Task2Baseline
About
Summary
Task 2 Baseline
Mechanism
My original idea for Task 2: Segmentation on MR-Linac MRIs. Apply nnU-Net’s pre-training and fine-tuning paradigm to fine-tune 3-fold ResEncMs using task 1’s best checkpoint and 50 labeled samples for task 2.
However, this strategy does not provide a competitive performance to significantly outperform the baseline method. Therefore, I return back to the baseline approach which utilizes 3-fold nnU-Net modules to perform the binary segmentation.
Different from the baseline's post-processing which uses only 1 fold of MRSegmentator for efficiency, I activate all 5 folds of MRSegmentator checkpoints. Although this is less computationally efficient, we can expect a better sketch of the pancreas organ, which helps reduce the false positives generated by the 3-fold nnU-Net modules.
I acknowledge this is only a minor modification of the baseline approach, but it does provide better segmentation performance.
Thank you for your time and reading.
Interfaces
This algorithm implements all of the following input-output combinations:
Validation and Performance
Challenge Performance
| Date | Challenge | Phase | Rank |
|---|---|---|---|
| July 27, 2025 | PANTHER | Open Development Phase Task 2 | 30 |
| Aug. 29, 2025 | PANTHER | Closed Testing Phase Task 2 | 2 |
Uses and Directions
This algorithm was developed for research purposes only.