Table 6.
Skin cancer segmentation results.
| Model | Accuracy (%) | Recall (%) | Specificity (%) | F-measure | Dice (%) | IoU (%) |
|---|---|---|---|---|---|---|
| U-Net (Augmentation) | 88.67 | 94.73 | 73.56 | – | 88.7 ± 0.42 | 81.5 ± 0.37 |
| U-Net (No Augmentation) | 89.67 | 97.08 | 70.93 | – | 86.2 ± 0.48 | 78.8 ± 0.43 |
| LBP | 98.80 | 95.84 | 99.20 | 0.95 | 83.5 ± 0.50 | 75.6 ± 0.48 |
| GLCM | 97.47 | 75.98 | 98.67 | 0.76 | 81.0 ± 0.54 | 72.9 ± 0.51 |
| Transfer learning | 85.39 | 94.38 | 80.45 | 0.82 | 87.6 ± 0.44 | 80.3 ± 0.39 |
U-Net and traditional methods (LBP and GLCM) results are shown with a clear indication of augmentation usage, facilitating direct comparison. Bold values indicate the best results obtained.