# nnU-Net: Self-adapting Framework for U-Net-Based Medical Image Segmentation

**Authors:** Fabian Isensee, Jens Petersen, Andre Klein, David Zimmerer, Paul F. Jaeger, Simon Kohl, Jakob Wasserthal, Gregor Koehler, Tobias Norajitra, Sebastian Wirkert, Klaus H. Maier-Hein  
**arXiv:** [1809.10486](https://arxiv.org/abs/1809.10486)  
**Date:** 2018-09-27  
**Summary page:** https://www.summarizepaper.com/en/paper/1809.10486v1/

## Key points

- U-Net introduced in 2015 is widely used in medical image segmentation due to its simple yet effective architecture
- Adapting U-Net to new challenges involves decisions about architecture, preprocessing, training, and inference that are interdependent and crucial for overall performance
- nnU-Net framework was developed as a robust and self-adapting solution based on 2D and 3D vanilla U-Nets
- nnU-Net focuses on stripping away unnecessary complexities from network designs and emphasizing key aspects that drive performance and generalizability
- Researchers led by Fabian Isensee evaluated nnU-Net in the context of the Medical Segmentation Decathlon challenge
- nnU-Net achieved the highest mean dice scores across all classes and excelled in seven phase 1 tasks on the online leaderboard of the challenge (except for class 1 in BrainTumour)

## Summary

, , , , 

The U-Net, introduced in 2015, has become a widely used benchmark in medical image segmentation due to its simple yet effective architecture. However, adapting the U-Net to new challenges involves making decisions about architecture, preprocessing, training, and inference that are interdependent and crucial for overall performance. In response to this complexity, the nnU-Net ('no-new-Net') framework was developed as a robust and self-adapting solution based on 2D and 3D vanilla U-Nets. The focus of nnU-Net is on stripping away unnecessary complexities from network designs and emphasizing the key aspects that drive performance and generalizability. A team of researchers led by Fabian Isensee, Jens Petersen, Andre Klein, David Zimmerer, Paul F. Jaeger, Simon Kohl, Jakob Wasserthal, Gregor Koehler, Tobias Norajitra, Sebastian Wirkert, and Klaus H. Maier-Hein evaluated the nnU-Net framework in the context of the Medical Segmentation Decathlon challenge. This challenge assesses segmentation performance across ten disciplines with diverse entities, image modalities, geometries, and dataset sizes without allowing manual adjustments between datasets. The results of their evaluation showed that nnU-Net achieved the highest mean dice scores across all classes and excelled in seven phase 1 tasks on the online leaderboard of the challenge (with the exception of class 1 in BrainTumour).