Table 9 Quantitative comparison between (SA + ResNet50) model and other classification networks (InceptionV3, VGG16, ResNet101, VGG19, and ResNet18 each with or without SA, and also ResNet50 without SA). For all of them we use Adam Optimizer during the training phase.
| Â | TP | TN | FP | FN | Precision | Sensitivity or Recall (%) | F1-Score (%) | Specificity (%) | Accuracy (%) | AUC (%) |
|---|---|---|---|---|---|---|---|---|---|---|
SA + ResNet50 | 129 | 393 | 3 | 3 | 0.977 | 97.7 | 97.7 | 99.2 | 97.7 | 99 |
ResNet50 | 126 | 390 | 6 | 6 | 0.9545 | 95.5 | 95.8 | 98.5 | 95.5 | 98.7 |
SA +VGG16 | 26 | 368 | 28 | 106 | 0.4815 | 19.7 | 28 | 94.5 | 47 | 71 |
VGG16 | 52 | 357 | 39 | 80 | 0.57 | 39.4 | 46.6 | 90.2 | 48.5 | 77.65 |
SA + InceptionV3 | 125 | 389 | 7 | 7 | 0.947 | 94.7 | 94.7 | 98.2 | 94.7 | 98 |
InceptionV3 | 126 | 391 | 5 | 6 | 0.96 | 95.5 | 95.8 | 98.7 | 96.2 | 98.76 |
SA + ResNet101 | 122 | 387 | 9 | 10 | 0.9313 | 92.42 | 97.7 | 92.8 | 92.4 | 98.62 |
ResNet101 | 113 | 377 | 19 | 19 | 0.8561 | 85.61 | 95.2 | 85.6 | 85.6 | 94.67 |
SA +VGG19 | 24 | 363 | 33 | 108 | 0.4211 | 18.18 | 91.7 | 25.4 | 27.3 | 51.86 |
VGG19 | 40 | 377 | 19 | 92 | 0.678 | 30.30 | 95.2 | 41.9 | 53 | 80.39 |
SA + ResNet18 | 128 | 392 | 4 | 4 | 0.9697 | 96.97 | 97 | 99 | 96.7 | 98.41 |
ResNet18 | 125 | 389 | 7 | 7 | 0.947 | 94.7 | 94.7 | 98.2 | 94.7 | 98.63 |
