Table 3. The evaluation metrics for the cubic kernel SVM classifier constructed with the fused DL features compared to SVM classifiers trained with each DL feature.
| CNN | Accuracy (std) | AUC (std) | Sensitivity (std) | Specificity (std) | Precision (std) | F1 score (std) | DOR (std) |
|---|---|---|---|---|---|---|---|
| AlexNet | 94.8% (0.001) | 0.99 (0) | 0.95 (0) | 0.948 (0.005) | 0.947 (0.005) | 0.949 (0.003) | 342.001 (30.593) |
| GoogleNet | 96.7% (0.003) | 0.99 (0) | 0.97 (0) | 0.963 (0.004) | 0.962 (0.005) | 0.966 (0.003) | 829.889 (113.608) |
| ShuffleNet | 96.3% (0.001) | 0.99 (0) | 0.97 (0) | 0.961 (0.001) | 0.96 (0.001) | 0.965 (0.001) | 776 (0) |
| ResNet-18 | 97.6% (0.003) | 1.00 (0) | 0.975 (0.006) | 0.975 (0.006) | 0.975 (0.006) | 0.975 (0.006) | 1,723.223 (714.441) |
| DL FUSION | 98.6% (0.001) | 1.00 (0) | 0.981 (0.001) | 0.99 (0) | 0.99 (0) | 0.986 (0) | 4,851 (0) |
Note:
Bold values indicate the highest results.