Tuning U-Net for Brain Tumor Segmentation

We propose a solution for BraTS22 challenge that builds on top of our previous submission—Optimized U-Net method. This year we focused on improving the model architecture and training schedule. The proposed method further improves scores on both our internal cross validation and challenge validation data. The validation mean dice scores are: ET 0.8381, TC 0.8802, WT 0.9292, and mean Hausdorff95: ET 14.460, TC 5.840, WT 3.594.