Segmentation of Brain Tumor Tissues with Convolutional Neural Networks

In this work, we investigate the possibility to directly apply convolutional neural networks (CNN) to segmentation of brain tumor tissues. As input to the network, we use multi-channel intensity information from a small patch around each point to be labelled. Only standard intensity pre-processing is applied to the input data to account for scanner differences. No post-processing is applied to the output of the CNN. We report promising preliminary results on the high-grade training data from the BraTS 2013 challenge. Work for the final submission will include architecture modifications, parameter tuning and training on the BraTS 2014 training corpus.