Look Ma, no code: fine tuning nnU-Net for the AutoPET II challenge by only adjusting its JSON plans
This work addresses medical image segmentation for PET scans, but it is incremental as it builds on the existing nnU-Net framework with minor adjustments.
The authors tackled the AutoPET II challenge by fine-tuning nnU-Net solely through JSON plan modifications, achieving a Dice score of 65.14 compared to the baseline of 33.28 via a residual encoder, increased batch size, and patch size, with a final ensemble submission.
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.