A Hybrid Approach to Automated Delineation of Brain Tissue in Alzheimer MR Images

Accurate delineation of brain tissue from Magnetic Resonance (MR) brain images is an important preprocessing task in many neuroimaging studies. Active contour methods have been widely used in the segmentation of brain. But they are semiautomatic as the user has to initialize the contour and their results greatly depend on the initialized contour. In this work, a method is described to combine a modified level set framework with Otsu algorithm for automatic and accurate segmentation of brain tissue and better approximation of initial contour. MR images used are acquired from open source online database OASIS. A mask image is obtained by performing Otsu thresholding on the MR image and the maximum connected region in the mask is given as the initial contour to the level set framework. A distance regularization term is included in the energy optimization function of the level set in order to maintain stability of the evolving contour. The results of segmentation are validated by comparing with the ground truth using similarity measures.