nnU-Net for Brain Tumor Segmentation

We apply nnU-Net to the segmentation task of BraTS 2020 challenge. The unmodified baseline configuration already achieves a respectable result. By incorporating BraTS-specific modifications regarding postprocessing, region-based training, more aggressive data augmentation as well several minor pipeline we are able improve its performance substantially. furthermore re-implement ranking scheme determine which our variants best fits requirements imposed by it. Our method took first place in competition with Dice scores 88.95, 85.06 and 82.03 HD95 values 8.498,17.337 17.805 for whole tumor, tumor core enhancing respectively.