A Computer-Aided Classification System for Utrasonic Liver Images

• Main Authors: Jen-Ya Wang, 王健亞
• Other Authors: Yung-Nien Sun
• Format: Others
• Language: zh-TW
• Published: 1994
• Online Access: http://ndltd.ncl.edu.tw/handle/13585854566761979676

碩士 === 國立成功大學 === 資訊及電子工程研究所 === 82 === Ultrasonic imaging is a popular and non-invasive tool frequently used in the diagnoses of liver diseases. However, the ultrasonic imaging condition, including focusing and time- gain control(TGC), and the heterogeneous textures, such as blood vessels and hepatic ducts in the acquired echo images, may confuse the echo texture analysis and induce incorrect texture measurement. To obtain suitable quality sonogram and reliable measurement, we have done many experiments for seeking the best suitable acquisition setting. The sonograms are acquired from inter-coastal view and the setting of focusing and TGC is selected and fixed specifically for our ultrasonic imaging machine. The heterogeneous textures are excluded from the liver texture measurement by using multiscale edge detector based on wavelet transform associated with B-spline interpolation. To improve the accuracy of echo texture analysis, the blood vessels and hepatic ducts are excluded in texture analysis. In texture analysis, a new method, called "Occurrence Probability of Texture Structure," is proposed in this thesis. The texture measures are computed based on the surface structure in a small local area. Besides, co-occurrence matrix, statistical feature matrix, texture spectrum, fractal dimension descriptor and this method are used to understand and describe the ultrasonic liver images, too. From all these texture descriptors, we can obtain 126 texture parameters. However, not all of these parameters are useful in liver diagnosis. Thus, we search for the effective classification parameters by using the forward sequential search algorithm. The useful parameters are sent into a probabilistic neural network for the classification of liver diseases. A relative score is computed to indicate the exacerbation of progressive liver diseases. The experimental results show that the correct classification rate is greatly improved in comparison with the conventional methods without removal of heterogeneous structures.