Table 8 Performance metrics of CancerDet-Net model under various conditions on nine types of cancer dataset.
From: Cross-platform multi-cancer histopathology classification using local-window vision transformers
Condition | Value | Accuracy (%) | Precision (%) | Recall (%) | F1-score (%) |
|---|---|---|---|---|---|
Input shape | 128 × 128 | 98.51 | 98.52 | 98.51 | 98.51 |
64 × 64 | 94.44 | 94.39 | 94.48 | 94.43 | |
224 × 224 | 97.63 | 97.45 | 97.80 | 97.62 | |
Attention | Axial | 85.88 | 87.03 | 86.53 | 86.77 |
Linearized Self | 83.77 | 85.11 | 84.61 | 84.86 | |
Cross | 95.06 | 95.75 | 95.25 | 95.50 | |
Multi-Head Self | 94.33 | 94.51 | 94.51 | 94.63 | |
Local-Window | 98.51 | 98.52 | 98.51 | 98.51 | |
Block | Without the HMSGA | 96.55 | 95.21 | 95.84 | 95.52 |
Without the CFE | 94.44 | 94.45 | 94.44 | 94.44 | |
Without ViT | 92.31 | 92.02 | 92.60 | 92.31 | |
With 1 ViT | 95.23 | 95.00 | 94.20 | 94.60 | |
With 2 ViT | 98.51 | 98.52 | 98.51 | 98.51 |
