# An attempt at beating the 3D U-Net

## TL;DR

This paper enhances the 3D U-Net architecture with residual blocks to improve medical image segmentation, achieving state-of-the-art results in the KiTS2019 challenge with a marginal performance gain.

## Contribution

The paper introduces residual and pre-activation residual blocks into 3D U-Net, demonstrating slight improvements and winning the KiTS2019 challenge.

## Key findings

- Residual 3D U-Net achieved higher dice scores.
- Outperformed all 105 competitors in KiTS2019.
- Won the challenge with a Composite Dice score of 91.23.

## Abstract

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.