We propose an anatomical convolutional module to couple anatomical information into deep neural network. We further develop a loss function based on the mass center of individual lesions called lesion-wise loss, which can regularize the network training, thereby improving the performance of lesion localization and segmentation. We validate our methods on a public dataset, ISBI-15 Multiple Sclerosis Lesion Segmentation Challenge [1], where the results showed that we achieved the best performance on all published methods.
This abstract and the presentation materials are available to members only; a login is required.