Graph cut based algorithm for corpus callosum segmentation from diffusion tensor images

Noises that usually appear in the process of diffusion tensor magnetic resonance imaging could result in voxel data distortion in diffusion tensor images and poor segmentation. A method based on graph cut is proposed to deal with this problem. In the process of solving the energy function of this method, T-link weights are computed using the median of the J-divergence values, one of tensor dissimilarity metrics, among non-seed points and the seed points that are regarded as hard constraints. Meanwhile, N-link weights are computed using a monotonically decreasing exponential function ranged in (0, 1]. At the same time, we construct a grid graph, and calculate the minimum cut through max-flow/min-cut algorithm to achieve the global optimal binary segmentation of images. Experimental results showed that the proposed method extracted corpus callosum more accurately from the white matter tissue, and statistical analysis of overlap rates with different values of parameters also proved the new algorithm has higher accuracy of the segmentation.