Performance Evaluation of Active Contour Models for Carotid Plaque Segmentation

Abstract

In this paper we compare four different snakes segmentation methods for segmenting the athrerosclerotic carotid plaque from longitudinal ultrasound images of the carotid artery. The accuracy and reproducibility of the methods were tested on 80 images and the results were compared with the manual delineations of an expert. The comparison showed that the Lai&Chin snakes segmentation method gave better results with the true positive fraction, TPF, true negative fraction TNF, Williams index, KI, and overlap index, to be 82.70%, 80.89%, 80.66%, and 69.30% respectively. Better false negative fraction, FNF, and false positive fraction, FPF, were given by the Balloon method with 13.90% and, 5.40% respectively.