MR image segmentation using a fuzzy-based neural network

Most techniques for segmentation of magnetic resonance images of the brain are extremely time consuming and/or require extensive user interaction. An automated segmentation procedure is presented, whereby the fuzzy c-means classification results are used to train a feedforward neural network. The cascade correlation algorithm is used to optimize the network training process. After applying a brain-extraction technique, the segmented images are then used for rendering computer-generated images of the brain's surface. Experimental results using real, 3D magnetic resonance images are presented, demonstrating the performance of the segmentation as well as the final surface rendering.