TABLE 4.
Classification performance compared to CNN and Transformer-based state-of-the-art models for APTOS 2019 dataset.
| Method | Accuracy | Precision | Sensitivity | F1-score |
|---|---|---|---|---|
| CapsNet + VGG16 (Kumar et al., 2021) | 75.50 | — | — | — |
| NASNet (Dondeti et al., 2020) | 77.90 | 76.00 | 77.00 | |
| MobileNet (Kassani et al., 2019) | 79.01 | — | 76.47 | — |
| DNN + Blended VGG and Xception (Bodapati et al., 2020) | 80.96 | — | — | — |
| Inception-ResNet-v2 (Gangwar and Ravi, 2021) | 82.18 | — | — | — |
| Composite Gated attention DNN (Bodapati et al., 2021a) | 82.54 | 82.00 | 83.00 | 82.00 |
| Modified Xception (Kassani et al., 2019) | 83.09 | — | 88.24 | — |
| Modified VGG16 (Bodapati et al., 2021b) | 84.31 | — | — | 84.00 |
| LA-NSVM (Shaik and Cherukuri, 2021) | 84.31 | 75.86 | 66.16 | 69.90 |
| TMILv4 (Yang et al., 2024) | 85.60 | — | 73.70 | — |
| SwAV | 87.00 | 86.00 | 87.00 | 86.00 |