Upload
trinhtram
View
214
Download
0
Embed Size (px)
Citation preview
Automatic volume estimation of the thyroid gland using 2D Ultrasound Imaging
presented by
SUBBARAO NIKHIL NARAYAN
School of EEE
Supervisor: Assoc. Prof. Pina Marziliano, School of EEE
Co-Supervisor: Prof. Nadia Magnenat Thalmann, Institute for Media Innovation
Clinical Collaborator: Dr. Christopher Hobbs, Dept. of Otorhinolaryngology, Tan Tock Seng Hospital,
Singapore
11-04-2013
Outline
• Fundamentals of Thyroid Gland
•Volume estimation
•Proposed method on 2D US imaging
• Experimental results
• Future work
Fundamentals Anatomy Neck Lumps
*Images taken from the internet
- Hormone Regulation
- Iodine Metabolism - Controls energy
usage in body
Functions Pathophysiology
- Hypothyroidism - Hyperthyroidism - Benign Thyroid Disease - Thyroid Cancer
Confirming Diagnosis
- Fine Needle Aspiration Biopsy
Thyroid Imaging
Normal Thyroid Thyroid with disorders
Thyroid Imaging
*Images taken from : www.thyroidimaging.com
Ultrasound
Radiography
Scintiscanning
CT Scan
MRI
Problem Statement
Accurate volume estimation of the thyroid gland using 2D Ultrasound Image / Image Sequences
Objectives:
1. Segmentation of thyroid gland and nodules for volume estimation
2. To develop a mathematical model for accurate estimation of thyroid gland volume.
Importance of Thyroid Volume Estimates
• Required for differential diagnosis
• Normal gland volume 5-20ml depending age, gender etc..
• Required for therapy using radioiodine 131 (131I) [1]
• Dosage calculation for treatment [1]
𝐷 = 𝑉 ∗100
𝑈 ∗ 𝐶
Where, D = 131I dosage (MBq), V = Thyroid volume (ml),
U = 24hr thyroidal 131I uptake(%) and C is a constant
[1]. van Isselt, J.W., et al., Comparison of methods for thyroid volume estimation in patients with Graves' disease. European Journal of Nuclear Medicine and Molecular Imaging, 2003. 30(4): p. 525-531.
Methods to Estimate Thyroid Volume
• Modality based methods
• Ellipsoid Method
• Planimetry Method
• Cavaliari Method
• Anthropometric Methods
• Computer Aided Methods
Ellipsoid Method
• Thyroid modeled as an ellipse
• 𝑉 = 𝜋
6 ∗ ℎ𝑒𝑖𝑔ℎ𝑡 ∗ 𝑤𝑖𝑑𝑡ℎ ∗ 𝑙𝑒𝑛𝑔𝑡ℎ [2-4]
• 𝑉 = 1.24101 ∗ 𝜋
6 ∗ ℎ𝑒𝑖𝑔ℎ𝑡 ∗ 𝑤𝑖𝑑𝑡ℎ ∗ 𝑙𝑒𝑛𝑔𝑡ℎ + 3.6627[5]
[2]. Kollorz, E.N.K., et al., Quantification of thyroid volume using 3-D ultrasound imaging. Ieee Transactions on Medical Imaging, 2008. 27(4): p. 457-466. [3]. Nygaard, B., et al., Thyroid volume measured by ultrasonography and CT. Acta Radiologica, 2002. 43(3): p. 269-274. [4]. van Isselt, J.W., et al., Comparison of methods for thyroid volume estimation in patients with Graves' disease. European Journal of Nuclear Medicine and Molecular Imaging, 2003. 30(4): p. 525-531. [5]. Ruggieri, M., et al., The estimation of the thyroid volume before surgery - an important prerequisite for minimally invasive thyroidectomy. Langenbecks Archives of Surgery, 2008. 393(5): p. 721-724.
Planimetry Method
• 𝑉𝑝 =𝜋
6× 𝐷𝑙 × 4
𝐴𝑙
𝜋×𝐷𝑙× 4
𝐴𝑡
𝜋×𝐷𝑡 [6]
where, Dl and Dt are the maximum depth in the longitudinal and transverse sections of the scan, Al and At are crosssectional areas in the longitudinal and
• Used frequently with 3D imaging modalities like CT/MRI/3D Ultrasound
[6]. V. Brauer, P. Eder, K. Miehle, T. Wiesner, H. Hasenclever, and R. Paschke, Interobserver variation for ultrasound determination of thyroid nodule volumes, Thyroid, vol. 15, no. 10, pp. 1169{1175, 2005.
At
Dt
Cavaliari Method
• Also known as point counting method [13]:
• 𝑉 𝑃𝐶 = 𝑇 ×𝑆𝑈×𝑑
𝑆𝐿
2× ∑𝑃
Where 𝑆𝑈 and 𝑆𝐿 are scaling parameters, 𝑃 is the number of points hitting the test grid and 𝑇 is the slice thickness
[13]. U. E. Vurdem, N. Acer, T. Ertekin, A. Savranlar, O. Topuz, and M. Keceli, Comparison of three volumetric techniques for estimating thyroid gland volume, Turkish Journal of Medical Sciences, vol. 42, no. 1, pp. 1299-1306, Dec 2012.
Anthropometric methods
• Metrics used [14]:
• 𝑎𝑔𝑒
• Body Surface Area (𝐵𝑆𝐴)
• log (𝑉) = 0.6089 + 0.832 × 𝐵𝑆𝐴 + 0.0403 × 𝑎𝑔𝑒
• log 𝑉 = 1.5988 + 0.7965 × 𝐵𝑆𝐴 − 0.1755 × 𝑔𝑒𝑛𝑑𝑒𝑟
where 𝑔𝑒𝑛𝑑𝑒𝑟 = 0 for male and 1 for female
• All constants obtained by statistical models built using software like IBM SPSS etc.
[14]. C. Veres, J. Garsi, C. Rubino, F. Pouzoulet, F. Bidault, J. Chavaudra, A. Bridier, M. Ricard, I. Ferreira, D. Lefkopoulos et al., Thyroid volume measurement in external beam radiotherapy patients using ct imaging: correlation with clinical and anthropometric characteristics, Physics in medicine and biology, vol. 55, no. 21, p. N507, 2010.
• [11] and [12] are the only papers that address the issue of Automatic Thyroid Segmentation and Volume estimation.
• Segmentation Results in [11]: • Volume estimation done using a new
scheme that makes use of Particle
swarm optimization
• Drawback: Uses information obtained
from CT scans to determine volume
of US image
• Volume Segmentation Results in [12]: • Drawback: Uses biased ellipsoid model
[11] C.Y. Chang, Y. F. Lei, C.H. Tseng, and S.R. Shih, Thyroid Segmentation and Volume Estimation in Ultrasound Images, IEEE Trans. on biomedical engineering, vol. 57, no. 6, June 2010 [12]. Kollorz, E.N.K., et al., Quantification of thyroid volume using 3-D ultrasound imaging. Ieee Transactions on Medical Imaging, 2008. 27(4): p. 457-466.
Computer Aided Methods
Drawbacks of Existing methods
• All volume estimates are lower than the actual volume
• Ellipsoid Method is biased [1-4,7]
• Doctors want new mathematical formulation of volume measurement – NONE FOUND SO FAR!
• Volume calculation inaccurate when a nodule(s) is present [1-4,7]
• Volumes are mostly manually determined and have a high inter-observer variance.
[1]. Nygaard, B., et al., Thyroid volume measured by ultrasonography and CT. Acta Radiologica, 2002. 43(3): p. 269-274. [2]. van Isselt, J.W., et al., Comparison of methods for thyroid volume estimation in patients with Graves' disease. European Journal of Nuclear Medicine and Molecular Imaging, 2003. 30(4): p. 525-531. [3]. Kollorz, E.N.K., et al., Quantification of thyroid volume using 3-D ultrasound imaging. IEEE Transactions on Medical Imaging, 2008. 27(4): p. 457-466. [4]. Ruggieri, M., et al., The estimation of the thyroid volume before surgery - an important prerequisite for minimally invasive thyroidectomy. Langenbecks Archives of Surgery, 2008. 393(5): p. 721-724. [7]. Trimboli, P., et al., A mathematical formula to estimate in vivo thyroid volume from two-dimensional ultrasonography. Thyroid, 2008. 18(8): p. 879-882.
Proposed scheme
• Use 2D of ultrasound images for volume estimation
Validation
Volume estimation
Thyroid gland segmentation
Preprocessing for artifact removal
Preprocessing for Artifact Removal
• Manually Induced artifacts in Ultrasound Images:
• Nodule markings by radiologists
• Gland description
• Machine induced markings
• 150 out of 190 images had artifacts
• Application of POC based methods [8-10].
[8]. P. Marziliano, M. Vetterli, Irregular sampling in approximation subspaces, SampTA, Loen, Norway, August 1999. [9]. Zhu, X., A.T.S. Ho, and P. Marziliano, Image authentication and restoration using irregular sampling for traffic enforcement applications. ICICIC 2006: First International Conference on Innovative Computing, Information and Control, Vol 3, Proceedings, ed. J.S. Pan, P. Shi, and Y. Zhao2006. 62-65. [10]. Zhu, X., A.T.S. Ho, and P. Marziliano, A new semi-fragile image watermarking with robust tampering restoration using irregular sampling. Signal Processing-Image Communication, 2007. 22(5): p. 515-528.
Artifact Removal - Results
Before artifact removal After artifact removal
N. S. Narayan, P. Marziliano and C. G. L. Hobbs, Automatic Removal of Manually Induced Artefacts in Ultrasound Images of Thyroid Gland, 35th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (IEEE EMBC), July 3-7, 2013.
Qualitative Analysis
• Restored image PSNR ~ 40dB
• Algorithm converges within 3 iterations
Convergence of POCS algorithm
Segmentation
• Novelty:
• Developed based purely on tissue echogenicity
• Fully Unsupervised utilizing the state of the art automatic cluster estimation techniques
• Speckle property of the image treated as a feature instead of noise
N. S. Narayan, P. Marziliano and C. G. L. Hobbs, Echogenicity Based Unsupervised Segmentation of Ultrasound Images of the Thyroid Gland, Submitted to The 16th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI), 2013.
Segmentation - Results
N. S. Narayan, P. Marziliano and C. G. L. Hobbs, Echogenicity Based Unsupervised Segmentation of Ultrasound Images of the Thyroid Gland, Submitted to The 16th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI), 2013.
Segmentation Accuracy
• Accurate to a tune of 90%
• Consistent cluster estimation up to very large thresholds
Stability of algorithm
Future work
• Pre-process for image based artifacts like Shadow and Enhancement artifacts
• Upgrade the segmentation algorithm
• Use of cluster validity schemes
• Active Contours to aid in segmentation
• Segment the thyroid gland from the image
• Developing model for volume estimation
• Thyroid Volume Calculation
• Validating the workflow