A novel framework for automatic segmentation of kidney from DW-MRI

The segmentation of the kidney tissues is a key step in developing any non-invasive computer-aided diagnostic (CAD) system for early detection of acute renal transplant rejection. This paper introduces a geometric (level-set)-based deformable model approach for the 3D kidney segmentation from diffusion-weighted magnetic resonance imaging (DW-MRI). The proposed deformable model is guided by a stochastic speed relationship based on an adaptive shape prior guided by the visual appearance of the DW-MRI data. The voxel-wise guiding of the level-sets is obtained by integrating these three image features into a joint Markov-Gibbs random field (MGRF) model of the kidney and its background. The segmentation approach was evaluated for 40 DW-MRI data sets acquired at b-values ranging from 0 to 1000 s/mm² and compared against other segmentation methods using three evaluation metrics: the Dice similarity coefficient (DSC), the 95-percentile modified Hausdorff distance, and the absolute kidney volume difference. Experimental results´ evaluation between manually drawn and automatically segmented contours confirm the robustness and accuracy of the proposed approach.