Performance Comparison Between Automatic Liver Segmentation in Arterial and Portal Contrast-Enhancement Phases

Abstract

The injection of contrast medium is fundamental for diagnosing and differentiating liver cancers in Computed Tomography (CT). Computer-aided diagnosis systems have been developed to support radiologists in detecting liver lesions. However, the injection of contrast medium may change features and thus affect the automatic segmentation of the liver. In this paper, the performance of an automatic segmentation algorithm using a posteriori information of the liver was compared between the arterial and portal contrast-enhanced CT phases. By performing the liver segmentation with a region growing algorithm based on one-class support vector machines, 28 CT scans in both arterial and portal phases were evaluated. In general, the results show that there were no differences between segmentation performances for both phases. Major segmentation errors were usually related to intrinsic characteristics of the liver. Therefore, based on the results presented in this paper, it was possible to conclude that different contrast-enhanced CT phases do not affect liver segmentation significantly.