14
Chien-Cheng Lee, Sz-Han Chen, Hong-Ming Tsai, Pau-Choo Chung, and Yu-Chun Chiang Department of Communications Engineering, Yuan Ze University Chungli, Taoyuan 320, Taiwan

Chien-Cheng Lee, Sz-Han Chen, Hong-Ming Tsai, Pau- Choo Chung, and Yu-Chun Chiang Department of Communications Engineering, Yuan Ze University Chungli,

Embed Size (px)

Citation preview

Page 1: Chien-Cheng Lee, Sz-Han Chen, Hong-Ming Tsai, Pau- Choo Chung, and Yu-Chun Chiang Department of Communications Engineering, Yuan Ze University Chungli,

Chien-Cheng Lee, Sz-Han Chen, Hong-Ming Tsai, Pau-Choo Chung, and Yu-Chun Chiang

Department of Communications Engineering, Yuan Ze University Chungli, Taoyuan 320,

Taiwan

Page 2: Chien-Cheng Lee, Sz-Han Chen, Hong-Ming Tsai, Pau- Choo Chung, and Yu-Chun Chiang Department of Communications Engineering, Yuan Ze University Chungli,

IntroductionThe accurate decision rate estimated by

using only simple visual interpretation of liver diseases is around 72%.

Page 3: Chien-Cheng Lee, Sz-Han Chen, Hong-Ming Tsai, Pau- Choo Chung, and Yu-Chun Chiang Department of Communications Engineering, Yuan Ze University Chungli,

In this papperThe diagnosis scheme includes two steps:

features extraction classification

Three kinds of liver diseases are identified: cyst, cavernous hepatomaHemangioma

Page 4: Chien-Cheng Lee, Sz-Han Chen, Hong-Ming Tsai, Pau- Choo Chung, and Yu-Chun Chiang Department of Communications Engineering, Yuan Ze University Chungli,

Features extractionGabor filters have the ability to model the

frequency and orientation sensitivity characteristic of the human visual system.

The features are optimal in the sense of minimizing the joint two-dimensional uncertainty in space and frequency.

Page 5: Chien-Cheng Lee, Sz-Han Chen, Hong-Ming Tsai, Pau- Choo Chung, and Yu-Chun Chiang Department of Communications Engineering, Yuan Ze University Chungli,

2D Gabor filter

Where,

2

22

2 2

yxexp

2

1y , xg

‧ 1j

Frequency: ψOrientation: θBandwidth: σ

θθψ‧θψ sinycosx2jexp y , xgy , xG , ,

Page 6: Chien-Cheng Lee, Sz-Han Chen, Hong-Ming Tsai, Pau- Choo Chung, and Yu-Chun Chiang Department of Communications Engineering, Yuan Ze University Chungli,

2D Gabor filter2-D convolution with image:

),(),(, ,, yxjGyxGyxGWhere IR θψ

yxG R , = θθψ‧ sinycosx2 cos y , xg

yxG I , = θθψ‧ sinycosx2sin y , xg

j i

, , j,iG jy,ixfy,x G θψ‧

The convolution is implemented using the mask of M×M sizes, which M is preferred to be an odd number.

Page 7: Chien-Cheng Lee, Sz-Han Chen, Hong-Ming Tsai, Pau- Choo Chung, and Yu-Chun Chiang Department of Communications Engineering, Yuan Ze University Chungli,

2D Gabor filterEnergy

Minimum Energy

22 ,,,,,,,,, θψθψ yxGyxGyxE IR

22 ,,,,,,,,, θψθψ yxGyxGyxEMin IR

Page 8: Chien-Cheng Lee, Sz-Han Chen, Hong-Ming Tsai, Pau- Choo Chung, and Yu-Chun Chiang Department of Communications Engineering, Yuan Ze University Chungli,

Supervised diseases classificationSupport Vector Machines

Train linear machines with marginsPreprocessing the data to represent patterns in

a high dimension with an appropriate nonlinear mapping Data from two categories can be linearly separable

Find the separating plane with largest margin The larger the margin, the better generalization of

the classifier

Page 9: Chien-Cheng Lee, Sz-Han Chen, Hong-Ming Tsai, Pau- Choo Chung, and Yu-Chun Chiang Department of Communications Engineering, Yuan Ze University Chungli,

SVM

Page 10: Chien-Cheng Lee, Sz-Han Chen, Hong-Ming Tsai, Pau- Choo Chung, and Yu-Chun Chiang Department of Communications Engineering, Yuan Ze University Chungli,

SVMFor linearly separable data set, the optimal

separating hyper-planes can be defined as follows:

where is a subset of the training patterns calledSupport Vectors (SVs).

Page 11: Chien-Cheng Lee, Sz-Han Chen, Hong-Ming Tsai, Pau- Choo Chung, and Yu-Chun Chiang Department of Communications Engineering, Yuan Ze University Chungli,

SVMThe coefficients and b are obtained by

solving the optimization problem:

The parameter C is a regularization parameter selected by the user. C corresponds to assigning a payment to the training errors.

Page 12: Chien-Cheng Lee, Sz-Han Chen, Hong-Ming Tsai, Pau- Choo Chung, and Yu-Chun Chiang Department of Communications Engineering, Yuan Ze University Chungli,

ExperimentThe images is 512 × 512 with contrast media

injection and the graylevel is stored at 12 bits per pixel, include76 liver cysts30 hepatomas 40 cavernous hemangiomas

Page 13: Chien-Cheng Lee, Sz-Han Chen, Hong-Ming Tsai, Pau- Choo Chung, and Yu-Chun Chiang Department of Communications Engineering, Yuan Ze University Chungli,
Page 14: Chien-Cheng Lee, Sz-Han Chen, Hong-Ming Tsai, Pau- Choo Chung, and Yu-Chun Chiang Department of Communications Engineering, Yuan Ze University Chungli,

Result