Early diagnosis of cirrhosis via automatic location and geometric description of liver capsule

Abstract

We propose in this paper an automatic method for early diagnosis of cirrhosis using high-frequency ultrasound images. Instead of analyzing image texture, our method exploits image characteristics of liver capsule. To this end, we first propose a novel spatial context-constrained multi-scale method to accurately extract the boundaries of the liver capsule. Our approach detects all the possible edges in scale space, and the irrelevant edges are then filtered out with a spatial context-based energy function. Secondly on this basis, two novel descriptors are proposed to characterize the geometric properties of liver capsule and the changes of liver capsule with the aggravation of liver cirrhosis. These two descriptors are used as features fed into a support vector machine classifier for quantitative analysis in automatic diagnosis. Experiment results show that the proposed method can reliably localize liver capsule and accurately classify ultrasound liver images into normal and cirrhosis classes automatically.