Robust segmentation of medical images using geometric deformable models and a dynamic speed function

Abstract

A number of methods based on geometric deformable models have been proposed recently to segment medical images. These methods require the definition of a speed function that governs model deformation. In this paper we propose a new speed function that is modified dynamically as the front pro­gresses. This new speed function is particularly well adapted to situations where edges are ill-defmed, adjacent structures have comparable intensity values, and images are noisy. We illustrate qualitatively the performance of our approach on a variety of MR, CT, and ultrasound images. The examples show the gener­ality of our approach and its insensitivity to parameters. We also evaluate it quantitatively on several CT scans of the liver.