Table 5 Sensitivity analysis of hyperparameters on SICAPv2 dataset.
From: Enhanced hierarchical attention mechanism for mixed MIL in automatic Gleason grading and scoring
Hyperparameter | Value | AUC | K-scores | Training time (h) |
|---|---|---|---|---|
Learning rate | 1e−5 | 0.9341 | 0.7924 | 12. 5 |
1e−4 | 0.9617 | 0.8504 | 10.2 | |
1e−3 | 0.9193 | 0.7612 | 8. 7 | |
Superpixel num | 100 | 0.9454 | 0.8118 | 9. 8 |
200 | 0.9617 | 0.8504 | 10.2 | |
300 | 0.9532 | 0.8308 | 14. 5 |
