Three-dimensional medical image segmentation using a graph-theoretic energy-minimisation approach

摘要

A new graph algorithm for multiscale segmentation of three-dimensional medical data sets is presented. It is a three-dimensional generalisation of an existing two-dimensional Mumford-Shah region-merging segmentation algorithm. The Mumford-Shah functional formulation leads to improved segmentation results compared to alternative approaches; the graph theoretic approach yields improved performance and simplified data structures; and the automated stopping estimation allows a fully non-supervised algorithm with no tuning parameters required (except for an optional parameter to select the depth of segmentation).