Computerized ultrasound image analysis for noninvasive evaluation of hepatic steatosis.

Abstract

Objective: To assess the diagnostic value of a particular set of local intensity parameters extracted from ultrasound liver images in conjunction with support vector machine (SVM) classifiers for liver steatosis grading in respect to the “gold standard” provided by liver biopsy. Material and methods: We prospectively enrolled in the study 228 patients with chronic hepatopathies. All the patients underwent liver biopsy and abdominal ultrasound examination. For quantitative ultrasound assessment of liver steatosis, an image analysis software was developed, which extracts three local intensity parameters from regions of interest (ROI) in the ultrasound section and analyzes their depth variation: the coefficient of variation of luminance (CVL), the median luminance (ml ), and the hepato-splenic attenuation index (HSAI). For steatosis grading, SVM classifiers were trained on the input feature spaces provided by the above mentioned parameters. The statistical significance of the steatosis grading was assessed on a significant test set using SVM classifiers, in terms of sensibility, specificity and through the ROC curves. Results: A cut-off value of 0.362 of the CVL of the liver performed the liver steatosis grading with an accuracy of 89.17% (p<0.0001). A cut-off value of 0.27 of the HSAI performed the prediction of the moderate-severe liver steatosis with an accuracy of 87%. Conclusions: The proposed computer analysis method of ultrasound images proved innovative and useful for the initial non-invasive assessment and grading of liver steatosis, with an additional advantage of reduced computational complexity and accessibility. The CVL provided a very good accuracy (89.17%) for an AUROC of 0.923 for the classification of liver steatosis in two severity categories (mild versus moderate-severe).