.

The goal in this paper is to enhance the resolution for CTimage using multi-images super-resolution technique. Unlikethe existing super-resolution methods that are usually done asa post-process, we propose to solve the SR problem in thesinogram space.

II. CT RECONSTRUCTION

There are two main types of CT reconstruction techniques.The first type is a Fourier-based technique, such as Filteredbackprojection and linogram. The other type is iterative-basedtechnique. The iterative based technique is algebraic and sta-tistical approaches [1]. Simply, these methods are trying to getthe closest approximation of the density function of the objectby using iterative techniques. Since filtered backprojection isthe most used algorithm in modern CT, we will adopt it in thereconstruction of HR CT images.

There are three main ways for CT reconstuctions usingbackprojection, namely, reconstruction of pencil-beam, fan-beam and conebeam CT. For simplicity and without loss ofgenerality, we will use pencil-beam reconstruction.

A. The Radon Transform

We will focus on explaining the Radon transform of animage function and discussing the inversion of the Radontransform in order to reconstruct the image [1].

We will discuss only the 2D Radon transform, althoughsome of the discussion could be readily generalized to the 3D


39

Radon transform. The Radon transform (RT) of a distributionf(x, y) is given by

p(ζ, φ) =

∫f(x, y)δ(x cosφ+ y sinφ− ζ)dxdx (1)

where δ is the Dirac delta function and x, y, ζ, and φ are thecoordinates. The task of tomographic reconstruction is to findf(x, y) given knowledge of p(ζ, φ).

B. Backprojection

Mathematically, the backprojection operation is defined as:

fBP (x, y) =

∫ π

0

p(x cosφ+ y sinφ, φ)dφ (2)

Geometrically, the backprojection operation simply propagatesthe measured sinogram back into the image space along theprojection paths.

III. RESOLUTION ENHANCEMENT OF CT IMAGES

A. Super-Resolution Problem Description

The multi-images super-resolution problem can be simplydescribed, in matrix-vector notation, as [9], [10]

Y k = DkBkWkX + V k, k = 1 : N (3)

WhereWk is the geometric motion operator between the HRframe X and theK−th LR frame, Y k. The system point spreadfunction (PSF) is modelled by the sparse matrix Bk, and Dk

is a sparse matrix that represents the decimation operator. Thevector V k is the system noise and N is the number of availableLR images. For convenience, concatenate all the measurementsin one vector as follows

Y = HX + V , (4)

From many available estimators, which estimate a HR imagefrom a set of noisy LR images, one may choose to find the mostprobable X, given the measurements, Y, that is the maximuma-posterior probability(MAP), which can be described as tomaximize

Pr{X|Y } (5)

Where Pr{X|Y } is the probability of the HR image, X , giventhe measurements Y

Pr{X|Y } = Pr{Y |X}Pr{X}Pr{Y }

(6)

Therefore, the solution for the maximum a-posteriori proba-bility is described as

X̂MAP = ArgMaxXPr{X|Y } (7)= ArgMaxXPr{Y |X}Pr{X} (8)

By assuming Gaussian distribution for noise and Gibbs distri-bution with some energy function A(X) for the prior infor-mation, we get

Pr{X} = Const. exp(−A(X)) (9)

Then

X̂MAP = ArgMaxXPr{Y ‖X}Pr{X}

= ArgMinX

N∑k=1

‖DBWkX − Y k‖2 + λA(X)

The last term A(X) represents the regularization term andλ represents the regularization parameter. There are manychoices for A(X), depending on the priori information, in-cluding Gaussian prior

A(X) = ‖X‖2 (10)

And the bilateral prior

A(X) =P∑

n=−P

P∑m=−P

amn(̇X − SnxSmy X) (11)

where Snx is a shifting operator by n pixels in x direction

[9]. There are two types of noise exist in this imaging model,namely, additive noise (usually assumed to be Gaussian) andregistration noise (registration error), which can be assumed asLaplacian noise [9]. Based on the modeling of the total noise,the data fidelity term will change.

• In case of Gaussian Noise assumption:

JMAP (X) =

N∑k=1

‖DkBkWkX − Y k‖22 + λA(X)

• In case of Laplacian Noise assumption:

JMAP (X) =N∑k=1

‖DkBkWkX − Y k‖11 + λA(X)

Where ‖.‖11 is the L1-norm.

B. Two-Steps-Based CT Super-Resolution

Resolution enhancement of medical image is usually donein two steps, namely, image reconstruction step and resolutionenhancement step. The image reconstruction can be modeledin matrix-vector multiplication as [12]

U = Sµ+ η (12)

Where S is the reconstruction matrix that relates the measuredsignal, U , with the pixel values, µ and η is the additive noise.As a reverse problem, estimating µ can be done by minimizing

J(µ) = ‖Sµ− U‖22 + λ‖µ‖22 (13)

After reconstructing N LR images, these LR images can befused to get a high-resolution image. The resolution enhance-ment problem is therefore described as

µk= DkBkWkX + V k, k = 1 : N (14)

which can be solved by minimizing the cost function

J(X) =N∑k=1

‖DkBkWkX − µk‖22 + λ(X)‖CX‖22 (15)

Where C is a sparse matrix representing a high-pass filteroperator, the last term is the Tikhonov regularization.


40

C. Reconstruction of High-Resolution CT

Instead of performing resolution enhancement in two stepsas shown in the previous Section, substituting from (14) into(12) leads to

Uk = SDkBkWkX + ηk+ SV k, k = 1 : N (16)

= SDkBkWkX + Ek k = 1 : N (17)

WithEk = η

k+ SV k, (18)

Then the HR CT reconstruction problem, by assuming thatthe total contaminating noise has Laplacian distribution, issimplified to minimize

J1(X) =N∑k=1

‖SDBkWkX − Uk‖11 + λ(X)‖CX‖11 (19)

Where X is the HR image. Uk is the measured signal. Withoutoss of generality, we assume that the blurring operator is thesame for all images, then Bk = B and that the downsamplingoperator is the same for all images, then Dk = D. Also con-sidering that the system function in MRI can be representingby Radon transformation, then S = R. Then (19) becomes

J1(X) =

N∑k=1

‖RDBWkX − Uk‖11 + λ(X)‖CX‖11 (20)

Where the regularization parameter, λ(X), can be calculatedadaptively as a function of the cost function as [10], then

λ(X) =

∑Nk=1 ‖RDBWkX − Uk‖11

1γ − ‖CX‖11

(21)

With1

γ= 2

N∑k=1

‖Uk‖11 (22)

As stated in Section III-A, the data fidelity term and there-fore the cost function depend on the noise model. The totalcontaminating noise Ek(Ek = k + SV k) is a result of manysources, including thermal noise, and registration error.

IV. SIMULATION RESULTS

A. Data Sets

In this section we present experiments illustrating theperformance of the proposed algorithm. The experiments areconducted on Phantom sequence. This sequence contains 4measured signals, in Sinogram space, to increase resolutionby factor 2× 2. The size of each LR image is 256× 256.

B. Experiment Setup

To test the efficiency of the proposed HR CT reconstructionalgorithm, we compared it with the reconstructed LR image.The cost function in (20), is solved iteratively using steepestdecent as

X̂n+1 = X̂n

− βN∑k=1

WTk B

TDTRTsign(RDBWkX̂n −Uk)

+ λCTsign(CX̂n)

The setup of these experiments is as follow, resolution en-hancement factor = 2, β = 4, λ = 0.02, and maximum-iteration = 10. The number of LR images used is 4. Therelative motions of the generated LR phantom images, withrespect to the reference image, are (0, 0), (0, 0.5), (0.5, 0) and(0.5, 0.5), respectively. The high-pass filter, C, is used torepresent Laplacian kernel with dimension 5× 5.

C. Results an Discussions

Figure 1, shows the reference phantom image (Fig. 1a)and its corresponding Sinogram (Fig. 1b). Figure 2 show theresults of adopting the proposed algorithm with the simpleback projection reconstruction algorithm. In Fig. 2a, zoomedparts of the reconstructed LR images is shown. The zoomedpart of the reconstructed HR CT using proposed algorithm.From these figures, we can see that the reconstructed HR CTusing proposed algorithm is sharper than the reconstructed LRimage.

Also the results of adopting the filtered back projectionwith the proposed algorithm is shown in Fig. 3. Iin Fig. 2a,zoomed parts of the reconstructed LR images is shown. IinFig. 3b, the zoomed part of the reconstructed HR CT usingproposed algorithm. From these figures, we can see that thereconstructed HR CT using proposed algorithm is sharperthan the reconstructed LR image. Moreover, it can be shownthat using filtered back projection is better than simple backprojection, which is logic as stated in the litreature.

V. CONCLUSION

In this paper, we proposed a HR CT Reconstructionalgorithm in the sinogram space. The proposed algorithmis solved SR reconstruction proplrm in th sinogram spacerather than in the pixel domain. The conventional SR CTalgorithms perform the enhancement in two steps, namely,reconstrut the CT image in pixel domain and then enhance theresolution by applying SR technique as a post-process. Basedon the simulation results, the proposed algorithm enhances theresolution compared to the reconstructed LR CT image.

REFERENCES

[1] Thorsten M. Buzug, Computed Tomography From Photon Statistics toModern Cone-Beam CT, Springer 2008.

[2] J. Hsieh et. al, Recent Advances in CT Image Reconstruction, CurrentRadiology Reports, Vol. 1, Issue 1, pp 39-51, March 2013.

[3] H. Greenspan, Super-Resolution in Medical Imaging, Computer Journal,vol. 52, No. 1, pp. 43-63, 2009.

[4] Ying Bai, Xiao Han, and Jerry L. Prince, Super-resolution Reconstructionof MR Brain Images, Proc. of 38-th Annual Conference on InformationSciences and Systems, Princeton, New Jersey, pp. 1358-1363, March2004.

[5] Ali Gholipour, Simon K. Warfield, Super-resolution Reconstruction ofFetal Brain MRI, Workshop on Image Analysis for the Developing Brain,London, UK, pp. 45-52, September 2009.

[6] D. Kouamo and M. Ploquin, Super-resolution in medical imaging: Anillustrative approach through ultrasound, IEEE International Symposiumon Biomedical Imaging: From Nano to Macro, pp. 249 - 252, 2009.

[7] H. Tang, T. Zhuang and Ed X. Wu, Realizations of fast 2-D/3-D imagefiltering and enhancement, IEEE Transactions on Medical Imaging, vol.20, no. 2, pp. 132-140, February 2008.

[8] G. M. Callic et. al, Analysis of fast block matching motion estimationalgorithms for video Super-Resolution systems, IEEE Transactions onConsumer Electronics, vol. 54, issue 3, 1430 - 1438, 2008.


41

Fig. 1. From up to down, a) LR CT original Phantom image, b) Thecorresponding Sinogram space.

[9] S. Farsiu, M. D. Robinson, M. Elad, and P. Milanfar, Fast and RobustMultiframe Super Resolution, IEEE Trans. On Image Processing,vol. 13,no. 10, Oct. 2004.

[10] O. A. Omer and T. Tanaka, Region-based weighted-norm with adap-tive regularization for resolution enhancement, Elsevier Digital SignalProcessing, doi:10.1016/j.dsp. 2011.02.005.

[11] J. Anthony Parker, Robert V. Kenyon and Donald E. Troxel, Comparisonof interpolating methods for image resampling, IEEE Transactions onMedical Imaging, vol. 2, no. 1, pp. 31-39, March 1993.

[12] K. P. Pruessmann, M. Weiger, M. B. Scheidegger, and P. Boesiger,SENSE: Sensitivity Encoding for Fast MRI, Magnetic Resonance inMedicine, Vol. 42, pp. 952-962, July 1999.

Fig. 2. From up to down, a) LR CT reconstructed phantom, b) result of HRCT reconstruction using simple back projection.


42

Fig. 3. From up to down, a) LR CT reconstructed phantom, b) result of HRCT reconstruction using filtered backprojection.


43

Finite element analysis of the lower extrtemity - hinge knee behavior under dynamic load

L. Zach, S. Konvickova and P. Ruzicka Department of Mechanics, Biomechanics and Mechatronics

CTU in Prague - Faculty of Mechanical Engineering Prague, Czech Republic [email protected]

Abstract— A key goal of joint endoprosthesis is to become a full-featured functional and anatomical replacement. The joint damage may occur for several reasons - primarily a disease of different nature and magnitude, resulting in gradual and irreversible changes and in an extreme solution in the implantation of artificial joints. However, there should be also mentioned accidents leading to joint destruction, which are often "trigger mechanism" of the disease. This work therefore presents a dynamic computational finite element analysis of a hinge-type knee replacement, which aim to streamline and accelerate the development of knee endoprosthesis. It tackles a question of the overall strength of the implant and detects sites of elevated concentrations of stresses that may be potential sources of implant damages. It also studies the behavior of the endoprosthesis under dynamic loads with emphasis on the study of the shape and size of the contact surfaces, which are closely related to the size of the contact pressure and material wear. Aside the hinged knee replacement, the computational model consisted of femur, fibula, tibia, patella and 25 most important muscles of the lower limb. Due to realistic definition of the boundary conditions, this model is suitable for investigation of in-vivo knee joint replacement behavior.

Keywords— Knee, knee replacement, finite element method, lower limb

I. INTRODUCTION Finite element method (FEM) is a common and an effective

tool used in mechanics for a development or a verification of various components or mechanisms. In biomechanics, using FEM means to undergo many compromises and simplifications. All these simplifications have to be reasonable and must take into account as many tissue characteristics as possible.

Considering this fact, there are two groups of FEA of lower extremity models. The first ones are used to simulate a behavior of a healthy knee joint in-vivo [1, 2, 3, 4] and the second group which deals with a knee joint after a total knee endoprosthesis (TKE) implantation [5,6].

The aim of this paper is to present the complex model of the lower limb, consisting of all bones of the knee and 25 main muscles of the lower limb and 8 ligament units of the knee. This complex model of lower limb model simulates dynamic behavior of the modular oncology hinged knee endoprosthesis

and predicts contact pressure and stress distribution for the TKE.

II. MATERIALS AND METHODS

A. Geometric model For the presented nonlinear dynamic analysis solved in

Abaqus CAE, a universal size of the hinged knee endoprosthesis by ProSpon [7] was chosen.

The modular hinged knee ProSpon is made up of several components to cover individual operation demands. For the presented model, all its main components were modeled, i.e. femoral component, femoral stabilizing rod, tibial compo-nent, tibial stabilizing rod, meniscal component (tibial plat-eau), hinge post, hinge pin, hinge lock and two hinge bush-ings (see Fig. 1). Position of the TKE on the corresponding bones respected the formerly designed mechanical axis and producer’s recommendations to a surgeon concerning an endoprosthesis implantation.

A bone anatomy was reconstructed based on the male cadaver CT scans of the Visible Human Project [8] provid-ed by the National Library of Medicine. A pelvic bone, necessary for muscles origins definition, was adopted from a model library of the BEL Repository, managed by the Istituti Ortopedici Rizzoli, Bologna, Italy [9].

Fig. 2. Lower limb geometric model


44

B. Material properties and mesh generation

For material definitions, only isotropic homogenous ma-terial models were used (see Table 1).

All materials excluding UHMWPE were supposed to be-have according to Hook’s low; the tibial plateau formed from UHMWPE has been defined as an elasto-plastic material (see Fig. 2). All muscles were represented by lines of actions with no material properties definitions. The patellar ligament was modeled as a spring with stiffness of 1000 N/mm.

A mash of elements was created semiautomatically using mixture of hexahedral, tetrahedral and wedge elements. The assembly consisted totally of 338 003 elements.

C. Boundary conditions The only contact defined in the analyses was the one

between the corresponding pairs of the hinge. All other contacts were considered as tie contacts which agreed with the ideal fixation of the TKE to the bone tissue.

Magnitudes of muscle forces were adopted from Vilímek [12] who calculated by a static optimalisation muscle forces for a group of 31 musculotendon actuators. Following 25 muscles took part of the presented FEA: two parts of gluteus medius (GLMED), two parts of gluteus minimus (GLMIN), semimembranosus (SM), semitendinosus (ST), biceps femoris long head (BFL), biceps femoris short head (BFS), sartorius (SR), adductor longus (ADL), adductor breve (ADB), tensor fascia lata (TFL), pectineus (PCT), gracilis (GRC), gluteus maximus (GLMAX), ilio-psoas (ILPS), rectus femoris (RF), vastus medalis (VM), vastus intermedius (VI), vastus lateralis (VL), medial gastrocnemius (MG), lateral gastrocnemius (LG), soleus (SOL), tibialis anterior (TA) and tibialis posterior (TP).

The resulting ground reaction force was also adopted from Vilimek [12].

TABLE I. MATERIAL PROPERTIES

Entity (Material) Young’s modulus [MPa]

Poisson’s ratio [-]

Bones 14 000 0.36

Femoral component (TiAl6V4) 113 800 0,34

Femoral stabilizing rod (TiAl6V4) 113 800 0,34

Tibial component (TiAl6V4) 113 800 0,34

Tibial stabilizing rod (TiAl6V4) 113 800 0,34

Meniscal component (UHMWPE) 820 0.44

Hinge post (TiAl6V4) 113 800 0,34

Hinge pin (TiAl6V4) 113 800 0,34

Hinge lock (TiAl6V4) 113 800 0,34

Hinge bushings (PEEK) 3 650 0.44

All shifts and rotations but flexion were constrained in case

of femoral head.

For the tibia, all rotations in an ankle were allowed as well as the proximal-distal shift.

The patella was allowed to move only in anterior-posterior axis direction which simulated a simplified articular capsule.

The dynamic FEA was run for the flexion of a hip joint up to 69.4 °.

III. RESULTS AND DISCUSSION The main goal of the FEA was to investigate an in-vivo

hinge knee behavior by dynamic load. The interface bone-endoprosthesis has not been studied in this analysis. The results pointed out on the most loaded parts and locations of the endoprosthesis. As supposed, the most critical component of the hinge knee is the PEEK bushing.

Fig. 3 illustrates the deformed assembly at 18.3 ° (9.46 °), 69.4 ° (28.3 °) and 93.1 ° (31.3 °) of the hip joint flexion (ankle joint flexion respectively).

Fig. 1. Assembly at 18.3 °, 69.4 ° and 93.1 ° of the hip joint

flexion

Fig. 3. UHMWPE elasto-plastic material model

Strain [%]

Stre

ss [M

Pa]


45

To evaluate the quality of the FE simulation, we also focused on the comparison of the angle of flexion in the ankle joint calculated by FEA and experimentally measured [12]. The maximum deviation of the angle of 14% was assessed as acceptable (Fig. 4).

ACKNOWLEDGMENT This research is supported by a grant of Technology

Agency of the Czech Republic TACR01010185 and a grant of Ministry of Industry of the Czech Republic FR-TI3/221.

REFERENCES [1] J.M.T. Penrose, “Development of an accurate three dimensional finite

element knee model,” in Comp. Meth. in Biomech. and Biomed. Eng. vol.5, 2002, pp. 291-300.

[2] T.L.H. Donahue, et al, “A finite element model of the Human knee joint for the study of tibio-femoral contact,” in J. Biomech. Eng. vol.124, 2002, pp. 279-280.

[3] J.A. Heegaard, “A computer model to simulate patellar biomechanics following total kneee replacemnet: the effects of femoral component alignment,” in Clinical Biomech. vol.16, 2001, pp. 415-423.

[4] P. Beillas, et al, “A new method to investigate in vivo knee behavior using a finite element model of the lower limb,” in J Biomech. vol.37, 2004, pp. 1019-1030.

[5] A.C. Godest, “Simulation of a knee joint replacement during a gait cycle using explicit finite element analysis,” in J. Biomech. vol.35, 2002, pp. 267-275.

[6] J.P. Halloran, “Explicit finite element modeling of total knee replacement mechanics,” in. J. Biomech. vol.38, 2004, pp. 323-331

[7] ProSpon,s.r.o. at http://www.prospon.cz [8] National Library of Medcine, Visible Human Project at

http://www.nlm.nih.gov/research/visible/visible_human.htm [9] Viceconti, Visible Human Male - Bone surfaces, From The BEL

Repository at http://www.tecno.ior.it/VRLAB/ [10] S.C. White, et al, “A Three Dimensional Musculoskeletal Model for

Gait Analysis. Anatomical Variability Estimates,” in J. Biomech. vol.22, 1989, pp. 885-893.

[11] R.A. Brand, et al., “A model of lower extremity muscular anatomy,” in J. Biomech. Eng. vol.104, 1982, pp. 304-310.

[12] M. Vilimek, “The challenges of musculotendon forces estimation in multiple muscle systems”, PhD Thesis. Prague, Czech Technical University in Prague - Fac. of Mechanical Engineering, 2005.

Fig. 4. Comparison of FEA and experiment [12]


46

Control of Upper Limb Active Prosthesis Using

Surface Electromyography

Muhammad Asim Waris, Mohsin Jamil, Syed Omer Gilani and Yasar Ayaz

School of Mechanical & Manufacturing Engineering (SMME)

National University of Sciences & Technology (NUST)

Islamabad, Pakistan

[email protected], [email protected], [email protected], [email protected]

Abstract— Electromyographic prosthesis with higher

degrees of freedom is an expanding area of research. In

this paper, active prosthesis with four degrees of freedom

has been investigated, which can be used to fit a limb with

amputation below elbow. The system comprises of multi-

channel inputs which correspond to the flexion and

extension as well as supination and pronation. To find

maximum surface neural activity, accurate placement of

electrodes has been carried out on 10 subjects aged

between 22-30 years. Signals (0-500 hertz) acquired from

contracting voluntary muscles with minimum cross talk

and common mode noise. Clean filtered EMG signal is

then amplified precisely. Finally digitization is being done

to drive bionic hand. Practical demonstration on a simple

DC motor proved providential using this method for the

two motions of an actual human arm. EMG Signals

emanating from muscles dedicated to individual fingers

have been recorded. Moreover modern classifiers; KNN

and NN have been investigated carefully with selected

features through different time and noise levels.

Keywords- Electromyography (EMG), flexion, extension,

amplification, supination, pronation, KNN (K nearest neighbor),

NN (neural network).

I. INTRODUCTION

The field of study that deals with the detection (from

needle, cup and surface electrodes), analysis (from picoscope)

and the use of electrical signals (for active prosthesis) is

known as Electromayography (EMG). It involves techniques

for analysis and recording the surface activity which produce

electric potential by skeletal muscles [1]. The device which

records EMG signals is called electromyograph. An

electromayograph detects the electrical potential generated by

muscle cells when these are contracted or neurologically

activated [2].

To develop innovative upper limb prosthesis, the real

challenge is to incorporate the device with intuitive, intelligent

and human like system without losing its functionality having

high deftness [3]. EMG surface activity stimulated by

voluntary contraction of the targeted muscle considered as the

intention of the mayo-electric prosthesis user [4]. Most

commonly Mean Absolute Value (MAV) is used to determine

the intension of the user, in which the absolute value of EMG

signal is compared with predetermined threshold value.

Human body consists of muscles, composed of fibers

having motor points on it. These points when activated

generate motor point active potential. A motor unit (MU) is

defined as an anterior horn cell, its axon and the muscle fibers

innervated by the motor neuron [5]. Motor unit action

potential (MUAP) is a train of pulses or summation of a group

of muscle fiber action potential (MFAP) where superimposed

information of muscle and generated pulses is determined by

each (MFAP).As long as force is maintained or even increased

motor unit generates pulses continuously and resultantly

muscle contracts [6]. Motor points remain activated as long as

muscles are being contracted, This continuous activation of

motor points superimpose to form EMG signal, when muscle

exerts more force greater number of motor points are activated

, so we can conclude that lifting a heavy weight fires more

motor points that lifting a lighter weight.

There are many factors which affect the EMG signals

emanating from the muscles, important aspects are the type of

muscle contraction that is occurring and the type of electrode

used for the detection of these signals, Judicious application of

recognized basis which can accurately identify the innervation

zone with bounding effects of noise and cross talk providing

immotility in signal and normalization of its amplitude which

further enhance the single by removing the effect of many

other variables [7]. EMG signal emanated can be used as a

symptom about: contraction of muscle, force produced when

muscle is being activated and tell us the fatigue variable of the

targeted muscle.

Advance electronics and micro controller based designs

brought new revolution is active prosthesis giving more

degree of freedom to the designer. More movements and

muscle activation can be controlled with the help of filter

algorithms giving more functionality and maneuverability to

the user.

II. MEASURING DEVICES

EMG due to its vast biomedical application becomes an

interest of many researches working in the field of prosthetics

[8]. Non-invasive techniques are the most desirable and

applicable technique for measuring EMG signals with help of

electrode. Which can be categorized in dry and gel type

electrodes [9].

Gelled electrode contains gel between skin and the

measuring electrode. These gelled electrodes are mostly

disposable and not very feasible as it cannot be used for longer

periods of time. Dry electrodes do not contain such medium


47

and can be used for longer periods of time, which are ideal for

active prosthesis.

Dry electrodes are further divided in to two types: active

and passive. Passive electrodes do not require energy or

current for their activation. Active electrodes require energy or

current for their activation, these electrodes often having high

input impedance and pre-amplification circuitry attached to it.

It’s important to prepare test subjects properly before testing

[10]. Skin surface should be clean, hairs should be removed to

avoid artifacts and electrode should be stationed properly

holding its place. Gelled electrodes were used in this research

with self-designed circuitry.

III. SIGNAL ACQUISTION AND POSITIONING OF

EMG ELECTRODES

Positioning of EMG electrode and acquisition of required

signal is an important feature of active prosthesis. Gelled

electrodes which were used in the study were carefully place

on the belly of the muscle, each electrode placed 1-2 cm from

each other. The established location of electrodes is between

innervation zone and the tendinous insertion [11].

Noise reduction is done for further enhancement of the

acquired signal then this signal is send to an instrumentation

amplifier for amplification purposes. Referenced signal must

be acquired, which is isolated electrically and must be

distanced from targeted muscle, the most desirable place for

reference electrode is the neck or it should be unrelated to the

muscles of forearm.

Placement of electrodes on the targeted muscles is an

important part; small difference from the motor point can

bring drastic effect on the amplitude of the signal. For flexion

of hand; flexor digitorum profundus, which fans out in to four

tendons connected to each of four fingers except thumb, is

preferred muscle for the placement of electrode and for the

extension, it should be on extensor digitorium communis.

Individual signals from each finger have also been observed

by placing electrode on the associated muscle of each finger.

Figure1. General placement of surface EMG and reference electrodes.

For flexion of thumb, preferred location of electrode is on

flexor pollicis longus, for its extension electrode should be

placed on extensor pollicis longus. In case of pinkie, middle

and ring finger electrode should be placed precisely on flexor

carpi ulinaris, flexor carpi radialis and flexor palmaris longus

respectively. Similarly for the extension of these fingers,

electrode should be on extensor carpi ulinaris.

Index finger which is the most dexterous and sensitive

finger controlling several motions of hand, for its flexion,

flexor digitorum superficialis is the preferred location of

electrode placement and for extension, recommended muscle

is extensor inidcis.

IV. NOISE REDUCTION TECHNIQUE

EMG signal has very low signal to noise (SNR) ratio, many

factor bring about these disturbances, and one of the major

portion of these noises are cardiac artifacts or

Electrocardiography (ECG). Non-stationary nature of EMG

signals keeps the amplitude ratio between EMG signals and

cardiac artifacts variable.

A signal processing technique based on finite impulse

response (FIR) adapter filter can be employed to reduce noise

in which multi-electrode array is used for signal acquisition

purposes [12]. In this method referencing is done with respect

to ECG signals then adaptive filter is applied to reduce power

line disturbances.

A. Noise Reference Estimation

Acquired signal Z (t) at the surface of the skin composed of

EMG signal ( )in t and ECG signal ( )is t . Band filter (20Hz:

40Hz) is then applied to the summation of the signal.

1

( ) [ ( ) ( )]N

i i

i

Z t n t s t

1 1

( ) ( ) ( )]N N

i i

i i

Z t n t s t

(1)

Where

( )in t Acquired Electromyographic (EMG) signal

( )is t Acquired Electrocardiographic (ECG) signal

N= Electrodes attached to limb

Autocorrelation function is given as

Putting equation 1 in 2

( ) [ ( ) ( )]zz E Z t Z t (2)

1 1 1 1

( ) ( ) ( ) ( ) ( )N N N N

zz i i i i

i i i i

E n t s t n t s t

, , , ,

( ) ( ) ( ) ( ) ( )l m l m m l l m

N N N N

zz n n n s n s s s

l m l m l m l m

As ECG and EMG signal has no co-relation so above equation

reduces.

, ,

( ) ( ) ( )l m l m

N N

zz n n s s

l m l m

High correlation of ECG signal at electrode channel yields

l m


48

2

1 ,

( ) ( ) ( ) ( )l l l m

N N

zz n n n n ss

i l m

N

2

,

( ) ( ) ( ) ( )l m

N

zz ss nn n n

l m

N N

The above equation proves that the signal to noise ratio

increase N times greater if we use array of electrode for the

acquisition of EMG signals.

B. Adaptive Filtering

Power line interference is one of the major causes which

decrease the quality of EMG signal significantly [13].

Adaptive filter can be really helpful in attenuating these

interferences.

Raw EMG signal contains power line interferences under

60Hz. If we know the characteristic of noise, a filter can be

designed to reduce that noise with high efficiency [14].

( ) ( ) ( ) ( ) ( ) ( )y k x k n k s k n k n k

Where

( )y k Output signal of noise canceler

( )x k Input raw signal

( )n k

Noise estimate

( )n k Noise influence on input raw signal ( )x k

With the help of noise estimate, noise influence on the

acquired signal can be minimized; the estimate noise can be

deduced by finite impulse response (FIR) adaptive filter

through which a sample of noise is given as input to the

controller. 1

0

( ) ( ). ( )N

i

i

n k w k r k i

1N Order of the filter

( )iw k Adjustable filter coefficients

( )r k Sample of noise or reference input

Digital

Filter

Adaptive

Algorithm

ˆ ( )n k

( )x k

( )r k

( )y k

Figure2. Block diagram of Adaptive filter

V. MODERN CLASSIFICATION TECHNIQUES

After the amplification, classification of these EMG signals

is done to identify the user’s intended motion. Many authors

investigated these classification strategies Sebelius et al. [15]

and Paul et al. [16] are one of those who faced and

investigated the issue of real-time implementation of

artificial active prosthesis. Segmentation, in accordance with a

flexion of the dedicated muscle, Pattern recognition, Feature

extraction, Classification of signals and simulating actual

prosthesis were the main subject of issue during the study

[17]. Here Neural Network and K nearest neighbor classifiers

have been studied.

A. Neural Network Classifier

In the recent past most of the study has been done on

multichannel signal processing. The electromyogrphic

signals from the multi-channel data acquisition system will

increase the classification efficiency with the increase in

classification accuracy but with increase of diminishing

effect in signals if the number of channels is increase to 4 or

more [18].

Many researchers have chosen multi-channel that is

multiple electrode can be used to perform some specific

function with only designated electrodes but some want to

move further ahead by leaving aside this strategy.

Furthermore the number of classes can be increased to

increase the classification accuracy. It is understood that the

accuracy will decrease because of the nature of accuracy and

when the output data flowing through different channels

increase which affects the quality of signal by affecting its

feature space. Therefore increasing number of channels will

certainly affect its feature space assimilated with each class

[19].

A back-propagation neural network is the solution of the

discussed problem, in which the EMG signal are acquired for

different hand movement earlier defined which can be flexion,

extension, supination and pronation or else. Calculated Time

frequency based parameters can be used as input to this

classifier ,Which can be Wavelet transform, Moving Average,

Auto regression, Root Mean Square, Fast Fourier Transform,

Variance, Standard deviation, Slope Sign Change, Willson

Amplitude, Zero crossing, Wave Length. Selection of these

features is the most important part of neural theory. Selecting

relevant features gives the pattern which then can further be

easily classified.

B. K Nearest Neighbor Classifier

Neural Network classifier is considered as slow and time

consuming. K nearest neighbor can be used to get accurate

results in span to time. In this classification technique, the

reference vectors from all the required motions can be used to

calculate the distance between the input vectors of present

state [20].

KNN first assigns class to the un-known events which

represents majority of its nearest neighbors. Assignment of

class is based on most suitable pattern nearest to the system

measured on the basis of Eculidean distance. Labeling is done

to classify segment that is most frequently represented among

the K nearest neighbor. At the end decision is made on the

basis of taking a vote and by examining the labels.

Discriminative approach been employed in KNN which is

more suitable when reliable probabilistic densities are difficult

to find.


49

VI. EXPERIMENTATION AND PROCEDURE

Acquired signal from the electrodes have frequency that

ranges between 0-500 Hz which have noise from different

sources such as cross talk, artifacts and above all power line

sources. These noise ranges from 50-60 Hz which has to be

removed before amplification is done. Signal is acquired from

two electrodes which are 1-2cm apart from each other; the

signal common to both is rejected with the help of differential

amplifier. The amplitude of the acquired signal ranges from 0-

20mV.

Amplification is one of the most important steps in active

prosthesis done with the help of instrumentation amplifier; this

amplification can be achieved with INA 121. It is an IC which

has vast applications in biomedical field, which has ability to

amplify up to 10,000 times [21]. Once the detection of EMG

signals is achieved through sensing electrodes, differential

technique is employed with the help of operational amplifier

to achieve first step in amplification [22].

Figure3. Differential Amplifier is device which rejects the common signal

of both the inputs provided to it.

Acquired analog signal whose frequency ranges form 50-

200 Hz and amplitude varies form 0-5 volt obtained after

carrying out the above mentioned procedure. This analog

signal has to be converted in to digital signal with the help of

Analog to Digital Converter (ADC) which is commonly used

in modern electronics. During digitization following things

have to be kept in mind resolution, range of conversion and

the sampling rate. The maximum voltage which an ADC can

convert in to digital format is known as range of conversion.

Sampling rate is kept high for the minimum loss of data;

dynamic range of conversion is kept high which keep the

amplification output small.

In the study 16 bit ADC has been used which come as a

peripheral with ATMEGA16 microcontroller [23]. This has

multiple channels with on chip-2 cycle multiplier. Above

information completely fulfill our requirement for amplified

signal between 1-4.8 volts. As flexion and extension of hand

take place, signal after amplification is put into mentioned

microcontroller for further digitization and motor control.

ADC convert amplified analog signal it to digital format,

where reference of the ADC is given 5 volts.

Thresholding technique is applied for the controlling

different movements. Peak values are measured from each

targeted muscle and threshold value is selected, threshold

value must be 2 volts less than the peak value in case of

flexion and extension. As the targeted muscles fire, the motor

unit get excited, signal detected by the gelled electrodes, pre-

amplification is done with differential amplifier, noise is

reduced by using high pass filter, then the amplification is

done with mentioned instrumentation amplifier, ADC convert

this amplified analog signal in to digital signal which is

between 0-5 volts and send it further for processing.

Comparison is being done between threshold value and the

system value, as it passes the threshold value it gives the

output as 1 which moves the motor in the desired direction.

Same procedure can be repeated for desired motion of

supination and pronation by assigning different motor points

on the targeted muscle. With accurate placement of electrodes,

we can able to differentiate between the motions for all the

fingers, hence giving more functionality as well degree of

freedom to the patient. Here are some experimental results

after finger classification being done with the help of gelled

electrodes.

S.No

Fingers and

Hand

movements

Acquired peak

voltage level

before

contracting

Acquired peak

voltage level

after

contracting

1 Hand Flexion 0.16 V 4.8 V

2 Hand

Extension

0.12 V 4.6 V

3 Index finger

Flexion

0.5 V 1.8 V

4 Thumb

Flexion

0.6 V 3.8 V

5 Ring finger

Flexion

0.2 V 3.1 V

TABLE: VOLTAGE LEVELS AFTER AMPLIFICATION

Table shows different voltage levels from different muscles of

hand and fingers which were explained earlier, these signals

were acquired after amplification, we can set the threshold

between these two peak values for which the motor can be

derived for the predefined value, that value should be nearer to

peak value contracting value.

ELECTRODE 1

ELECTRODE 2

PRE-AMPLIFICATION

(DIFFERENTIAL

AMPLIFIER)

NOISE REDUCTION

(BAND PASS FILTER)

AMPLIFICATION

(INSTRUMENTATION

AMPLIFIER)

DIGITIZATION

(ADC)

MICROCONTROLLER

(ATMEGA16)

MOTOR DRIVE

(DC MOTOR)

ARTIFICIAL LIMB

(BIO-MECHANICAL)

REFERENCE

ELECTRICALLY

ISOLATED

Figure4. Block diagram indicating all step involved in active prosthesis


50

The amplitude of EMG signal emanating from belly of the

muscles varies person to person. A muscular man can emanate

EMG signal of more amplitude than a normal man.

Acquisition of superimposed signals, amplitude from the

contracting muscles will be higher as compared to isolated

muscle that is valid in case of flexion and extension. It will

have low amplitudes when placed on the dedicated muscle for

each fingers. The given study is taken on normal built.

VII. CONCLUSION

A novel method for flexion and extension of mechanical

arm is successfully achieved with the help of EMG. Electrode

placement on the targeted muscle for the clean signal is

necessary. During study it was revealed that each muscle form

a set pattern of signal whenever the flexion or extension take

place. Extracting the statistical features from EMG signals for

classifying the motions through described classifiers, different

wavelet function can be used for enhancing the classification

rate. Amplitude of EMG signal varies as number of muscles

fire increase or decrease. This study is a step forward towards

achieving active prosthesis which is not only light weight, cost

effective but also a successful replacement of upper limb

amputation. Noise reduction techniques have been emphasized

for future work.

REFERENCES

[1] A.G. Outten, S.J. Roberts and M.J. Stokes “Analysis of human muscle

activity”, Artificial Intelligence Methods for Biomedical Data

Processing, IEEE Colloquium, London, 26 April,1996.

[2] M.L. Harba and G.E. Chee “Muscle Mechanomyographic and

electromyographic signals compared with reference to action potential

average propagation velocity”, Engineering in Medicine and Biology Society, 19th Annual International Conference of the IEEE, Vol.3, and

6th August 2002.

[3] G. Matrone, C. Cipriani, M. C. Carrozza and G. Magenes “Two-Channel Real-Time EMG control of a Dexterous Hand Prosthesis”, 5th

International IEEE EMBS Conference on Neural Engineering, Cacun,

Mexico, April 27-May 1, pages: 554-557, 2011. [4] G.W. Choi, G.H. Choe, I.H. Moon and M.S. Mun

“Development of Surface Mayoelectric Sensor for Mayoelectric Hand

Prosthesis”, Power Electronics Specialists Conference, 2006. PESC ’06. 37th IEEE, Pages 1-5, 2006.

[5] P. Zhu “Design of Surface Electromayography Detection Circuit”,

International Conference on Future Information Technology and Management Engineering, volume.1, pages: 459-462, 2010.

[6] S. Shahid, J. Walker “Application of higher order statistics techniques to

EMG signal to characterize the motor unit action potential”,IEEE Transactions on Bio-medical Engineering, Vol.52, NO.7,July 2005.

[7] C.J. De Luca “Electromayography Encyclopedia of Medical Devices

and Instrumentation”, (John G. Webster Ed), John Wiley Publisher, 2006.

[8] C.N. Huang, C.H. Chen and H.Y. Chung “The Review of Applications

and Measurements in Facial Electromayography”, Journal of Medical and Biological Engineering, Vol.25, 23rd November, 2004.

[9] Dr. Scott Day “Important Factors in Surface EMG Measurement”,

Bortec Biomedical Incorporated. [10] N. Masso, F. Rey, D. Romero, G. Gual, L. Costa and A. German

“Surface Electromayography and Applications in Sport” Apunts

Medicina De L’Esport, Vol.45:127-136, February 5, 2010. [11] C.J. De Luca “Surface Electromayography: Detection and Recording”,

Delsys Incorporated, 2002.

[12] S. Yacuab, P.Y. Gumery, K. Raoof “A Novel Signal Processing Method

for Multi-electrode Surface Electromayography”, Engineering in Medicine and Biology Society. Engineering Advances: New

Opportunities for Biomedical Engineers, Proceedings of the 16th Annual

International Conference of the IEEE, Vol.2, pages 1336-1337, 1994. [13] M. Malboubi, F. Razzazi, Ma. Sh, A. Davari “Power Line Noise

Elimination from EMG Signals Using Adaptive Laguerre Filter with

Fuzzy step size”, 17th Iranian Conference of Biomedical Engineering, pages 1-4, 2010.

[14] R.L. Ortalon, R.N. Mori, R.R. Pereira, C.M.N. Cabral, J.C. Pereira, J.A

Cliquet “Evalution of Adaptive/Nonadaptive Filtering and Wavelet Transform Techniques for Noise Reduction in EMG Mobile Acquisition

Equipment”, Transactions on Neural Systems and Rehablitation

Engineering, IEEE, Issue: 1, pages 60-69, 2003. [15] F. Sebelius, M. Axelsson, N. Danielsen, J. Schouenborg, and T. Laurell,

“Real Time Control Virtual Hand”, Technology And Disability”, vol.17

no. 3, pages 131-141, 2005. [16] J. Pons, R. Ceres, E. Rocon, S. Levin, I. Markovitz, B. Saro, D.

Reynaerts, W.V. Moorleghem and L. Bueno,”Virtual Reality Training

and EMG Control of the MANUS Hand Prosthesis”, Robotica, vol.23, no.03, pages 311-317, 2005.

[17] F.E.R. Mattioli, E.A. Lamounier, A. Cardoso, A.B. Soares, A.O.

Andrade.”Classification of EMG signals using Artificial Neural Networksfor Virtual hand Prosthesis Control”, Annual International

Conference of the IEEE, Engineering in Medicine and Biology Society,

Pages 7254-7257, 2011. [18] G. Tsenov, A.H. Zeghbib, F. Palis, N. Shoylev and V. Mladenov

“Neural Networks for Online Classification of Hand and Finger Movements using Surface EMG signals”, in Neural Networks

Applications in Electrical Engineering, 2006,NEUREL 2006 .8th

Seminar, pages 167-171, 2007. [19] M.R. Ahsan, M.I. Ibrahimy, O.O. Khalifa “EMG Motion Pattern

Classification through Design and Optimization of Neural Network”,

International Conference on Biomedical Engineering, ICOBE, pages 175-179, 2012.

[20] P. Geethanjali, K.K. Ray, P.V. Shanmuganathan “Actuation of

Prosthetic Drive Using EMG”, TENCON 2009, IEEE Region 10 Conference , pages 1-5, 2009.

[21] Datasheet INA121 “FET input Low Power Instrumentation Amplifier”,

Burr-Brown Incorporated. [22] Y. Shimomura, K. Iwanaga, H. Harada and T. Katsuura “Evaluation and

Design of a Small Portable EMG Amplifier”. Volume: 18, 61-67, 1999.

[23] Datasheet ATMEGA16 “8-bit AVR microcontroller with 16kBytes In-System Programmable Flash”, ATMEL Incorporated.


51

Development of a System for Measurement on Asymmetric Sitting Posture

Ji-Yong Jung Department of Healthcare Engineering

College of Engineering, Chonbuk National University Jeonju, Republic of Korea

[email protected]

Yonggwon Won School of Electronics and Computer Engineering

College of Engineering, Chonnam National University Gwangju, Republic of Korea

[email protected]

In-Sik Park Department of Healthcare Engineering

College of Engineering, Chonbuk National University Jeonju, Republic of Korea

[email protected]

Tae-Kyu Kwon Division of Biomedical Engineering

Chonbuk National University Jeonju, Republic of Korea

[email protected]

Jung-Ja Kim Division of Biomedical Engineering

Chonbuk National University Jeonju, Republic of Korea

[email protected]

Abstract— Sitting posture measurement system using the unstable board with accelerometer was developed. And, postural balance was assessed to determine the effect of asymmetry on sitting posture between patients with pelvic asymmetry and healthy subjects. 10 subjects (pelvic asymmetry patients:5, healthy controls:5) were participated in this study. We performed experiment under static and dynamic sitting condition. Angular variation in the anterior-posterior and left-right direction was measured in both two conditions. Also, intra class correlation coefficient was used to evaluate the reliability of the system. The value of angle of pelvic asymmetry patients was more tilted significantly to the left side than right side during static and dynamic sitting. The reliability of the system was excellent. This paper suggested that a system for measurement on asymmetric sitting posture can be utilized to provide useful information about patients with pelvic asymmetry in rehabilitation medicine. Furthermore, results from this study can be used to develop the new clinical quantitative measurement system.

Keywords—asymmetric sitting posture; pelvic asymmetry; leg length discrepancy; accelerometer; unstable board

I. INTRODUCTION Postural control is a processing in which complex

interaction about various tissues inside the human body and external force is generated [1], [2]. Maintaining correct posture is essential to provide normal biomechanical function of the body effectively in daily life. Postural imbalance which is associated with habitual bad posture during sitting may result

in low back pain (LBP), scoliosis, and musculoskeletal disorder caused by asymmetry of pelvis and trunk muscle [3]-[5]]. Also, bad sitting posture over a long period of time can lead to long term complications such as osteoarthritis [6].

Prevalence of patients with pelvic asymmetry induced by leg length discrepancy (LLD), is defined as a condition in which a disparity of length between the legs, increased by approximately 40~70% in the general population [7]. Several studies have suggested that LLD cause asymmetry in the lower extremity and pelvis, leading to arthritic changes in the lumbar spine, LBP, pelvic tilt, altered lordosis, and postural change, depending on the discrepancy of 10mm or less [8]-[11]. If the postural asymmetry leads to changed movement patterns that might negatively affect the individual’s activities, then there is a need to better understand about this.

Most people actually have mild asymmetry with no noticeable symptoms and they spend more time sitting with change of working conditions. Prolonged sitting during working can influence on forming bad sitting posture, and it is connected to continuous functional damage to balance control system [12]. Although there is evidence that how pelvic asymmetry affect the postural stability in static standing [13-15], sitting [16], [17], and during walking [18], [19], the differences of postural balance in unstable sitting posture between symptomatic and asymptomatic person has not been studied.

In recent years, many studies related to the measurement of physical activity using accelerometer have been conducted [20-


52

21]. Accelerometers are commonly utilized to detect the motion in various clinical fields as its convenience and effectiveness. Bliley et al. [22] measure body posture and movement using MEMS accelerometer. Luo et al. [23] developed posture monitoring system based on an accelerometer for training people to improve posture and demonstrated that this device can be used to detect postural changes. Curone et al. [24] detected human activity using new algorithm based on real-time three-axis accelerometer data placed on the trunk. Clinical quantitative measurement system using accelerometer which is a promising technique can be used to provide postural information while sitting for individual and to prevent progression in patients with pelvic asymmetry.

The aim of this study was to assess the seated balance of patients with pelvic asymmetry and healthy subjects using new measurement system with accelerometer. Also, we confirmed the reliability of this system for evaluation its usefulness in clinical medicine.

II. METHODS & MATERIALS

A. Measurement Insturment Shape and appearance of sitting posture measurement

system was hemisphere (radius: 320 mm), creating instability and maximum around 20 degrees in all directions as shown in Fig. 1. Seat surface of this unstable board was covered with soft material to provide comfort during sitting. MEMS accelerometer (MMA7331L, Freescale Semiconductor Inc., Austin, Texas) measure acceleration in a range of ±4 g and sensitivity was about 86.3 mV/g. And, it was positioned to middle bottom of the board. This position of sensor facilitated measurement on neutral and asymmetry sitting posture. Photo sensors (SG-23FF, Kodenshi Co., Tokyo, Japan) were also attached to the surface of both sides in board to check sitting state of subjects by measuring the gap between the tip and the plate.

Accelerometer output included the acceleration of gravity, vibration, and acceleration transformation. Therefore, a third-order digital finite impulse response (FIR) low-pass filter at 2 Hz was used to correct the sensor output. Tilting angle was calculated using acceleration of gravity.

B. Subjects 5 male pelvic asymmetry patients (PA) and 5 male pelvic

symmetry subjects (PS) were participated in the experiment. Their mean age was 14.4±1.34 years, mean height 165.8±10.54 cm, and mean body mass 61.4±12.48 kg. The patients, which were diagnosed pelvic asymmetry with LLD, were recruited from an outpatient foot clinic. Height of the right pelvis was larger than left pelvis and the difference of length between the legs was 6.99±2.91 mm. Subjects in the PA group were excluded if they had pain of the lower extremity, had experience of pelvic or LLD correction, or had any postural training. Subjects in the PS group had no history of injury in the musculoskeletal system or disease related to asymmetry of the lower extremity. All subjects were informed a full explanation regarding the protocol and provided written consent prior to their participation.

C. Experimental Protocol

To measure the asymmetry of sitting posture, experiment procedure was divided into two conditions: static and dynamic sitting. In static sitting condition, subjects were instructed to sit in their usual manner on the sitting posture measurement system, which is located in the center of stool, with their arms crossed on contra-lateral shoulders for 30 seconds as shown in Fig. 2. In dynamic sitting condition, subjects were asked to perform anterior, posterior, left, and right pelvic rotation with trying to fix their upper trunk, and then sitting posture was hold for 5 seconds, respectively, as shown in Fig. 3. A foot support was used to prevent the influence of leg movement, and it was adjusted to support the feet by keeping knee and ankle angles at 90° [25]. Before the experiment, all subjects practiced all testing procedures until they could understand about all postures. To prevent fatigue, subjects took a 5 minute rest in between experiments.

D. Data Analysis Angle variation data (sampling rate: 100 samples/s) in the

frontal and sagittal planes collected by sitting measurement system were analyzed using LabVIEW 2010 (National Instrument Co., Austin, Texas). X-axis was presented to right (+) and left (-) direction in frontal plane, and Y-axis was presented to anterior (+) and posterior (-) direction in sagittal plane. The COP (center of pressure) of the subject was computed using accelerometer.

Fig. 2. Static sitting posture

Fig. 1. Sitting posture measurement system


53

To assess the repeatability of the system, the average distance of the COP sway path and total COP sway area was analyzed [26]. COP sway path and sway area was calculated as follow Eq. (1) ~ (6):

∑−

=++ −+−=

1

1

211 )()(

N

n

MLn

MLn

APn

APn COPCOPCOPCOPCOPpath (1)

22 )()( MLn

APnn COPCOPa += (2)

21

21 )()( ML

nML

nAP

nAP

nn COPCOPCOPCOPb −+−= ++ (3)

21

21 )()( ML

nAP

nn COPCOPc ++ += (4)

2nnn

ncba

S++

= (5)

2nnn

ncba

S++

= (5)

∑−

=

−⋅−⋅−⋅=1

1

)()()(n

nnnnnnnn cSbSaSSCOParea (6)

where N was the total number of samples, n was the sample number, an was the length from center (0,0) to n, bn was the length from n to n+1, cn was the length from center (0,0) to n+1, and sn was total sum of an, bn, and cn divided by 2. Data from all 3 trials were analyzed.

Statistical analysis was performed using SPSS 18.0 statistical software (SPSS Inc., Chicago, IL). Independent t-test was used to examine the difference in angle variation between PA and PS group, at p < 0.05 level. Also, intra-class correlation coefficient (ICC) was analyzed to evaluate the reliability of the system.

III. RESULTS

A. Static Sitting Posture Mean angular variation in anterior-posterior (AP) and left-

right (LR) direction is shown in Fig. 4 and 5, respectively. Mean angle of both groups were tilted to posterior and left side. Tilting angle of PA group was smaller than PS group in AP direction. In contrast, tilting angle of PA group was significantly larger than PS group in LR direction (p=0.013).

B. Dynamic Sitting Posture There was a difference in angular variation between PA and

PS group during anterior and posterior pelvic rotation as shown in Fig. 6. Posterior pelvic tilt angle of PA group and anterior pelvic tilt angle of PS group was larger than pelvic tilt angle in angle between PA and PS group.

Fig. 6. Anterior and posterior tilt angle in dynamic sitting

Fig. 5. Left and right tilt angle in static sitting

Fig. 4. Anterior and posterior tilt angle in static sitting

Fig. 3. Dynamic sitting posture

(a) Anterior pelvic rotation, (b) Posterior pelvic rotation, (c) Left pelvic rotation, (d) Right pelvic rotation


54

Difference in angular variation during left and right pelvic

rotation is presented in Fig. 7. The major difference in the angle was evident in the left and right pelvic tilt of PA group. In PA group, value of angle was significantly more tilted to left than right side while there is a little difference in tilting angle between left and right side in PS group

C. Repeatability of the system Average distance of COP sway path and area did not differ

significantly in the first test as compared with the second test as shown in Fig. 8. By observing the ICC of COP sway path and area between first and second test for static and dynamic sitting posture, we could confirm the reliability of the system. The ICC values for COP sway path ranged from 0.88 to 0.96 and COP area ranged from sitting posture measurement system ranged from 0.81 to 0.95.

I. DISCUSSTION We compared static and dynamic sitting posture between

PA and PS group using measurement system with accelerometer. Height of right pelvis was high compare with left pelvis in PA group. In static sitting, PA group showed an asymmetry which is more tilted 1.01° to left side than right side, and there is a significant difference between both groups.

Postural asymmetry during static sitting may affect to perform anterior, posterior, left, and right pelvic rotation. Significant angular difference in LR direction was 1.97° with left tilt in PA group during dynamic sitting. Generally, in case of difference in length exist between the legs, length asymmetry is not reduced while it can cause pelvis deformation in frontal and sagittal plane, and negatively influence on standing posture as well as sitting posture by compensation [27], [28]. From these results, PA group have asymmetrical balance induced by pelvic asymmetry during static and dynamic sitting.

It has been suggested that ICC values in a range of 0.75 to 1.00 was considered as excellent; 0.60 to 0.74 as good; 0.40 to 0.59 as fair; less than 0.40 as poor [29]. Our results showed excellent reliability in both static and dynamic sitting. Test-retest reliability of the measurement system demonstrated high ICC values ranged from 0.81 to 0.96. It means that this sitting posture measurement system may be useful for measuring the postural asymmetry.

II. CONCLUSION In this study, we developed sitting posture measurement

system using accelerometer and evaluated the difference in posture between patients with pelvic asymmetry and healthy subjects during unstable sitting. The value of angle was tilted with the degree of asymmetry of the pelvis in both static and dynamic sitting condition. The reliability results for the measurement system were excellent. The results indicate that measurement system for asymmetric sitting posture can be used to assess the sitting postural balance for individual and to evaluate the postural change of patients with disease such as LBP, scoliosis, and disc. Further research is required to compare the trunk muscle activation pattern and kinematics between patients with pelvic asymmetry and normal subjects.

ACKNOWLEDGMENT This research was supported by Basic Science Research

Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (NRF-2012R1A1B3003952).

REFERENCES [1] A. Shumway-Cook and M. Woollacott, Motor control: translating

Research into Clinical Practice, 3rd ed. Philadelphia: Lippincott Williams & Wilkins, 2006.

[2] A. Shumway-Cook and M. Woollacott, “Attentional demands and postural control: the effect of sensory context,” J Gerontol A Biol Sci Med Sci, vol. 55, no. 1, pp. M10-16, Jan. 2000.

[3] I. Kingma and J. H. van Dieen, “Static and dynamic postural loading during computer work in females: sitting on an office chair versus sitting on an exercise ball,” Appl Ergon, vol. 40, no. 2, pp. 199-205, Mar. 2009.

[4] D. Lanzetta, D. Cattaneo, D. Pellegatta, and R. Cardini, “Trunk control in unstable sitting posture during functional activities in healthy subjects and patients with multiple sclerosis,” Arch Phys Med Rehabil, vol. 85, no. 2, pp. 279-283, Feb. 2004.

[5] L. E. Dunne, P. Walsh, S. Hermann, B. Smyth, and B. Caulfield, “Wearable monitoring of seated spinal posture,” IEEE Transactions on Biomedical Circuits and Systems, 2008, vol. 2, no. 2, pp. 97-105.

[6] L. Twomey and J. Taylor, Physical therapy of low back, 3rd ed. Austrailia: Churchill Livingstone, 2000.

[7] E. Al-Eisa, D. Egan, K. Deluzio, and R. Wassersug, “Effects of pelvic skeletal asymmetry on trunk movement: three-dimensional analysis in

Fig. 7. Left and right tilt angle in dynamic sitting (*p < 0.05)

Fig. 8. Test-retest of COP sway path and area (*p < 0.05)


55

healthy individuals versus patients with mechanical low back pain,” Spine, vol. 31, no. 3, pp. E71-79, Feb. 2006.

[8] B. Gurney, “Leg length discrepancy,” Gait & posture, vol. 15, no. 2, pp. 195-206, Apr. 2002.

[9] G. Cummings, J. P. Scholz, and K. Barnes, “The effect of imposed leg length difference on pelvic bone symmetry,” Spine, vol. 18, no. 3, pp. 368-373, Mar. 1993.

[10] D. O. Mannello, “Leg length inequality,” J Manipulative Physiol Ther, vol. 15, no. 9, pp. 576-590, Mar. 1992.

[11] R. L. Blake and H. Ferquson, “Limb length discrepancies,” J Am Podiatr Med Assoc, vol. 82, no. 1, pp. 33-38, Jan. 1992.

[12] H. C. Wu and T. Chen, “Developing intelligent software for diagnosing computer-related health issues,” in 5th WSEAS International Conference on Applied Computer Science, 2006, pp. 832-837.

[13] A. Mouzat, M. Dabonneville, P. Bertrand, and P. Vaslin, “Postural asymmetry in human stance: the mean center of pressure position,” in 2001 23rd Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), 2001, pp. 1159-1162.

[14] T. Sims, T. W. Findley, Hans. Chaudhry, Ji. Zhiming, and B. Bukiet, “Postural stability: mathematical modeling to evaluate interventions,” in 2003 29th Annual Northeast Bioengineering Conference, 2003, pp.209-210.

[15] M. D. Bussey, “Does the demand for asymmetric functional lower body postures in lateral sports relate to structural asymmetry of the pelvis?,” J Sci Med Sport, vol. 13, no. 3, pp. 360-364, May. 2009.

[16] E. Al-Eisa, D. Egan, K. Deluzio, and R. Wassersug, “Effects of pelvic asymmetry and low back pain on trunk kinematics during sitting: a comparison with standing,” Spine, vol. 31, no. 5, pp. E135-143, Mar. 2006.

[17] E. Al-Eisa, D. Egan, and A. Fenety, “The association between lateral pelvic tilt and asymmetry in sitting pressure distribution,” J Man Manip Ther, vol. 12, no. 3, pp. 133-142, Sept. 2004.

[18] K. R. Kaufman, L. S. Miller, and D. H. Sutherland, “Gait asymmetry in patients with limb-length inequality,” J Pediatr Orthop, vol. 16, no. 2, pp. 144-150, Mar. 1996.

[19] F. G. Delacerad and O. D. Wikoff, “Effect of lower extremity asymmetry on the kinematics of gait,” J Orthop Sport Phys Ther, vol. 3, no. 3, pp. 105-107, 1982.

[20] S. Sprager and D. Zazula, “Gait identification using cumulants of accelerometer data,” in 2nd WSEAS International Conference on Sensors, and Signals and Visualization, Imaging and Simulation and Materials Science, 2009, pp. 94-99.

[21] J. Musić, R. Kamnik, V. Zanchi, and M. Munih, “Human body model based inertial measurement of sit-to-stand motion kinematics,” WSEAS Trans. System, vol. 7, no, 3, pp. 156-164, Mar. 2008.

[22] K. E. Bliley, D. J. Schwab, D. R. Holmes, P. H. Kane, J. A. Levine, E. S. Daniel, and B. K. Gilbert, “Design of a compact system using a MEMS accelerometer to measure body posture and ambulation,” in 19th IEEE International Symposium on Computer-Based Medical Systems (CBMS), 2006, pp. 335-340.

[23] E. Lou, M. Bazzarelli, D. Hill, and N. Durdle, “A low power accelerometer used to improve posture,” in Proceeding of the IEEE Canadian Conference on Electrical & Computer Engineering, 2001, pp. 1385-1389.

[24] D. Curone, G. M. Bertolotti, A. Cristiani, E. L. Secco, and G. Magenes, “A real-time and self-calibrating algorithm based on triaxial acclerometer signals for the detection of human posture and activitiy,” IEEE Transaction on Information Technology in Biomedicine, 2010, vol. 14, no. 4, pp. 1098-1105.

[25] J. H. van Dieen, L. L. Koppes, and J. W. Twisk, “Postural sway parameters in seated balancing; their reliability and relationship with balancing performance,” Gait & Posture, vol. 31, no. 1, pp. 42-46, Jan. 2010.

[26] J. Han, Z. Moussavi, T. Szturm, and V. Goodman, “Application of nonlinear dynamics to human postural control system,” in 2005 27th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), 2005, pp. 6885-6888.

[27] O. Friberg, “Clinical symptoms and biomechanics of lumbar spine and hip joint in leg length inequality,” Spine, vol. 88, no. 6, pp. 643-651, Sep. 1983.

[28] J. P. Gofton, “Persistent low back pain and leg length disparity,” J Rheumatol, vol. 12, no. 4, pp. 747-750, Aug. 1985.

[29] A. J. Ronchi, M. Lech, N. F. Taylor, and I. Cosic, “A reliability of the new back strain monitor based on clinical trials,” in 30th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBS), 2008, pp. 693-696.


56

Abstract- Carbonic anhydrase (CA; EC 4.2.4.1), metallo-enzyme, can catalyze reversible hydration of CO2 (CO2 + H2O ↔ H+ + HCO3

-) with high efficiency (kcat ~106 s-1) and plays fundamental roles in many biological processes like photosynthesis, respiration, pH homeostasis and ion transport. Recently, CA has been considered as an important biocatalyst for CO2 sequestration technology because the accumulation of CO2 is the main cause for global climate change and it is critical to develop technologies that can reduce atmospheric CO2 level. This review deals with the classes and mechanisms of several CAs as CO2 capture agents.

Keywords— Carbonic Anhydrase; Classes, Mechanism

I. INTRODUCTION ince 20th century, industrialization and civilization have developed extensively. Anthropogenic interventions have

caused a rise in concentration of greenhouse gases, especially, carbon dioxide (CO2) which has led to several undesirable consequences such as global warming and related changes [1]. The most easily addressable source of CO2 is from the coal-fired power plants, which currently account for almost the half of the electricity generation. Due to industrial progress, concentration of CO2 in the atmosphere is still increasing and it is unlikely that there will be a dramatic change in this situation in near futures. The outcome of this increase has already witnessed a profound effect on global environment. The climate changes and increasing environmental awareness have been driving global policy maker to find a solution for global warming [2, 3]. Thus, actions are being taken to alleviate the greenhouse gas emissions, especially in the case of large point source emissions, through various mechanisms and protocols.

In technology, there are various methods of reducing CO2

concentration in the atmosphere. These methods are normally classified as (1) reduction of formation of CO2 and (2) reduction of emission of CO2. In case of reducing CO2 emissions, one

This work was supported by the Marine biomaterials Research Center grant from Marine Biotechnology Program funded by the Ministry of Land, Transport and Maritime Affairs, Korea..

promising approach is CO2 capture and storage (CCS)- CO2 is captured at a power plant and sequestered for long-term storage in any of a variety of suitable geologic forms [4]. CO2 capture from coal-fired power generation can be achieved by various approaches: post-combustion capture, pre-combustion capture, and oxy-combustion. A wide variety of separation techniques is being pursued, including gas phase separation, absorption into a liquid, and adsorption on a solid, as well as hybrid processes, such as adsorption / membrane systems. Although these approaches may be prominent, it is still too costly, so the efficiency of capture/sequestration should be increased. Much effort should be made not only for improving the state-of-the-art technologies but also for developing several innovative concepts, such as metal organic frameworks, ionic liquids, and enzyme-based systems.

Biologically based CO2 capture systems are one of innovative approaches to improve the capture efficiency. These systems are based upon highly-efficient biological CO2 mechanisms, which are naturally-occurring in living organisms. One of the most prominent possibilities is the use of the enzymes involved in such biological CO2 reactions; the respiratory system in mammalian cells or photosynthetic systems in plant cells. On the basis of CO2-catalyzing enzymes, we could make “bio-mimic” CO2 capture systems, which can show high efficiency or performance in capture and release of CO2

comparable to biomechanisms. Carbonic anhydrase (CA, EC 4.2.1.1) is one of the enzymes,

which can be employed for CO2 sequestration technology development. CA is metallo-enzyme (mostly, a zinc-containing) and ubiquitous in nature, that is, found in animals and plants and even in the microbes. They exist in different forms, with different structures and molecular weights, and their activities vary from one to another [5]. CA catalyzes a reversible hydration of carbon dioxide: CO2 + H2O ↔ H+ + HCO3

- (1.1) including varieties of other reactions. This enzyme is involved in basic cell processes, such as photosynthesis, respiration, transport of inorganic carbon (Ci) and ions, calcification, and regulation of acid–base balance [6]. Interestingly, CA has the ability to catalyze the hydration of 600,000 molecules of carbon dioxide per molecule of CA per second comparable to a theoretical maximum rate of 1,400,000 [5, 6]. They are the fastest enzymes known e.g. it is reported that one molecule of hCA II from the human body can catalyze 1.4 ×106 molecules of CO2 per sec. CA can fix large quantities of CO2 into CaCO3 in presence of suitable cations at modest pH values in vitro.

Since CA has such unique CO2-catalyzing properties as described above, CAs has been attended as prominent biocatalysts for CO2 sequestration technology development. In this review, several classes CA are presented. This review will

Bashistha Kumar Kanth and Seung Pil Pack* Department of Biotechnology and Bioinformatics

Korea University Sejong, 339-700 Korea

E-mail: [email protected]

Carbonic Anhydrase as CO2 capturing agent: its Classes and Catalytic Mechanisms

S


57

be helpful for design of CA-based CO2 sequestration systems.

II. HISTORICAL VIEW In 1933, Carbonic anhydrase was independently discovered

by Meldrum and Roughton (Meldrum and Roughton, 1933) and Stadie and O'Brien. CA was first characterized while searching a catalytic factor necessary for fast transportation of HCO3

- from the erythrocyte to pulmonary capillary. Meldrum and Roughton purified the erythrocyte carbonic anhydrase. Keilin and Martin presented role for Zn in catalysis by finding the fact that activity was directly proportional to the Zn content; thus, carbonic anhydrase was the first Zn metalloenzyme identified. The enzyme has been found to exist in all animals after examining for its presence. Neish reported its existence in 1939 in plants but the existence was confirmed by sulfhydryl protectants when used to preserve its activity during purification. CAs from plants differ with that from animals by size and decreased sensitivity to the sulfonamide family of inhibitors.

In 1963, Veitch and Blankenship became first to report CA from a prokaryote. They found in the nasal exudates of patients suffering from respiratory infections. Several microbes like Lactobacillus, nine strains of Neisseria, and Streptococcus salivarius were examined for its origin. The first CA enzyme was reported from Neisseria sicca strain and was found to share many of the same properties as the human CA. The first sequence of CA from E. coli was reported in1992. The CA from Methanosarcina barkeri was first detected in the Archaea domain and an enzyme was found closely related to CA from M. thermophila in 1994. Roberts et al. (1997) reported CA from Thalassiosira weissflogii [7].

III. CLASSES AND MECHANISMS OF CA Carbonic anhydrases are ubiquitous enzymes, present in

prokaryotes and eukaryotes. There are five distinct CA families (α, β, γ, δ and ζ). These families have no significant similarity in amino acid sequence i.e. evolutionarily unrelated and in most cases are believed to be an example of convergent evolution [8-11]. Alpha-class carbonic anhydrase (α-CA)

Alpha-class CAs (α-CAs) are present in vertebrates, bacteria, algae and cytoplasm of green plants [6, 12-17]. Those CA enzymes found in mammals have different sub-cellular localization and tissue distribution. They are of 16 different and divided into four broad subgroups; cytosolic CAs (CA-I, CA-II, CA-III, CA-VII and CA-XIII), mitochondrial CAs (CA-VA and CA-VB), secretory CAs (CA-VI) and membrane-associated CAs (CA-IV, CA-IX, CA-XII, CA-XIV and CA-XV). There are three additional "acatalytic" CA isoforms (CA-VIII, CA-X, and CA-XI), whose functions remain unclear. CO2 + H2O ↔ HCO3 - + H+ (1.1)

CO.NH + H2O ↔ H2NCOOH (1.2)

NH.C.NH + H2O ↔ H2NCONH2 (1.3)

RCHO + H2O ↔ RCH(OH)2 (1.4)

RCOOAr + H2O ↔ RCOOH + ArOH (1.5)

RSO3Ar + H2O ↔RSO3H + ArOH (1.6)

ArF + H2O ↔ HF + ArOH (Ar: 2,4 - dinitrophenyl) (1.7)

PhCH2OCOCl + H2O ↔ PhCH2OH + CO2 + HCl (1.8)

RSO2Cl + H2O ↔ RSO3H + HCl (R: Me, Ph) (1.9)

Besides physiological reactions, the reversible hydration of CO2 to bicarbonate [reaction (1.1)], α-CAs catalyze a variety of other reactions, such as hydration of cyanate to carbamic acid, or of cyanamide to urea [reactions (1.2) and (1.3)]; the aldehyde hydration to gem-diols [reaction (1.4)]; the hydrolysis of carboxylic, or sulfonic [reactions (1.5), (1.6)], as well as other less investigated hydrolytic processes, such as those described by reactions (1.7) - (1.9).

+CO2

OH2

Zn(II)His119

His94

His96

(D) O

Zn(II)His119

His94His96

O

O-H(C)

+H2O

-HCO3-

-BH+B

Zn(II)His119

His94His96

-OH(A)

OC

O(B)

Zn(II)His119

His94His96

-OH

Fig. 1. Mechanism of Alpha-type CA The metal ion Zn(II) in all α-CAs is essential for catalysis. In

case of α-CA, human carbonic anhydrase II, hCAII (pdb: 2VVB) therefore was exemplified to explain structure-based mechanism of α-CA in detail. The metal ion, Zn(II), is situated at the bottom of a 15 Å deep active site cleft, being coordinated by three histidine residues (His94, His96 and His119) and a water molecule / hydroxide ion. The zinc-bound water is also engaged in hydrogen bond interactions with the hydroxyl moiety of Thr199, which in turn is bridged to the carboxylate moiety of Glu 106; these interactions enhance the nucleophilicity of the zinc-bound water molecule, and orient the substrate (CO2) in a favorable location for the nucleophilic attack (Fig. 1). The active form of the enzyme is the basic one, with hydroxide bound to Zn(II) (Fig. 1(A)). This strong nucleophile attacks the CO2 molecule bound in a hydrophobic pocket in its neighbourhood (substrate-binding site comprises residues Val121, Val143 and Leu198 in hCAII) (Fig.1(B)), leading to the formation of bicarbonate coordinated to Zn(II) (Fig. 1(C)). The bicarbonate ion is then displaced by a water molecule and liberated into solution, leading to the acid form of


58

the enzyme, with water coordinated to Zn(II) (Fig. 1(D)), which is catalytically inactive. To regenerate the basic form (A), a proton transfer reaction from the active site to the environment takes place, which may be assisted either by active site residues (such as His64) or by buffers present in the medium. The process may be schematically represented by reactions (1.10) and (1.11) in below:

E-Zn(II)- OH - + CO2 E-Zn(II)- HCO3

- E-Zn(II)- OH2 + HCO3 - (1.10)

E-Zn(II)-OH2 E-Zn(II)- OH - + H+ (1.11)

The reaction (1.11) i.e., the proton transfer that regenerates the zinc-hydroxide species of the enzyme is the rate limiting step [18]. In the catalytically-active isozymes, such as CA II, CA IV, CA VII and CA IX, the process is assisted by a histidine residue placed at the entrance of the active site (His64 in hCAII), as well as by a cluster of histidines, which protrudes from the rim of the active site to the surface of the enzyme, assuring thus a very efficient proton transfer process for the most efficient CA isozyme, CA-II [18].

Two main classes of CA inhibitor are known: the metal complexing anions, and the unsubstituted sulfonamides, which bind to the Zn(II) ion of the enzyme either by substituting the non-protein zinc ligand or add to the metal coordination sphere generating trigonal-bipyramidal species. Beta-class carbonic anhydrase (β -CA)

Beta-class CAs (β-CAs) are present predominantly in bacteria, algae and chloroplasts of both mono- as well as dicotyledons. β-CA have been obtained from a red alga (P. purpureum), a plant chloroplast (P. sativum), bacterium (E. coli), an Archaea (M. thermoautotrophicum), two enzymes from pathogenic bacteria (M. tuberculosis), a carboxysome (H. neapolitanus), gram negative bacteria (Haemophilus influenza) and so on.

All β-CAs share a unique α/β fold not found in any other proteins. Although β-CA can adopt a variety of oligomeric states with molecular masses ranging from 45 to 200 kDa, the fundamental structural unit appears to be a dimer or its structural equivalent. β-CAs are generally formed of 2-6 monomers of molecular weight of 25-30 kDa. Some well-known β-CAs are PPCA (pdb code: 1DDZ) available from Porphyridium purpureum, PSCA (1EKJ) from Pisum sativum, ECCA (1I6P) from Escherichia coli and MTCA (1G5C) from Methanobacterium thermoautotrophicum.

The Porphyridium purpureum CA (PPCA) is a pseudo-tetramer composed of two pseudo-dimers. Its monomer is composed of two internally repeating structures, being folded as a pair of fundamentally equivalent motifs of an α/β domain and three projecting α-helices. The motif is very distinct from the motif of either α- or γ-CAs. This homodimeric CA looks like a tetramer [19]. The Zn (II) ion is essential for catalysis in both α- or β-CA families, but its coordination is different and rather variable for the β-CAs. Depending upon organization of active site region (ligation state of active site zinc ion, and the

orientation & arrangement of nearby residues) in uncomplexed state, β-CAs have two structural classes- type I & type II (Rowlett, 2010). Thus, in the prokaryotic β-CAs, the Zn(II) ion is coordinated by two cysteinate residues, an imidazole from a His residue and a carboxylate belonging to an Asp residue (Fig. 2B(I), whereas the chloroplast CA has the Zn(II) ion coordinated by the two cysteinates, the imidazole belonging to a His residue, and a water molecule (Fig. 2B(II) (Kimber and Pai 2000; Kisker et al., 1996; Mitsuhashi et al., 2000).

His205/459Cys208/462

Zn(II)

S

S

OO Cys149/403

Asp151/405HO

H

His205/459Cys208/462

Zn(II)S

S

O

HO

Cys149/403

Asp151/405

HO

Cys208/462

O Asp151/405

O

HOZn(II)

O

His205/459

Cys149/403

H O(D)

(A)(B)

+H2O -HCO3-

-H+CO2

(C)

His205/459Cys208/462

Zn(II)S

S

O

HO

Cys149/403

Asp151/405

HOO

C

O

Fig. 2. Mechanism of Beta-type CA As arranged above, the proposed catalytic mechanism of

β-CA is explained here by exemplifying PPCA. Since there are two homologous repeats in PPCA, two Zn(II) ions are coordinated by the four amino acids. In this case these pairs are Cys149/Cys403, His205/His459, Cys208/Cys462, and Asp151/Asp405 [19]. A water molecule is also present in the neighborhood of each metal ion, but it is not directly coordinated to it, forming a hydrogen bond with oxygen belonging to the zinc ligand Asp151 / Asp405 [Fig. 2(A)]. A proton transfer reaction may occur from this water molecule to the coordinated carboxylate moiety of the aspartate residue, with generation of a hydroxide ion which may be then coordinated to Zn(II) which acquires a trigonal bipyramidal geometry [Fig. 2(B)]. Thus, the strong nucleophile which may attack CO2 bound within a hydrophobic pocket of the enzyme is formed [Fig. 2(C)], with generation of bicarbonate bound to Zn (II) [Fig. 2(D)]. In the intermediate, the aspartic acid residue originally coordinated to zinc is proposed to participate to a hydrogen bond with the coordinated bicarbonate [Fig. 2(D)]. In the last step, the coordinated bicarbonate is released in solution, together with a proton, the aspartate generated re-coordinates the Zn(II) ion, and the accompanying water molecule forms a hydrogen bond with it. The enzyme is thus ready for another cycle of catalysis.

Gamma-class carbonic anhydrase (γ-CA)

Gamma-class CAs (γ-CAs) are mainly found in Archaea and some in bacteria. The γ-CA (Cam) has been isolated from the methanogenic archaeon Methanosarcina thermophile growing


59

in hot springs [20]. The protein fold is composed of a left-handed β-helix motif interrupted by three protruding loops and followed by short and long α-helixes. The Cam monomer self-associates in a homo-trimer with the approximate molecular weight of 70 kDa [20]. The Zn(II) ion remain within the active site and is coordinated by three histidine residues, like in α-CAs, the active site of this γ-CA contains additional metal-bound water ligands, so that the overall coordination geometry is trigonal bipyramidal for the zinc-containing Cam. Two of the His residues coordinating the metal ion belong to one monomer whereas the third one is from the adjacent monomer. Thus, the three active sites are located at the interface between pairs of monomers [20]. The catalytic mechanism of γ-CAs has been proposed to be similar with α-class enzymes. It is matter of controversy whether Zn(II) is not tetra-coordinated as originally reported or penta-coordinated [20] with two water molecules bound to the metal ion. At this moment, the zinc hydroxide mechanism is accepted as being valid for γ-CAs, as it is probable that equilibrium exists between the trigonal bipyramidal and the tetrahedral species of the metal ion from the active site of the enzyme.

Ligands at the active site make contacts with the side chain of Glu62 suggesting that this side chain is probably protonated. In the uncomplexed zinc-containing Cam, the side chains of Glu62 and Glu84 appear to share a proton; additionally, Glu84 exhibits multiple conformations and Glu84 may act as a proton shuttle because histidine as an active site residue generally plays this function, usually His64. Anions and sulfonamides were shown to bind to Cam [21].

Delta-class carbonic anhydrase (δ-CA)

Delta-class CA (δ–CA) has been found in diatoms. In 1997, Morel et al. reported the purified CA (represented by TWCA1) from the marine diatom Thalassiosira weissflogii. The active site of TWCA1 is similar to that of mammalian α-CAs. The diatom carbonic anhydrase do not show significant sequence similarity with other carbonic anhydrases and represent convergent evolution at the molecular level [7, 22, 23]. The zinc has three histidine ligands and a single water molecule, being quite different from the β-CAs of higher plants in which zinc is coordinated by two cysteine thiolates, one histidine, and a water molecule. TWCA1 enzyme is upregulated by low pCO2 [13, 23] and under Zn-limited conditions, the zinc ion at the active site can be substituted by Co(II) in vivo.

Zeta-class carbonic anhydrase (ζ-CA)

Zeta-class CA (ζ-CA) is found in the same species Thalassiosira weissflogii. The marine diatom, further growing under low concentration of Zn(II) ions (like in the sea water) and in the presence of Cd2+ or low CO2 pressure, produces this zeta-class CA (represented by CDCA1) [7,13, 23]. ζ-CA occurs exclusively in bacteria, in particular a few chemolithotrophs, marine cyanobacteria containing cso-carboxysomes (So et al., 2004) and diatoms. ζ-CA naturally uses Cd(II) as its catalytic metal ion [22, 23]. CDCA1 consists of three tandem CA repeats

(R1-R3), which share 85% identity in their primary sequences [23].

Three-dimensional structural analysis suggests that CDCA1 bears some structural resemblance to β-CA, particularly near the metal ion site. Thus, the two forms may be distantly related, though amino acid sequence has diverged considerably. CDCA1 plays crucial role in CO2 fixation in marine diatom. This CA is involved in the use of cadmium as an algal nutrient in the sea water though cadmium is biologically toxic. Another structural analysis of CDCA1 shows its unique ability to exchange Cd(II) with Zn(II) keeping for both metal ions high catalytic efficiency, while the other classes CAs are severely inhibited by the Cd(II) ions [24].

IV. REMARKS Carbonic anhydrase (CA) is a zinc-containing

metallo-enzyme present in virtually every tissues, cell type, subcellular organelles, and in organisms ranging from unicellular microbes to mammals. Presently, CAs are divided to five classes (α, β, γ, δ, ζ), which have no sequence similarity each other and are supposed to be evolutionally independent. CA catalyzes a rapid inter-conversion of CO2 and water to HCO3

- and protons. The high efficiency of CA is fundamental to many biological processes like photosynthesis, respiration, pH homeostasis, ion transport, water and electrolyte balance, etc. Recently, CA has been considered as an important biocatalyst for CO2 sequestration technology because the accumulation of CO2 is the main cause for global climate change and it is critical to develop technologies that can reduce atmospheric CO2 level [25-27]. This review deals with the classes and mechanisms of several CAs as CO2 capture agents.

V. REFERENCE [1] IPCC. in: Metz, B., Davidson, O., Coninck, H. C. D., Loos, M. et al.,

(Eds). IPCC Special Report on Carbon Dioxide Capture and Storage. Cambridge University Press, 2005.

[2] Bond, G. M., Stringer, J., Brandvold, D. K., Simsek, F. A. et al., Development of integrated system for biomimetic CO2 sequestration using the enzyme carbonic anhydrase. Energy Fuels 2001, 15, 309-316.

[3] McKibbin, W. J., Wilcoxen, P. J., Climate Change Policy after Kyoto: Blueprint for A Realistic Approach. Brookings Institution Press, Washington 2002, 215–230.

[4] Gough, C., State of the art in carbon dioxide capture and storage in the UK: an experts’ review. Int. J. Greenhouse Gas Control 2008, 2, 155–168.

[5] Trachtenberg, M. C., Tu, C.K., Landers, R. A., Wilson, R. C. Carbon dioxide transport by proteic and facilitated transport membranes. Life Support Biosphy. Sci. 1999, 6, 293-302.

[6] Chegwidden, W. R., Carter, N. D., Edwards, Y. H., The Carbonic Anhydrase New Horizons. Birkhauser Verlag, 2000.

[7] Roberts S. B., Lane T., Morel F. M. M. Carbonic anhydrase in the marine diatom Thalassiosira weissflogii (bacillariophyta). J Phycol 1997, 33, 845-50.

[8] Supuran, C. T., Carbonic Anhydrases - An Overview. Curr. Pharm. Design. 2008, 14, 603-614.

[9] Supuran, C.T.; Scozzafava, A.; Conway, J. Carbonic Anhydrase - Its inhibitors and Activators, in: Supuran, C. T. (Ed): Carbonic Anhydrases: Catalytic and Inhibition Mechanisms, Distribution and Physiological Roles, CRC Press, Boca Raton (FL), USA, 2004. pp. 6-12.

[10] Supuran, C. T., Carbonic anhydrases as drug target- An overview. Curr. Top. Med. Chem. 2007, 7, 825-833.


60

http://en.wikipedia.org/wiki/Chemolithotroph

[11] Viparelli, F., Monti, S. M., De Simone, G., Innocenti, A., Inhibition of the R1 fragment of the cadmium-containing zeta-class carbonic anhydrase from the diatom Thalassiosira weissflogii with anions. Bioorg. Med. Chem. Lett. 2010, 20, 4745-4748.

[12] Hilvo, M., Tolvanen, M., Clark, A., Shen, B., Characterization of CA XV, a new GPI-anchored form of carbonic anhydrase. Biochem. J. 2005, 392, 83-92.

[13] Lane, T. W., Morel, F. M. M., A biological function for cadmium in marine diatoms. Proc. Natl. Acad. Sci. USA 2000, 97, 4627-4631.

[14] Smith, K. S., Ferry, J. G., Prokaryotic carbonic anhydrases. FEMS Microbiol. Rev. 2000, 24, 335-366.

[15] Supuran, C.T., Scozzafava, A., Casini, A., Carbonic anhydrase inhibitors. Med. Res. Rev. 2003, 23,146-189.

[16] Supuran, C.T., Scozzafava, A., Conway, J., Carbonic Anhydrase – Its Inhibitors and Activators. USA: CRC Press; 2004. 1-363.

[17] Supuran, C. T., Scozzafava, A., Carbonic anhydrase inhibitors and their therapeutic potential. Expert Opin. Ther. Pat. 2000, 10, 575-600.

[18] Briganti, F., Mangani, S., Orioli, P., Scozzafava, A., Carbonic anhydrase activators: X-ray crystallographic and spectroscopic investigations for the interaction of isozymes I and II with histamine. Biochemistry. 1997, 36, 10384-10392.

[19] Mitsuhashi, S., Mizushima, T., Yamashita, E., Yamamoto, M., X-ray structure of beta-carbonic anhydrase from the red alga, Porphyridium purpureum, reveals a novel catalytic site for CO2 hydration. J. Biol. Chem. 2000, 275, 5521-5526.

[20] Iverson, T.M., Alber, B. E., Kisker, C., Ferry, J. G., Ree,s D. C. A closer look at the active site of gamma-class carbonic anhydrases: high-resolution crystallographic studies of the carbonic anhydrase from Methanosarcina thermophila. Biochemistry 2000, 39, 9222-31.

[21] Innocenti, A., Zimmerman, S., Casini, A., Ferry, JG, Scozzafava, A, Supuran, CT. Carbonic anhydrase inhibitors. Inhibition of the zinc and cobalt gamma-class enzyme from the archaeon Methanosarcina thermophila with anions. Bioorg Med Chem Lett 2004, 14, 3327-31.

[22] Tripp, B. C., Smith, K., Ferry, J. G. Carbonic anhydrase: New insights for an ancient enzyme. J Biol Chem 2001, 276, 48615-18.

[23] Lane, T. W., Saito, M. A., George, G. N., Pickering, I. J., Prince, R. C., Morel, F. M. M. Biochemistry: A cadmium enzyme from a marine diatom. Nature 2005, 435, 42.

[24] Xu, Y., Feng, L., Jeffrey, P. D., Shi, Y., Morel F. M. Structure and metal exchange in the cadmium carbonic anhydrase of marine diatoms. Nature 2008, 452, 56-61.

[25] Ki, M. R., Kanth, B. K., Min, K. H., Lee, J-W., Pack, S. P.. Increased expression level and catalytic activity of internally-duplicated carbonic anhydrase from Dunaliella species by reconstitution of two separate domains. Process Biochem 2012; 47:1423–27.

[26] Ki, M. R., Min, K.H., Kanth, B. K., Lee, J-W., Pack, S. P.. Expression, reconstruction and characterization of codon-optimized carbonic anhydrase from Hahella chejuensis for CO2 sequestration application. Bioprocess Biosyst Eng 2013; 36, 375-381

[27] Kanth, B. K., Min, K. H., Kumari, S., Jeon, H-C., Jin, E. S., Lee, J-W., Pack, S. P.. Expression and characterization of codon-optimized carbonic anhydrase from Dunaliella species for CO2 sequestration application. Appl Biochem Biotechnol 2012, 167, 2341–56.


61

Feasibility of the C60 Fullerene Antioxidant Properties: Study with Density Functional Theory

Computer Modeling V.A. Chistyakov, Yu.O. Smirnova, I. Alperovich

Southern Federal University, Rostov-na-Donu, Russia [email protected]

Fullerene C60 compound was recently found to be a potent anti-oxidant, which may be envisioned as a result of alteration of the inner mitohondria membrane electric potential with protons transport boosted by fullerenes. Here we briefly report on the theoretical test of the very possibility of protons to pass through the surface of C60 fullerene to become confined within latter thus possibly decreasing the transmembrane electric field gradient when fullerene crosses the mitochondria membrane. Quantum-chemical calculations within Density Functional Theory are employed as a means of checking described scenario.

fullerene; C60; Sculachev ions; mitochondria membrane potential; antioxidant activity; Density Functional Theory; DFT

I. INTRODUCTION Fullerene C60 may be a powerful antioxidant demonstrating

anti-aging activity. Recently Baati et al. showed that fullerene prolonged rat's life span approximately twice [1]. Besides, rats treated with fullerene C60 demonstrated high resistance to carbon tetrachloride capable of triggering generation of huge amounts of harmful reactive oxygen species. Consequently, fullerene C60 was proposed to have high antioxidant activity in vivo. Geroprotective activity of C60 fullerene found experimentally in [1] is much higher than those of the most powerful reactive oxygen species scavengers.

Reactive oxygen species may cause oxidative damage to DNA, lipids and proteins leading to malfunctioning of cell components and eventually its death. Such mechanism is considered to be the main cause of organisms’ aging, and currently free-radical theory plays a pivotal role in modern biological concepts of aging [2].

As of today, it is still unknown how fullerene exactly acts as an antioxidant. Fullerene C60 is known to be able to inactivate hydroxyl radicals by attaching to the double bonds [3]. However, this mechanism cannot explain substantial (near two-times!) increase in the lifespan of rats. We propose there is an additional mechanism determining fullerene anti-aging activity. This mechanism is based on mild uncoupling of respiration and phosphorylation processes in a cell related to the proton gradient alteration on the mitochondrial membrane. The majority of reactive oxygen species is generated in mitochondrial respiratory chain; the latter is the source of superoxide anion that triggers a reaction chain resulting in formation of other radicals. That is why most effective antioxidants are mitochondrial-targeted. Among such

compounds are alipophilic cations (so-called Skulachev ions) with antioxidant load [4].

Because of electron transport chain activity, transmembrane potential drop is generated by creating the difference in concentrations of protons inside and outside of the inner membrane. The outer side of the internal mitochondria membrane has positive charge, while the inner side is negatively charged. This difference in electric potentials makes accumulation of Skulachev ions in the mitochondria possible. Therefore, lipophilic cations concentrate in mitochondria via electric field forces [4].

Fullerene C60 is also a lipophilic compound [5]. Theoretical simulations with molecular dynamics indicated that C60 may penetrate into membrane and accumulate there [6]. Although such simulations do not account for the charge of the mitochondrial membrane one can suppose that fullerene is indeed able to cross the inner mitochondrial membrane.

Let us assume that fullerene can absorb protons and obtain a positive charge, which allows it to cross the membrane. Next, fullerene could enclose protons within and transfer them through the inner mitochondrial membrane, which leads to the decrease in transmembrane electric potential. Korshunov et al. demonstrated [7] that even small (about 10%) decrease in transmembrane potential leads to the tremendous ten-fold decline in superoxide anion-radical generation. Therefore, such so-called mild uncouplers of oxidative phosphorylation can facilitate proton movement inside the mitochondria and thereby possess an excellent oxygen radicals-protective effect, although they are not antioxidants in terms of chemistry [5].

In order to prove outlined mechanism theoretically, we performed theoretical analysis of the fullerene C60 ability to acquire positive charge and absorb protons using Density Functional Theory (DFT). We found that the proposed mechanism indeed might take place.

II. METHOD OF CALCULATION Density Functional Theory (DFT) is a powerful and widely

used method for quantum-chemistry calculations [8]. DFT enables one to accurately compute electronic and structural properties of a variety of molecular systems. In the present work, DFT implemented in ADF 2012 program suite [9] was used to simulate the interaction between the single fullerene and surrounding proton(s) by searching for the most probable


62

Figure 1. The results of DFT geometry optimization for one (left) and six (right) protons and fullerene.

Figure 2. The distribution of charge for two, four and six protons inside the fullerene, from the left to the right, respectively.

atomic configuration of the whole system. Such configuration is found by minimizing the total energy of the system during the so-called process of the geometry optimization, which means evaluation of the system stable configuration corresponding to the minimum on the total energy hypersurface among multitude of close structures.

We applied General Gradient Approximation (GGA) for the exchange-correlation part of potential in two forms: GGA-BLYP [10] and GGA-BLYP-D3 [11]. Slater-type basis sets called double-zeta and triple-zeta with polarization function were used; frozen core approximation was employed to shorten computation time.

III. RESULTS AND DISCUSSION Initially, interaction between single proton and fullerene

was simulated. The proton was placed outside of the C60 above one of the carbon pentagons at the distance about 1Å from the pentagon plane. Calculation resulted in the proton passing through the fullerene and finally appearing inside C60 (Fig. 1, left panel). Next, more and more protons were consecutively added by allocation them above both carbon pentagons and hexagons in the initial configuration before the geometrical optimization. In all cases when there were from one to six protons present, they penetrated into the fullerene. However, the seventh proton added to the system failed to enter space inside the fullerene. One may conclude then that maximum amount of protons fullerene can accommodate is equal to six (Fig. 1, right panel).


63

To be able to penetrate into the mitochondrial membrane according to the mechanism described in the Introduction the fullerene must have positive charge distributed on its surface. Thus it is crucial to confirm the positive charge distribution over C60 surface for each tested configuration of protons. Fig. 2 illustrates charge distribution obtained with DFT within Mulliken scheme for two, four, and six protons caught inside the fullerene. One can note that when amount of protons stuck inside fullerene increases, its surface accumulates more positive charge.

As the next step, the presence of water molecules in real cell was taken into account by surrounding fullerene with water molecules. The simulation was carried out for a fullerene with single proton placed above a pentagon and 47 water molecules randomly distributed around the fullerene. We found water molecules do not influence the fullerene’s capability of absorbing protons.

To sum up, DFT simulations allowed us to propose the following mechanism. C60 fullerene molecules enter the space between inner and outer membranes of mitochondria, where there is excess of protons due to diffusion. In this compartment fullerenes are loaded with protons and acquire positive charge distributed over their surface. Such "charge-loaded" particles can be pushed through the inner membrane of the mitochondria due to the potential difference generated by the inner membrane, using electrochemical mechanism described in detail by Skulachev et al. [4, 12]. In this case, the transmembrane potential reduces, that, in turn, significantly lowers the intensity of superoxide anion radical production. The key role of mitochondria in the cellular regulation makes such “charge-loaded” fullerenes of great scientific interest not to say about prospective route for development of novel anti-aging drugs. There are number of issues to be addresses further including fullerenes’ toxicity and consideration of the more realistic theoretical models. Simulations of the fullerenes’ anti-oxidant activity are under way and to be published elsewhere.

REFERENCES [1] T. Baati, F. Bourasset, N. Gharbi, et al.,“The prolongation of the lifespan

of rats by repeated oral administration of [C60] fullerene,” Biomaterials, vol. 33, no. 12, pp. 4936-4946, 2012.

[2] A.Bratic, N.G. Larsson,“The role of mitochondria in aging,” Journal of Clinical Investigation, vol. 123, no. 3, pp. 951-957, 2013.

[3] G.V. Andrievsky, V.I. Bruskov, A.A. Tykhomyrov, S.V. Gudkov, “Peculiarities of the antioxidant and radioprotective effects of hydrated C60 fullerene nanostructures in vitro and in vivo,” Free Radicals Biology and Medicine, vol. 47, no. 6, pp. 786-793, 2009.

[4] M.V. Skulachev, Y.N. Antonenko, B.V. Chernyak, D.A. Cherepanov, V.A. Chistyakov, M.V. Egorov, et al., “Mitochondrial-targeted plastoquinone derivatives. Effect on senescence and acute age-related pathologies,” Current Drug Targets, vol. 12, no. 6, pp. 800-826, 2011.

[5] Y. Xiao, M.R. Wiesner, “Characterization of surface hydrophobicity of engineered nanoparticles,” Journal of Hazardous Materials, no. 215-216, pp. 146-151, 2012.

[6] J. Wong-Ekkabut, S. Baoukina, W. Triampo, I. Tang, D.P. Tieleman, L. Monticelli, “Computer simulation study of fullerene translocation through lipid membranes,” Nature Nanotechnology, no. 3, pp. 363-368, 2008.

[7] F.F. Severin, I.I. Severina, Y.N. Antonenko et al., “Penetrating cation/fatty acid anion pair as a mitochondria-targeted protonophore,” Procedings of NationalAcademy of Science USA, vol. 107, no. 2, pp. 663-668, 2010.

[8] F.M. Bickelhaupt, “Kohn-Sham density functional theory: predicting and understanding chemistry,” Reviews in Computer Chemistry, vol. 15, no. 1, pp. 1-86, 2000.

[9] G.T. Velde, F.M. Bickelhaupt, E.J. Baerends et al., “Chemistry with ADF,” Journal of Computer Chemistry, vol. 22, no. 9, pp. 931-967, 2001.

[10] C. Lee, W. Yang, R.C. Parr, “Development of the Colle-Salvetti correlation-energy formula into a functional of the electron density,” Physical Review B, vol. 37, no. 2, pp. 785-789, 1998.

[11] S. Grimme, J. Anthony, S. Ehrlich, H. Krieg, “A consistent and accurate ab initio parametrization of density functional dispersion correction (DFT-D) for the 94 elements H-Pu,” Journal of Chemical Physics, vol. 135, no. 15/154104, 2010.

[12] V.P. Skulachev, “How to clean the dirtiest place in the cell: cationic antioxidants as intramitochondrial ROS scavengers,” IUBMB Life, vol. 57, no. 4-5, pp. 305-310, 2005.


64

Voice Pathologies Classification Using GMM And SVM Classifiers.

AMARA Fethi Department of electronic

University of Badji Mokhtar Annaba, Algeria

[email protected]

FEZARI Mohamed Department of electronic

University of Badji Mokhtar Annaba, Algeria

[email protected]

Abstract—In this paper we investigate the proprieties of automatic speaker recognition (ASR) to develop a system for voice pathologies detection, where the model does not correspond to a speaker but it corresponds to group of patients who shares the same diagnostic. One of essential part in this topic is the database (described later), the samples voices (healthy and pathological) are chosen from a German database which contains many diseases, spasmodic dysphonia is proposed for this study. This problematic can be solved by statistical pattern recognition techniques where we have proposed the mel frequency cepstral coefficients (MFCC) to be modeled first, with gaussian mixture model (GMM) massively used in ASR then, they are modeled with support vector machine (SVM). The obtained results are compared in order to evaluate the more preferment classifier. The performance of each method is evaluated in a term of the accuracy, sensitivity, specificity. The best performance is obtained with 12 coefficientsMFCC, energy and second derivate along SVM with a polynomial kernel function, the classification rate is 90% for normal class and 93% for pathological class.This work is developed under MATLAB

Keywords-Speech pathologies detection, voice disorders, classifiction, machine learning, laryngeal diseases.

I.INTRODUCTION Assessment voice quality is an important tool for dysphonia

evaluation; it is based on perceptual analysis [1] and instrumental evaluation which comprise acoustic and aerodynamic measure [2], the first one is subjective because of the variability between listeners, although the second is objective it is invasivefor one hand , on the other hand it is has a limited reliability.

This is why the development of automatic system for classification is proposed as a complementary tool with the other mentioned technics, we distinguish three principal approaches: acoustic, parametric and non-parametric approach and statistical methods. The first approach consist to compare acoustic parameters between normal and abnormal voices such as fundamental frequency, jitter, shimmer, harmonic to noise ratio, intensity [3-6]. The evaluation of acoustic parameters depends on the fundamental frequency; the evaluation of the latter is difficult particularly in the presence of Pathology. MDVP and PRAAT are two available software to calculate these parameters [7].

The second approach is the parametric and non-parametric features extraction [8-9].

The classification of voice pathology can be seen as pattern recognition so statistical methods are an important tool to discriminate between normal and pathological voice or to know the disease from a speech signal. The statistical methods are used to mimic the brain comportment where we can recognize persons from their voice. Many researches are realized for this task, the conception of these systems has the same principal steps starting by feature selection then training and at the last testing. Support vector machine (SVM) is applied to test the effectiveness and reliability of the short term cepstral and noise parameters [10] and it is applied on discrete wavelet transform it gave a very promising results[11], GMM is used as classifier with MFCC [12], in [13] the training is supported byHidden Markov Model (HMM). The neural network is massively used for this topic in [14] the MFCCs are proposed to be the input of multi-layerperceptron (MLP).

In this paper the conception of our detector is inspired from a system of ASR [15]. 12 MFCCs, energy, dynamic parameters (first derivate and second derivate) are extracted to be the input of GMM and SVM. The main idea behind this work is to test the efficiency of the cepstral analysis to characterize pathological voices and to compare the performances of the two classifiers.

Although, the developed system is inspired from ASR system there are a principals differences betweenthe two systems which we cannot ignored, we can limited theme in two essential key point

•In ASR the model corresponds to a speaker while the model in second system corresponds to the group of patients with the same diagnostic.

•In voice pathologies detection samples used for train are different from samples used for test unlike in ASR where the two set is similar.

This paper is organized as a follow: in second section is dedicated to describe different steps to develop the systemand how we designed the two classifiers based on GMM and SVM, theexperiments are presented in section 3. The results are presented in section 4 and the last section is reserved for the conclusion and future work.


65

II.METHODOLOGY Our system will pass by the same steps to concept a system

for ASR, we will describe theme step by step, the block diagram in “fig1” show different steps adapted to our system.

Figure 1 Block diagram adapted to the detector.

A. Speech signal: In this work the creation of the data base is not our goal so

we will not discuss the speech acquisition but we will describe the database which the results are built around it.

The database presents an essential factor to develop a detector where the use of standard one helps to compare the obtained results in order to test the effectiveness and the reliability of methods [14].

In this work we have choose a German database for voice disorder developed by PUTZER [16] which contain healthy and pathological voices, where each one pronounce vowels [i, a, u] /1-2 s in wav format at different pitch (low, normal, high), it contain alsophrase and electroglottographsignal (EGG). All files are sampled at 50 KHz.

From this large database we have select patients suffer from neurological pathology (spasmodic dysphonia), this disease affects women than men that is why we have choose a female voice for training and testing step, Table.1 show the selected samples. As mentioned above the recording files contain phrase, this study is built around the phrase “good morning how are you” pronounced in Germany. The goal to use phrase in one hand is to get more data for training where GMM need an important quantity of data particularly when use a high number of mixture (Gaussian), in other hand the diversity of data enhance the accuracy of a system.

Table1. Description of dataset

Training set Test set Number Age Number Age

Normal 52 20-60 11 20-60 Pathological 29 30-82 9 30-82

Those files are down sampled to 25 KHz in order to get optimal analysis where the speech signal is considered stationary by frame of 10 to 30s so the use of a very high frequency oblige the use of a large window to get a stationary frame which minimize the size of the extracting features.

B. Pre- processing: Pre-processing of Speech Signal serves various purposes in

any speech processing application. It includes Noise Removal, Endpoint Detection, Pre-emphasis, Framing, Windowing and silence remove. In this this study we are interesting to remove silence knowing that the efficient features are included in speech portion [17].

C. Features extraction: Features extraction means finding good data allows to

categorize the healthy status of patient, features selection make a boundary between each class.

Spasmodic dysphonia is a disorder of vocal function, characterized by spasms of the muscles of the larynx that disrupt or impede the regular flow of voice this leads us to choose the MFCCs parameters in order to split the glottal source from the effect of cavities or filter in order to have a parameters with significant difference between pathological and healthy voices.

MFCC parameters are obtained calculating the Discrete Cosine Transform (DCT) over the logarithm of the energy in several frequency band given by:

= (1)

Temporal derivatives In order to use the proprieties of the dynamic behavior of

speech signal the analysis can be extended to compute the temporal derivate of the MFCC parameters, first derivate (∆) is given by:

(2)

The second derivate (∆∆) are calculated with the same equation.These parameters are calculated thanks to the toolbox voice box with melcepst function.

D. Training This study use two well-known classifier in statistical

pattern recognition, GMM and SVM, for one hand, the main idea behind the use of the SVM is that this classifier is performed with organics pathologies (10). In other hand, the comparison with the GMM is recommended where theprevious

Speech signal

Removesilence

MFCC+E (MFCC+E)+∆

(MFCC+E)+∆+∆∆

Training

Model

Decision

T E S T

Normal Spasmodic


66

work was based in GMM classifier. We describe in the follow subsection how to design the two classifiers.

GMM In pattern recognition (machine learning) the learning is

supported by the statistical classifier, Gaussian mixture model (GMMs) consist to represent the data (features) obtained at last step by a simple Gaussian curve described by:

P ( (2)

(3)

λ is the model.

Each component has the general form:

(4)

Each cluster is represented by a Gaussian as in “fig 3”

Figure 3. Scatterplot of a two-dimensional (2-D) cepstral vector and its

approximation by means of a 2-D Gaussian mixture.[10]

∑is the d-by–d covariance matrix and |∑| is its determinant itcharacterizes the dispersion of the data on the d-dimensions of the feature vector. The diagonal element σii is the variance of xi, and thenon-diagonal elements are the covariances between features. Often, the assumption is made that the features are independent. Thus, ∑ is diagonal and p(x) can actually be written as the product ofthe univariate probability densities for the elements of x. the proposed model must be optimal, one ideal way to get optimal model this is the use of Maximum likelihood estimation (MLE) given by:

= (5)

Maximizing the likelihood of observing x as being produced by the patient. Nevertheless, in the case where all the parameters are unknown, the maximum likelihood yields useless singular solutions. Thus there is a need for an alternate method. In literature the use of Expectation Maximization (EM) is the most used solution for this problem. EM is an iterative algorithm starts from initial model calculated here with a K-means algorithm for clustering.

SVM SVM is a two-class classifier that maximizes the

distancebetween nearest points of the two classes. Our task is to predict whether a test sample belongs to one of two classes. We receive training examples of the form :{xi, yi}, i = 1,…,n and xi∈Rd, yi∈{1; +1}. We call {xi} the co-variates or input vectors and {yi} the response variables or labels. We consider a very simple example where the data are in fact linearly separable we can draw a straight line such that all cases with fall on one side and have ) < 0 and cases with fall on the other and have ) > 0

Figure 4. Support vector machine with linear separation

When a data is not linearly separable a kernel function is proposed for better separation as mentioned in “fig5”

Figure5. SVM with polynomial kernel function.

E. Test step: Once models are created and that we have managed to train

the classifier, we can proceed to the classification test.

For a GMM: anew feature vector Xt is said to belong to an appropriate model if it maximizes p (Xt | λ) for every possible class. For SVM we could classify new test cases according to the ruleytest = sign(xtest).

In order to evaluate the performance of the system the results are presented by a confusion matrix represented in “Table 2” Table2. Typical aspect of a confusion matrix

System’s decision

Actual diagnosis Pathological Normal

Pathological True positive (TP) False positive (FP) Normal False negative (FN) Truenegative (TN)


67

True positive (TP) or sensitivity, is the ratio between pathological files correctly classified and the total number of pathological voices. False negative rate (FN) is the ratio between pathological files wrongly classified and the total number of pathological files. True negative rate (TN), sometimes called specificity, is the ratio between normal files correctly classified and the total number of normal files. False positive rate (FP) is the ratio between normal files wrongly classified and the total number of normal files.

The final accuracy of the system is the ratio between all the hits obtained by the system and the total number of files.

III.EXPERIMENTALPROTOCOLS:

As mentioned above the sample voice (normal and spasmodic) is divided in two set one for the training and one for test.Some experiments are realizedin order to evaluate the effect of different factors in our system; two groups of experiments are compared. One is based on the GMM classifier, whereas the other is using SVM classifier.

GMM classifier

- Use 12 MFCCs, energy, their derivate (∆)and(∆∆).

- Use of different number of Gaussian (power of 2).

SVM classifier

- Use 12 MFCCs, energy, their derivate (∆)and(∆∆).

- Use different kernel function.

In this study, the K-mean algorithm for clustering is used before training SVM so we will not separate features but we will separate their centers or cluster in order to assure convergence of SVM training and to reduce the cost of computation.

IV.RESULT AND DISCUSSION In our experiment we need to know the optimal model

which give best classification rate, this is obtained by a model with proprieties: 64 centers (Gaussian) for GMM and with a SVM with a polynomial kernel function with 39 MFCCs. The results are represented in confusion matrix in table 3. Table3 Confusion matrix.

System’s decision

Actual diagnosis GMM SVM

Pathological Normal Pathological Normal Pathological 79.92% 18.10 % 93 % 10 % Normal 20.08% 81.90% 7 % 90%

The results of the classification given in a frame that means the rate of classification represent the number of known frame by the total of the frame.

If we test each file (normal and pathological) separately, we get an accuracy of 100% for the two classes, by setting up a threshold to the number of classified frames. If more than 70% of the frames of a file are assigned to a certain class, then the whole file is assumed to belong to that class.

• Discussion: In this subsection, we discuss some experimental results

obtained from the proposed analysis methods.

GMM

-The classification rate depend to the number of Gaussian and the number of parameters MFCCs as mentioned in “fig6”

Figure 6. Classification rate for different mixtures and parameters for normal

and abnormal class.

-From the two curve we note that when we increase the number of Gaussian with the increase of the MFCCs coefficients the classification rateimproves

-Modeling by GMM requires a large number of data for the training, particularly when we use a high number of Gaussian to create a model, this prevents us to use more than 64 Gaussians particularly with the abnormal class which contains a small number of file.

SVM

As with GMM, classification rate improve with the increase of MFCC coefficient the results are represented in figure 7


68

Figure7 Classification with SVM

-The precedent figure shows that SVM is more preferment than GMM.

V.CONCLUSION This work is focused on pathological voices detection

(spasmodic dysphonia) and it is built around a system for automatic speaker recognition based onGMM and SVM as classifier.

A good classification rate needs efficient features to characterize each class, in this work, on one handthe accuracy of system increase with the of the number of parameters (best accuracy with 39 coefficients) that means that the difference between normal and abnormal become noticeable with second derivate (∆∆) of MFCC and energy more than the others, on the other handthe effect of the number of Gaussian which makes up the model is important where a sufficient number of mixtures allows to represent data (features) optimally.We can deduce also that the quantity of data used for training a system is very important.Both GMM and SVM the best accuracy is obtained with (∆∆) dynamics parameters while SVM is more preferment than GMM where the accuracy for an abnormal class is 93% and 87% for the normal class.

The very promising result motivates us to improve this work;the future work will be concerned on the use ofanother database to assess the independence of the method used for the database. We will also validate this work with other pathologies for example organic pathologies.We will interest to the hybrid approach.

REFERENCES [1] Ghio A. Dufour S. Rouaze M. Bokanowski V. Pouchoulin G. Révis J.

Giovanni A. ‘’ Mise au point et évaluation d’un protocole d’apprentissage de jugement perceptif de la sévérité de dysphonies sur de la parole naturelle’’. REV LARYNGOL OTOL RHINOL.2011;132,1:1-9.

[2] Antoine Giovanni1, Pirng Yu2, Joana Révis1, Marie-Dominique Guarella1, Bernard Teston3, Maurice Ouaknine1 ’’Analyse objective des dysphonies avec l’appareillage EVA’’. Fr ORL - 2006 ; 90 : 183

[3] Darcio G. Silva, Luıs C. Oliveira and Mario Andrea ‘’Jitter Estimation Algorithms for Detection of Pathological Voices’’ Hindawi Publishing Corporation, EURASIP Journal on Advances in Signal Processing Volume 2009, Article ID 567875, 9 pages.

[4] Miltiadis Vasilakis, Yannis Stylianou ‘’Voice Pathology Detection Basedeon Short-Term Jitter Estimations in Running Speech’’ Folia Phoniatr Logop 2009;61:153–170.

[5] Sonu, R. K. Sharma ‘’ Disease Detection Using Analysis of Voice Parameters’’ International Journal of Computing Science and Communication Technologies, VOL.4 NO. 2, January 2012.

[6] Jacques Koremana, Manfred Pützer, Manfred Just ‘’Correlates of Varying Vocal Fold Adduction Deficiencies in Perception and Production: Methodological and Practical Considerations ‘’ Folia Phoniatr Logop 2004;56:305–320

[7] Miltiadis Vasilakis, Yannis Stylianou ‘’Voice Pathology Detection Based eon Short-Term Jitter Estimations in Running Speech’’ Folia Phoniatr Logop 2009;61:153–170.

[8] Raissa Tavares , Nathália Monteiro , Suzete Correia , Silvana C. Costa , Benedito G. Aguiar Neto (2) and Joseana Macêdo Fechine ‘’Optimizing laryngeal pathology detection by using combined cepstral features’’ Proceedings of 20th International Congress on Acoustics, ICA 2010 23-27 August 2010, Sydney, Australia ICA 2010.

[9] Julián D. Arias-Londoño, Juan I. Godino-Llorente, Germán Castellanos-Domínguez, Nicolás Sáenz-Lechón, Víctor Osma-Ruiz ‘’Complexity Analysis of Pathological Voices by means Markov Entropy measurements’’31st Annual International Conference of the IEEE EMBS Minneapolis, Minnesota, USA, September 2-6, 2009

[10] Juan Ignacio Godino-Llorente, Pedro Gómez-Vilda,Nicolás Sáenz-Lechón1, Manuel Blanco-Velasco, Fernando Cruz-Roldán, and Miguel Angel Ferrer-Ballester” Support Vector Machines Applied to the Detection of Voice Disorders” Springer-Verlag Berlin Heidelberg pp. 219 – 230, 2005.

[11] Nafise ErfanianSaeedi, FarshadAlmasganj, FarhadTorabinejad ‘’Support vector wavelet adaptation for pathological voice assessment’’ Computers inBiologyandMedicine41(2011)822–828

[12] Juan Ignacio Godino-Llorente*, Member, IEEE, Pedro Gómez-Vilda, Member, IEEE, andManuel Blanco-Velasco, Member, IEEE “Dimensionality Reduction of a Pathological Voice Quality Assessment System Based on Gaussian Mixture Models and Short-Term Cepstral Parameters” IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 53, NO. 10, OCTOBER 2006

[13] Alireza A. Dibazar, Theodore W. Berger, and Shrikanth S. Narayanan” Pathological Voice Assessment”IEEE EMBS 2006 NEW YORK.

[14] Nicolas Saenz-Lechon, Juan I. Godino-Llorente, Vıctor Osma-Ruiz, Pedro Gomez-Vilda ‘’Methodological issues in the development of automatic systems for voice pathology detection’’ Biomedical Signal Processing and Control 1 (2006) 120–128.

[15] G. Pouchoulin, C. Fredouille1, J.-F. Bonastre, A. Ghio, M. Azzarello, A. Giovanni ‘’Modélisation Statistique et Informations Pertinentes pour la Caractérisation des Voix Dysphonies’’ Actes des XXVIes journ´ees d’´etudes sur la parole Dinard, juin 2006.

[16] Manfred Putzer & Jacques Koreman ‘’A german databse for a pattern for vacal fold vibration ‘’ Phonus 3, Institute of Phonetics, University of the Saarland, 1997, 143-153.

[17] Ayaz Keerio, Bhargav Kumar Mitra, Philip Birch, Rupert Young, and Chris Chatwin“On Preprocessing of Speech Signals”On Preprocessing of Speech Signals” World Academy of Science, Engineering and Technology 47 2008


69

The use of starch matrix-banana fiber composites for biodegradable maxillofacial bone plates

Lamis R. Darwish Department of Biomedical Enginneering,

Faculty of Engineering. Helwan University Cairo, Egypt.

Mohamed Tarek El-Wakad Department of Biomedical Enginneering,

Faculty of Engineering. Helwan University Cairo, Egypt.

Mahmoud Farag Department of Mechanical Engineering

The American University in Cairo Cairo, Egypt

Mohamed Emara Department of manufacturing Engineering,

CIC- Canadian International Cairo, Egypt

Abstract— Starch based green composites have been studied

as potential materials to be used in several biomedical applications. This paper explores utilizing starch based composites reinforced with pseudostem banana fibers in fabricating biodegradable maxillofacial bone plates. Corn starch plasticized by 30 wt.% glycerin and 20 wt.% distilled water was used as a matrix. The produced thermoplastic starch (TPS) matrix is reinforced with pseudostem banana fibers at different weight fractions using hot pressing at 5 MPa and 160ºC for 30 minutes. Our experimental results showed that increasing the banana fibers weight fraction progressively improved the mechanical properties reaching a maximum at 50 wt.% fibers. The improvement in the mechanical properties of starch/banana fibers composite was attributable to the strong interaction between fibers and the starch matrix, as evidenced by a series of scanning electron micrographs of the fracture surface. Furthermore, experiments investigating thermal properties and water uptake also showed that the best results are achieved at the 50 wt.% banana fibers. The experimental results show that the starch matrix-banana fiber composites satisfy the maxillofacial bone fixation requirements.

Keywords- Biodegradable composites; Thermoplastic cornstarch; Banana fibers; maxillofacial bone plates.

Metallic implants made of titanium alloys are widely used for internal fixation of bone fractures of the human skeleton. These implants are regarded as the gold standard for the majority of the fracture fixation treatments due to their outstanding mechanical properties [1-3]. However, metallic implants have several significant drawbacks such as stress shielding effect which leads to bone resorption, implant loosening and, consequently, the need for a second operation to remove the implants especially in pediatrics. This second operation has several risks such as infection, removal problems of jammed implants, implants migration and associated extra health care costs [2-4]. Consequently, in order to eliminate stress shielding it has been investigated for many years to use a material that would degrade and gradually lose its strength at the same rate of bone healing.

Starch based green biomaterials have recently been proposed as having a great potential for several applications in the biomedical field as they are totally biodegradable, biocompatible, and inexpensive [5-7]. These materials were

shown to have a great potential to be used in the bone related therapy applications [8-10], ranging from bone tissue engineering scaffolds [11], to bone cements [12].

Starch is the major polysaccharide constituent of photosynthetic tissues and of many storage organs in plants [13]. Worldwide corn represents the major commercial source of starch. It is a semicrystalline polymer composed of a mixture of amylose, a linear polysaccharide and amylopectin, a highly branched polysaccharide [4]. However, before being thermally processable as for thermoplastic polymers, starch must be converted to thermoplastic starch (TPS) by the addition of specific plasticizers combined with the application of heat and shear forces [14]. A plasticizer is required for preparing TPS because the plasticizer can reduce brittleness and avoid the cracking of starch based materials during handling and storage [15]. Various plasticizers have been used for TPS such as formamide, ethanolamine, sorbitol, sugars, and glycerol [16].

However, TPS has disadvantages when compared to synthetic biodegradable polymers, such as it has a strong ability to absorb water, and has poor mechanical and thermal properties [15]. The reinforcement of the TPS matrix with natural lignocellulosic fibers seems to be the logical alternative in order to increase their mechanical performance and to preserve the green character of the final material [17].

Consequently, this work proposes (invistigates) using glycerol plasticized starch matrix reinforced with pesudostem banana fibers (BFs) at different weight fractions in the fabrication of maxillofacial bone plates. Pseudostem banana fiber is a hard fiber obtained from the leaf sheaths. It is considered to be the strongest commercially available hemicelluloses fiber due to its high cellulose content [18-20]. Extensive experiments have been conducted to investigate the resulting plates.

I. MATERIALS AND METHODS

A. Materials Regular native cornstarch (28% amylose) with 11%

moisture and reagent grade glycerin (99.7% purity) were used as received. Banana pseudostem was received from a local


70

supplier. The pseudostem was chopped manually into 30 mm length pieces. The raw banana fiber was obtained by mechanical separation of the fiber from these chopped pieces. The extracted banana fiber was treated with 0.5% NaOH solution at 90 °C for 30 min, and then washed with running tap water until the PH value reached 7. The alkali treated fiber was then oven dried.

B. Sample preparation Corn starch was plastisized by 30 wt.% glycerin and 20

wt.% distilled water. More details about the plasticization technique can be found elsewhere [21]. The resulting thermoplastic starch matrix was emulsified with distilled water then reinforced with alkali treated banana fibers at 40%, 50% and 60% weight fractions using hot pressing at 5 MPa and 160 ºC for 30 minutes.

C. Morphological characterization Morphological characteristics of the tensile fracture

surface of TPS/banana fiber composites were examined using ZISS Scanning Electron Microscope (SEM). Images were captured in a liquid nitrogen cooled atmosphere.

D. Mechanical tests Composite samples (specimen) were cut from the

compression molded sheets considering the dimensions ratio mentioned in ASTM D5083-10. Tensile and three point flexural tests were performed using a universal testing machine (Instron 3382) with a cross head speed of 5 mm/min. Tests were performed at room temperature. Ten specimens were tested for each fiber weight fraction.

E. Water absorption measurements Composite samples were oven dried at 100 °C for 4 hours.

These samples were weighted immediately after their removal from the oven. Then they were stored for 7 days in desiccators set up at 100% RH using distilled water. The increase in weight was measured on daily basis. The water absorption can be calculated using the following equation:

Water absorption (%)=[(wf-wi)/wi]*100 (1)

where wf is the final sample weight and wi is the initial sample weight. All measurements were performed in duplicate.

F. Thermal properties Thermogravimetric analysis (TGA) was performed using a

TA Q500 TGA analyzer (TA Instruments, New Castle, DE, USA). The samples were heated from room temperature to 700 °C at a heating rate of 20 °C/min using synthetic air flow of 60 mL/min. The derivative of TGA curves (DTG) was obtained using TA analysis software.

II. RESULTS AND DISSCUSION

A. Morphological characteristics Fig. 1 shows the scanning electron micrographs of the

tensile fracture surfaces of TPS/BF composites at 40%, 50%, 60% BF wt.%. The good interfacial adhesion between TPS matrix and treated BF can be clearly seen in fig. 1 (a-d). This good adhesion was evident by the fracture of the BF at the surface, the decreased number of fibers pull outs and holes, and the absence of gap between the BF and the TPS matrix. This good adhesion is mainly due to the similarity in polarity between the BF and the TPS.

On the contrary, increasing the BF wt.% up to 60% caused deterioration in the wettability of the BFs with emulsified TPS matrix. This deterioration is due to the agglomeration of the BFs. This was evident by the increased number of pulled out fibers and holes (fig. 1 (e)), also, the surface cleanliness of the pulled out fibers (fig. 1 (f)).

B. Mechanical properties The ultimate tensile strength (UTS) and young’s modulus

(YM) for the TPS/BF composites with different BF content are shown in fig. 2. Both of the UTS and YM increase linearly with increasing the BF content up to 50%. The UTS and YM were 30 MPa and 4.6 GPa, respectively, in the samples with 50% BF. The only study in the literature used pseudostem banana fibers as reinforcement for glycerol plasticized cornstarch was by Guimarães et al. [22]. The maximum banana fibers wt.% reported in this study was 35% and it gave 3.56 MPa and 74.35 MPa for UTS and YM, respectively.

Besides the flexural strength and modulus for the same fiber content patterns followed the same behavior of UTS and YM. The flexural strength and modulus are shown in fig. 3. The maximum flexural strength and modulus were 80 MPa and 7.5 GPa, respectively, in the samples with 50% BF.

Further addition of BF causes deterioration in the mechanical properties of the composite due to the agglomeration of the BF. This agglomeration decreases the wettability of the fiber with the emulsified TPS matrix, thus decreases the attachment between the fibers and the matrix. These results are in agreement with the SEM fracture surface investigation.

C. Water absorption Water absorption is an important factor for the application

of the TPS composites in the biomedical field. The results of water absorption for the TPS/BF composites with different BF wt.% are shown in fig. 4 . It can be clearly noticed from the figures that increasing the BF wt.% decreases the water absorption of the TPS/BF composites. These results are ascribed to the fact that starch is more hydrophilic than cellulose, thus the presence of fibers reduce the hydrophilicity of the starch based composites. The enhancement in the water absorption property of TPS/BF composites and its cause conforms to the results documented in previous studies [22-25].


71

Fig.1. SEM micrographs of the tensile fracture surfaces of the (a,b) 40% BF, (c,d) 50% BF, and (e,f) 60% BF composite

a b

c d

e f


72

Fig. 2. Tensile properties of the TPS/BFs composites at different BFs wt.% (a) tensile strength (MPa) and (b) Young’s modulus (MPa)

Fig. 3. Flexural properties of the TPS/BFs composites at different BFs wt.% (a) flexural strength (MPa) and (b) flexural modulus (MPa)

D. Thermogravietric analysis Fig. 5 shows the TGA curves and their corresponding

derivative (DTG) curves of the 40%, 50% 60% BF composites. The first weight loss that occurred just above the room temperature till 100 °C is related to the evaporation of water. It appeared as a small peak around 60 °C and 100 °C in the DTG curves of 40% and 50% BFs composite. This peak disappeared in the higher BFs wt.% (60%), which indicates that less water was absorbed by the composite. Wattanakornsiri et al [26] reported the same peak disappearance in the high fibers content. The weight loss in

the temperature range of 150 °C to 200 °C is related to the evaporation of glycerin. This glycerin evaporation appeared in the DTG curves of the TPS and the TPS/BF composites as a small peak right before 200 °C. These peaks were also observed by other authors [24-27]. Peaks group (I) observed in the DTG curves of the TPS/BF composites at the temperature range of 325 °C to 335 °C are attributed to the thermal degradation of starch. The weight loss in the temperature range of 365 °C to 377 °C, which appeared in the DTG curves as peaks group (II) are related to the degradation of the treated BFs. Starch and BFs decomposition peaks were observed by other authors [24].

a b

a b


73

It can be noticed from the TGA curves and table I that the onset degradation temperature of starch increased with increasing the BFs wt.%. Prachayawarakorn et al. [28-29] reported the same observation and assigned it to the possible formation of hydrogen bond linkage between TPS and cellulosic fibers.

Furthermore, The weight loss of the 40% ,50% and 60% BFs composites at the peaks group (I) temperatures, which are related to the decomposition of starch, are 41%, 38.7% and 38.3%, respectively. It can be noticed that the weight loss decreased by increasing the BFs content.

Thus, The decrease in the weight loss as well as the increase in the onset degradation temperature of starch indicate that the thermal stability of the TPS/BFs composites improves with increasing the BFs wt.%. Similar observations were also reported by other research groups [24-27]. They attributed this improvement in the thermal stability of TPS by increasing the fibers content to the higher thermal stability of cellulosic fibers compared to starch and the good compatibility of both polysaccharides.

Fig. 4. Effect of BFs wt.% on (a) water absorption of the composites over a period of 7 days and (b) water absorption saturation level of the composites after 7 days

Fig. 5. (a) TGA and (b) DTG of TPS/BFs composites with different BFs wt.%

a b

a b


74

III. CONCLUSION

This paper addressed the developing of green cornstarch composites for maxillofacial bone plates’ fabrication. Glycerol plasticized starch was reinforced with pseudostem banana fibers at different weight fractions. Extensive experiments had been conducted in order to investigate the morphological, thermal and mechanical properties for the proposed composite. The optimum mechanical properties were obtained at 50 wt.% BFs. Furthermore, Incorporating BFs into the TPS matrix improved the thermal properties of the composite. Thus the mechanical and thermal properties of this composite nominate it to be used in the fabrication of maxillofacial bone plates. However the water resistance property was the major drawback of the fabricated composite. Future work may explore blending biodegradable synthetic polymer with the TPS matrix in order to overcome the water resistance drawback.

TABLE I. Onset and peak temperatures of TPS/BFs composites

BF wt.% Tonset (°C) Temperature region >350 °C <350 °C

Peak (°C) Weight loss (%) Peak (°C)

40% BF 281.25 325 41 365.9

50% BF 306.25 333.8 38.75 377

60% BF 309.375 334.9 38.33 377.5

REFERENCES [1] Leenslag, J. W., Pennings, A. J., Bos, R. R., Rozema, F. R., &

Boering, G. “Resorbable materials of poly (L-lactide). VI. Plates and screws for internal fracture fixation”. Biomaterials, vol. 8(1), 70-73.J. Clerk Maxwell, A Treatise on Electricity and Magnetism, 3rd ed., vol. 2. Oxford: Clarendon, 1892, pp. 68-73, 1987.

[2] Hu, Q., Li, B., Wang, M., & Shen, J., “Preparation and characterization of biodegradable chitosan/hydroxyapatite nanocomposite rods via in situ hybridization: a potential material as internal fixation of bone fracture”. Biomaterials, vol. 25(5), pp. 779-785, 2004.

[3] Eglin, D., & Alini, M., “Degradable polymeric materials for osteosynthesis: tutorial”. Eur Cell Mater, vol. 16, pp. 80-91, 2008.

[4] Zargar Kharazi, A., Fathi, M. H., Bahmani, F., & Fanian, H. (2011). Partially resorbable composite bone plate with controlled degradation rate, desired mechanical properties and bioactivity. Polymer Degradation and Stability,96(12), 2055-2063.

[5] Alberta Araújo, M., Cunha, A. M., & Mota, M. (2004). Enzymatic degradation of starch-based thermoplastic compounds used in protheses: identification of the degradation products in solution. Biomaterials, 25(13), 2687-2693.

[6] Marques, A. P., Reis, R. L., & Hunt, J. A. (2002). The biocompatibility of novel starch-based polymers and composites: in vitro studies. Biomaterials, 23(6), 1471-1478.

[7] Lu, D. R., Xiao, C. M., & Xu, S. J. (2009). Starch-based completely biodegradable polymer materials. Express Polym Lett, 3(6), 366-75.

[8] Azevedo, H. S., Gama, F. M., & Reis, R. L., 2003, “In vitro assessment of the enzymatic degradation of several starch based biomaterials”, Biomacromolecules, 4(6), pp. 1703-1712.

[9] Duarte, A. R. C., Mano, J. F., & Reis, R. L., 2010, “Enzymatic degradation of 3D scaffolds of starchpoly-( ɛ-caprolactone) prepared by supercritical fluid technology”, Polymer Degradation and Stability, 95(10), pp. 2110-2117.

[10] Mano, J. F., Koniarova, D., & Reis, R. L., 2003, “Thermal properties of thermoplastic starch/synthetic polymer blends with potential biomedical applicability”, Journal of Materials Science: Materials in Medicine, 14(2), pp. 127-135.

[11] Gonçalves, A., Costa, P., Rodrigues, M. T., Dias, I. R., Reis, R. L., & Gomes, M. E., 2011, “Effect of flow perfusion conditions in the chondrogenic differentiation of bone marrow stromal cells cultured onto starch based biodegradable scaffolds”, Acta Biomaterialia, 7(4), pp. 1644-1652.

[12] Boesel, L. F., & Reis, R. L., 2008, “A review on the polymer properties of Hydrophilic, partially Degradable and Bioactive acrylic Cements (HDBC)”, Progress in Polymer Science, 33(2), pp. 180-190.

[13] Alves, N. M., Saiz-Arroyo, C., Rodriguez-Perez, M. A., Reis, R. L., & Mano, J. F. (2007). Microhardness of starch based biomaterials in simulated physiological conditions. Acta Biomaterialia, 3(1), 69-76.

[14] Wattanakornsiri, A., Pachana, K., Kaewpirom, S., Sawangwong, P., & Migliaresi, C. (2011). Green composites of thermoplastic corn starch and recycled paper cellulose fibers. Sonklanakarin Journal of Science and Technology, 33(4), 416.

[15] Kaewtatip, K., & Thongmee, J. (2012). Studies on the structure and properties of thermoplastic starch/luffa fiber composites. Materials & Design, 40, 314-318.

[16] Hulleman, S. H., Janssen, F. H., & Feil, H. (1998). The role of water during plasticization of native starches. Polymer, 39(10), 2043-2048.

[17] Girones, J., Lopez, J. P., Mutje, P., Carvalho, A. J. F., Curvelo, A. A. S., & Vilaseca, F. (2012). Natural fiber-reinforced thermoplastic starch composites obtained by melt processing. Composites Science and Technology, 72(7), 858-863.

[18] Benítez, A. N., Monzón, M. D., Angulo, I., Ortega, Z., Hernández, P. M., & Marrero, M. D., 2012, “Treatment of Banana Fiber for Use in the Reinforcement of Polymeric Matrices”, Measurement, 46(3), pp. 1065-1073.

[19] Chaudhary, A. K., Gope, P. C., & Singh, V. K., 2011, “Studies on Fracture Performance of Bio-fiber-Silica-glass Fiber Reinforced Epoxy Hybrid Composites” In Experimental and Applied Mechanics, T. Proulx ed., Springer, New York, Vol. 6, pp. 363-368.

[20] Mohanty, Amar K., Manjusri Misra, and Lawrence Thaddeus Drza, 2005, Natural fibers, biopolymers, and biocomposites, CRC Press, Boca Raton, Florida, pp. 90-110, Chap. 2.

[21] Elsayed, H., Farag, M., Megahed, H., Mehanny, S., 2012, “Influence of flax fibers on properties of starch based composites”, IMECE2012-89628, Proc. of ASME 2012 international conference of mechanical engineering, Huston, Texas, USA.

[22] Guimarães, J. L., Wypych, F., Saul, C. K., Ramos, L. P., & Satyanarayana, K. G. (2010). Studies of the processing and characterization of corn starch and its composites with banana and sugarcane fibers from Brazil. Carbohydrate Polymers, 80(1), 130-138.

[23] Curvelo, A. A. S., De Carvalho, A. J. F., & Agnelli, J. A. M. (2001). Thermoplastic starch–cellulosic fibers composites: preliminary results.Carbohydrate Polymers, 45(2), 183-188.

[24] Prachayawarakorn, J., Chaiwatyothin, S., Mueangta, S., & Hanchana, A. (2012). Effect of jute and kapok fibers on properties of thermoplastic cassava starch composites. Materials & Design.

[25] Prachayawarakorn, J., Sangnitidej, P., & Boonpasith, P. (2010). Properties of thermoplastic rice starch composites reinforced by cotton fiber or low-density polyethylene. Carbohydrate Polymers, 81(2), 425-433.

[26] Wattanakornsiri, A., Pachana, K., Kaewpirom, S., Sawangwong, P., & Migliaresi, C. (2011). Green composites of thermoplastic corn starch and recycled paper cellulose fibers. Sonklanakarin Journal of Science and Technology, 33(4), 416.

[27] Kaewtatip, K., & Thongmee, J. (2012). Studies on the structure and properties of thermoplastic starch/luffa fiber composites. Materials & Design, 40, 314-318.

[28] Prachayawarakorn, J., Limsiriwong, N., Kongjindamunee, R., & Surakit, S. (2012). Effect of agar and cotton fiber on properties of thermoplastic waxy rice starch composites. Journal of Polymers and the Environment, 20(1), 88-95.


75

[29] [Prachayawarakorn, J., Ruttanabus, P., & Boonsom, P. (2011). Effect of cotton fiber contents and lengths on properties of thermoplastic starch composites prepared from rice and waxy rice starches. Journal of Polymers and the Environment, 19(1), 274-282.


76

Authors Index

Alperovich, I. 62

Kim, J.-J. 52

Amara, F. 65

Konvickova, S. 44

Arathi, T. 19

Kwon, T.-K. 52

Ayaz, Y. 47

Omer, O. A. 39

Chistyakov, V. A. 62

Pack, S. P. 57

Darwish, L. R. 70

Parameswaran, L. 19

El-Wakad, M. T. 70

Park, I.-S. 52

Emara, M. 70 Farag, M. 70

Ruzicka, P. 44

Fezari, M. 65 Gilani, S. O. 47

Setayeshi, S. 30

Hosseinkhani, E. 30

Smirnova, Y. O. 62

Teshnehlab, M. 30

Jamil, M. 47

Waris, M. A. 47

Jung, J.-Y. 52

Won, Y. 52

Kanth, B. K. 57

Zach, L. 44


77

Documents

RECENT ADVANCES in BIOLOGY, MEDICAL PHYSICS, MEDICAL ... · Prof. Wasfy B Mikhael (IEEE Fellow, University of Central Florida Orlando,USA) Prof. Sunil Das (IEEE Fellow, University