Computerized Liver Segmentation from CT Images using Probabilistic Level Set Approach

Abstract

Accurate segmentation of patient’s liver from his/her computed tomography–angiography (CTA) images is the preliminary component for a reliable computerized liver evaluation system. Flawlessness in liver diagnosis relies upon the precision in the segmentation of liver region from all the slices/images in a given patient dataset. Nevertheless, with the challenges like intensity similarity, partial volume effect of liver with its adjacent abdominal organs and liver shape variability across patients, achieving automated optimal liver region segmentation from acquired CT scans is difficult. This paper proposes a semisupervised liver segmentation technique, which adjusts the segmentation parameters for each patient through continuous learning of patient’s CTA dataset properties in a Bayesian level set framework to address all the aforementioned challenges. In this framework, Bayesian probability model with spatial prior is utilized to initiate the level set and to derive an enhanced variable force and edge indication function that helps level set evolution to reach genuine liver boundaries in reduced time. The proposed model has been validated on standard MICCAI liver dataset, producing accuracy score of 79.