Evaluation of texture analysis techniques for quantitative characterization of ultrasonic liver images

In this study, the authors attempt to determine the efficacy of computer assisted ultrasonic liver tissue characterization using texture analysis techniques. Two different algorithms were used in this study; the gray level difference statistics (GLDS) and the fractal dimension texture analysis (FDTA). Both techniques were applied on three sets of ultrasonic liver images, normal-hepatoma-cirrhosis, all histologically proven. In all images, 32×32 pixel rectangular regions-of-interest were selected by specialized physicians and used in the analysis. FDTA was able to differentiate hepatoma from cirrhosis and normal liver with an accuracy of 90% and the GLDS was able to differentiate cirrhosis from normal with an accuracy of 75%. The combination of the two techniques proved to differentiate the three types with an overall accuracy of 81.7%