PET-guided liver segmentation for low-contrast CT via regularized Chan-Vese model

In this paper, we propose an automated liver segmentation method to overcome the challenging issue of similar intensities shared by liver and its surrounding tissues in low-contrast CT images. Our approach takes advantage of PET data to initialize the CT liver region of interest (ROI), and then applies anisotropic diffusion on the CT liver ROI to suppress the intensity values of adjacent structures and hence to highlight the liver region. The regularized 3D Chan-Vese level-set model with distance regularized term is introduced to segment the CT liver volume. Experimental results on 40 clinical PET-CT studies demonstrated that without relying on any training datasets, our method achieved accurate and robust normal liver segmentation in low-contrast CT volumes from PET-CT scanners.