Table 2 Performance evaluation of different CNN models on test images
Input | Model | Sensitivity | Specificity | Accuracy | AUC | Kappa |
|---|---|---|---|---|---|---|
T1-WI | Resnet-50 | 55.9% | 76.9% | 66.9% | 0.714 | 0.331 |
T2-WI | Resnet-50 | 40.0% | 78.8% | 60.9% | 0.649 | 0.193 |
CE-T1 | Resnet-50 | 74.2% | 80.9% | 77.6% | 0.855 | 0.552 |
T1-WI | Efficientnet-b2 | 86.4% | 69.2% | 77.4% | 0.854 | 0.552 |
T2-WI | Efficientnet-b2 | 72.9% | 72.7% | 72.8% | 0.769 | 0.455 |
CE-T1 | Efficientnet-b2 | 81.8% | 76.5% | 79.1% | 0.890 | 0.582 |
T1-WI | Inception-v3 | 76.3% | 76.9% | 76.6% | 0.844 | 0.532 |
T2-WI | Inception-v3 | 74.1% | 82.5% | 78.6% | 0.874 | 0.568 |
CE-T1 | Inception-v3 | 83.3% | 75.0% | 79.1% | 0.871 | 0.583 |