Level set method based on a statistical shape constraint for MRI brain segmentation

A reliable automatic segmentation of medical images requires a priori knowledge on the anatomical structures to be extracted. In this paper, we propose a new segmentation technique based on the level set method with the integration of a priori knowledge on the studied structures. This knowledge is represented as a statistical shape constraint, deduced from a training set of known shapes. This constraint is incorporated into the segmentation step guide the evolution of the initial curve and ensures that the deformation is limited in an allowable shape domain. The developed method has been tested on a database of 20 MR images in order to extract some internal brain structures: caudate, putamen and Thalamus. The results show that integrating the statistical shape constraint provides a marked improvement in the accuracy of segmentation.