Interactive Differential Segmentation of the Prostate using Graph-Cuts with a Feature Detector-based Boundary Term

In this paper we present a modified boundary term for Graph-Cuts, which enables the latter to couple with feature detectors that return a confidence with respect to the detected image feature. Such detectors lead to improved localisation of boundaries in challenging images, which are often undetected by the implicit intensity-based edge detection scheme of the original method. This is particularly true for medical image segmentation, due to complex organ appearance, partial volume effect and weak intensity contrast at boundaries. The novel term is validated via its application to the differential segmentation of the prostate. The results demonstrate considerable improvement over classical Graph-Cuts of the Central Gland / Peripheral Zone separation when it is coupled with a SUSAN edge detector.