Ultrasonic image analysis for liver diagnosis

Abstract

The authors propose a new ultrasonic image analysis system that can be utilized as an effective tool in classifying liver states as normal, hepatitis, or liver cirrhosis. In this system, the authors first define suitable settings for the ultrasonic device, then remove the inhomogeneous structures from the area of interest in the image, and then, by using the forward sequential search method, look for the useful texture parameters from the co-occurrence matrix, the statistical feature matrix, the texture spectrum, and the fractal dimension descriptors. Finally, the selected parameters are fed into a probabilistic neural network for the classification of liver disease. Experimental results are presented that show the classification rate with and without the inclusion of the inhomogeneous structures.