An attempt at beating the 3D U-Net

Authors
Isensee, Fabian
Maier-Hein, Klaus H.

Issue Date
2019

Publisher
University of Minnesota Libraries Publishing

Type
Article

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.

Identifiers
doi: 10.24926/548719.001