Cerebral White Matter Segmentation using Probabilistic Graph Cut Algorithm

Abstract

Accurate and efficient computer-assisted brain image segmentation methods are of great interest to both scientific and clinical researchers of the human central neural system. Cerebral white matter segmentation from Magnetic Resonance Imaging (MRI) data of brain remains a challenging problem due to a combination of several factors: noise and imaging artifacts, partial volume effects, intrinsic tissue variation due to neurodevelopment and neuropathologies, and the highly convoluted geometry of the cortex. We propose here a probabilistic variation of the traditional graph cut algorithm (IEEE international conference on computer vision, pp 105–112) with an improved parameter selection mechanism for the energy function, to be optimized in a graph cut problem. In addition, we use a simple yet effective shape prior in form of a series of ellipses to increase the computational efficiency of the proposed algorithm and improve the quality of the segmentation by modeling the contours of the human skull in various 2D slices of the sequence. Qualitative as well as quantitative segmentation results on T1-weighted MRI input, for both 2D and 3D cases are included. These results indicate that the proposed probabilistic graph cut algorithm outperforms some of the state-of-the art segmentation algorithms like the traditional graph cut (IEEE international conference on computer vision, pp 105–112) and the expectation maximization segmentation (IEEE Trans Med Imaging 20(8):677–688, 2001).