Abstract
Automated brain tumour segmentation has the potential of making a massiveimprovement in disease diagnosis, surgery, monitoring and surveillance.However, this task is extremely challenging. Here, we describe our automatedsegmentation method using 2D CNNs that are based on U-Net. To deal with classimbalance effectively, we have formulated a weighted Dice loss function. Wefound that increasing the depth of the 'U' shape beyond a certain level resultsin a decrease in performance, so it is essential to choose an optimum depth. Wealso found that 3D contextual information cannot be captured by a single 2Dnetwork that is trained with patches extracted from multiple views whereas anensemble of three 2D networks trained in multiple views can effectively capturethe information and deliver much better performance. Our method obtained Dicescores of 0.79 for enhancing tumour, 0.90 for whole tumour, and 0.82 for tumourcore on the BraTS 2018 validation set and its performance is comparable to thestate-of-the-art methods.