Genetic Algorithm Driven Statistically Deformed Models for Medical Image Segmentation

We present a novel evolutionary computing based approach to medical image segmentation. Our method complements the image-pixel integration power of deformable shape models with the high-level control mechanisms of genetic algorithms (GA). Specifically, the GAs alleviate typical deformable model weaknesses pertaining to model initialization, deformation parameter selection, and energy functional local minima through the simultaneous evolution of a large number of models. Furthermore, we constrain the evolution, and thus reduce the size of the search-space, by using statistically-based deformable models whose deformations are intuitive (stretch, bulge, bend) and driven in-terms of principal modes of variation of a learned mean shape. We demonstrate our work through the application to segmentation of the corpus callosum (CC) in mid-sagittal brain magnetic resonance images (MRI).