Table 1 Results for the BCCC dataset, the ablation study refers to the model which was trained from scratched and tested on the BCCC dataset
Task BCCC dataset | Sub-class | Balanced accuracy (%) | Sensitivity (%) | Specificity (%) | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
Pretrained | Finetuned | Ablation study | Pretrained | Finetuned | Ablation study | Pretrained | Finetuned | Ablation study | ||
2 Classes | 0 - No Tumor | 71.1 | 91.7 | 81.0 | 62.6 | 89.5 | 81.6 | 79.5 | 95.0 | 80.5 |
1 - Tumor | 79.5 | 95.0 | 80.5 | 62.6 | 89.5 | 81.6 | ||||
3 Classes | 0 - No tumor | 63.1 | 82.1 | 69.0 | 60.5 | 89.5 | 90.0 | 85.0 | 88.6 | 67.0 |
1 - Low risk | 60.0 | 81.8 | 41.8 | 73.9 | 88.7 | 89.3 | ||||
2 - High Risk | 68.9 | 72.2 | 57.8 | 85.3 | 94.5 | 91.7 | ||||
5 Classes | 0 - No Tumor | 50.0 | 75.3 | 63.6 | 64.7 | 92.5 | 86.3 | 84.0 | 90.3 | 73.0 |
1 - Type IB | 49.1 | 80.0 | 21.8 | 82.1 | 95.0 | 94.0 | ||||
2 - Type IA | 40.0 | 80.0 | 58.2 | 91.0 | 96.6 | 95.5 | ||||
3 - Type II | 32.3 | 61.5 | 53.8 | 91.4 | 95.0 | 83.7 | ||||
4 - Type III | 60.0 | 80.0 | 20.0 | 91.5 | 96.8 | 100 | ||||