Look Ma, no code: fine tuning nnU-Net for the AutoPET II challenge by only adjusting its JSON plans

TL;DR

This paper demonstrates that simple modifications to nnU-Net's JSON configuration, such as changing the architecture and training parameters, can significantly improve performance on the AutoPET II challenge without altering the core code.

Contribution

The authors show that tuning nnU-Net via its JSON plans alone can lead to substantial performance gains, simplifying the process of model optimization.

Findings

Modified nnU-Net configuration achieved a Dice score of 65.14 compared to 33.28 baseline.

Ensembling top configurations further improved results.

Increased compute requirements for training.

Abstract

We participate in the AutoPET II challenge by modifying nnU-Net only through its easy to understand and modify 'nnUNetPlans.json' file. By switching to a UNet with residual encoder, increasing the batch size and increasing the patch size we obtain a configuration that substantially outperforms the automatically configured nnU-Net baseline (5-fold cross-validation Dice score of 65.14 vs 33.28) at the expense of increased compute requirements for model training. Our final submission ensembles the two most promising configurations.