Liver fibrosis grade classification with B-mode ultrasound

B-mode images of 20 fresh postsurgical human liver samples were obtained to evaluate ultrasound ability in determining the grade of liver fibrosis. Image features derived from gray level concurrence and nonseparable wavelet transform were extracted to classify fibrosis with a classifier known as the support vector machine. Each liver sample subsequently underwent histologic examination and liver fibrosis was graded from 0 to 5 (i.e., six grades total). The six grades were then combined into two, three, four and six classes. Classifications with the extracted image features by the support vector machine were tested and correlated with histology. The results revealed that the best classification accuracy of two, three, four and six classes were 91%, 85%, 81% and 72%, respectively. Thus, liver fibrosis can be noninvasively characterized with B-mode ultrasound, even though the performance declines as the number of classes increases. The elastic constants of 16 samples out of a total of 20 were also correlated with the image features. The Pearson correlation coefficients indicated that the image features are more strongly correlated with the fibrosis grade than with the elastic constant.