Patient-level enhancement detection on 1,109 studies (test)
83Balanced AccuracynnU-Net
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| nnU-NetMethodology=bootstrap mean ± SD (1,000 resamples, seed 42)2025.08 | 83 | 91.5 | 91.5 | 74.4 | 96.8 | 94.1 | |
| SegResNetMethodology=bootstrap mean ± SD (1,000 resamples, seed 42)2025.08 | 74.8 | 76.2 | 76.2 | 73.5 | 96.1 | 84.9 | |
| SwinUNETRMethodology=bootstrap mean ± SD (1,000 resamples, seed 42)2025.08 | 71.1 | 74.6 | 74.6 | 67.6 | 95.1 | 83.6 | |
| Expert radiologistsn=11, Methodology=aggregate across 1,100 blind reads (SD across per-rater metrics)2025.08 | 69.8 | 74.9 | 74.9 | 64.7 | 68 | 71.3 |