An attempt at beating the 3D U-Net
This work addresses medical image segmentation for kidney and tumor analysis, but it is incremental as it builds on the established U-Net architecture with minor modifications.
The authors tackled the problem of kidney and tumor segmentation by augmenting a 3D U-Net with residual blocks, achieving a Composite Dice score of 91.23 on the test set, which won the KiTS2019 challenge by outperforming 105 competing teams.
The U-Net is arguably the most successful segmentation architecture in the medical domain. Here we apply a 3D U-Net to the 2019 Kidney and Kidney Tumor Segmentation Challenge and attempt to improve upon it by augmenting it with residual and pre-activation residual blocks. Cross-validation results on the training cases suggest only very minor, barely measurable improvements. Due to marginally higher dice scores, the residual 3D U-Net is chosen for test set prediction. With a Composite Dice score of 91.23 on the test set, our method outperformed all 105 competing teams and won the KiTS2019 challenge by a small margin.