Table 2 Performance of proposed models and other SOTA methods on different dataset.
Dataset | Models | ACC | SEN | SPE | PRE | F1-score |
|---|---|---|---|---|---|---|
Public dataset collected from seven public dermatology atlas websites: DermNet, DermNet NZ, AtlasDerm, DermIS, SD-260, Kaggle, and DanDerm | (a) Proposed models | |||||
ResNet34 | 84.43% | 84.97% | 83.60% | 85.38% | 85.17% | |
ResNet50 | 85.32% | 85.45% | 85.19% | 86.67% | 86.06% | |
ResNet101 | 86.82% | 87.79% | 85.71% | 87.38% | 87.58% | |
Swin Transformer Base | 90.80% | 90.61% | 91.01% | 91.90% | 91.25% | |
Swin Transformer Large | 92.78% | 92.96% | 92.59% | 93.40% | 93.18% | |
(b) Reference29 | ||||||
VGG | 96.77% | 97.20% | 96.30% | 96.73% | 96.80% | |
ResNet | 95.27% | 95.10% | 95.60% | 96.19% | 95.80% | |
DenseNet | 96.27% | 96.20% | 96.10% | 96.70% | 96.20% | |
Person dataset provided by Department of Cosmetic Laser Surgery in the Hospital for Skin Disease and Institute of Dermatology, Chinese Academy of Medical Sciences (CAMS) & Peking Union Medical College | (a) Proposed models | |||||
ResNet34 | 81.65% | 82.04% | 81.00% | 87.72% | 84.78% | |
ResNet50 | 82.40% | 82.63% | 82.00% | 88.46% | 85.45% | |
ResNet101 | 83.52% | 84.43% | 82.00% | 88.68% | 86.96% | |
Swin Transformer Base | 86.89% | 87.43% | 86.00% | 91.82% | 89.57% | |
Swin Transformer Large | 87.64% | 88.02% | 87.00% | 91.88% | 89.91% | |
(b) Reference30 | ||||||
YOLO V3 | 85.02% | 92.91% | 72.00% | 84.70% | 88.62% | |
