Global Planar Convolutions for improved context aggregation in Brain Tumor Segmentation with MR images

Brain tumors pose a significant social and economic burden worldwide. A key to improve the quality and expectancy of life of patients with brain tumors is to automate the process of delineation of tumoral structures. In this work we propose the Global Planar Convolution module, a building-block for Convolutional Neural Networks that enhances the context perception capabilities of segmentation networks for brain tumor segmentation. We show that such modules achieve similar performance to equivalent networks with increased depth, and provide an initial inspection of their behavior via interpretation of intermediate feature maps.