Training ROI selection based on MILBoost for liver cirrhosis classification using ultrasound images

Abstract

Ultrasound images are widely used for diagnosis of liver cirrhosis. In most of liver ultrasound images analysis, regions of interest (ROIs) are selected carefully, to use for feature extraction and classification. It is difficult to select ROIs exactly for training classifiers, because of the low SN ratio of ultrasound images. In these analyses, training sample selection is important issue to improve classification performance. In this article, we have proposed training ROI selection using MILBoost for liver cirrhosis classification. In our experiments, the proposed method was evaluated using manually selected ROIs. Experimental results show that the proposed method improve classification performance, compared to previous method, when qualities of class label for training sample are lower.