A PC-based system for soft tissue classification

The author presents an algorithm and instrumentation for ultrasound classification of simulated human-liver tissue abnormalities. The tissue is simulated by a liver phantom that mimics the tissue acoustically. The instrumentation used is a 50 MHz microcomputer-based data acquisition and analysis system. The system digitizes the ultrasound backscattered signal from selected regions of the phantom and processes the digitized data for feature measurement. The algorithm is based on a three-layer backpropagation artificial neural network. The network is trained to differentiate between simulated normal and abnormal tissue and to classify three types of simulated abnormalities. The results of this study show that out of twenty-eight cases the system classifies twenty five correctly and fails to classify three cases. The reasons for this are discussed along with recommendations to increase the accuracy of classification.<>