Table 2. The multi-class prediction scores for the Herlev and SIPaKMeD Cervical Cancer datasets under evaluation criteria, that is, Accuracy, Precision, Recall, F-Beta, and Kappa Score.
Multi-class classification Convolution Neural Network used the K-fold cross-validation techniques, which showed a substantial increase in scores. The EfficientNet-B3 in consultation with Transfer Learning and Progressive resizing generated Benchmark scores for Whole-Slide images of the SIPakMeD dataset.
| Model | K-Fold CV | Accuracy (%) | Precision (%) | Recall (%) | F-Beta (%) | Kappa Score (%) |
|---|---|---|---|---|---|---|
| Herlev | ||||||
| VGG19 | 5-Fold | 85.18 ± 6.90 | 88.22 ± 4.66 | 87.24 ± 7.00 | 87.20 ± 6.76 | 91.26 ± 5.40 |
| ResNet-34 | 5-Fold | 91.94 ± 8.73 | 93.36 ± 7.41 | 92.81 ± 8.43 | 92.86 ± 8.28 | 95.31 ± 5.05 |
| ResNet-101 | 5-Fold | 93.14 ± 8.78 | 94.56 ± 7.01 | 93.98 ± 8.08 | 94.05 ± 7.93 | 95.57 ± 5.55 |
| EfficientNetB3 | 5-Fold | 91.40 ± 10.25 | 92.74 ± 8.63 | 92.20 ± 9.36 | 92.19 ± 9.36 | 92.99 ± 8.04 |
| EfficientNetB4 | 5-Fold | 93.03 ± 8.95 | 94.31 ± 7.31 | 94.27 ± 7.62 | 94.25 ± 7.58 | 96.40 ± 4.51 |
| EfficientNetB5 | 5-Fold | 92.16 ± 9.25 | 93.19 ± 9.13 | 93.21 ± 8.23 | 93.12 ± 8.28 | 95.29 ± 4.63 |
| SIPaKMeD | ||||||
| VGG19 | 5-Fold | 98.65 ± 0.57 | 98.79 ± 0.64 | 98.44 ± 0.48 | 98.90 ± 0.48 | 99.09 ± 0.41 |
| ResNet-34 | 3-Fold | 96.56 ± 0.13 | 97.19 ± 0.29 | 97.27 ± 0.17 | 97.22 ± 0.14 | 98.19 ± 0.13 |
| ResNet-101 | 5-Fold | 98.55 ± 1.11 | 98.57 ± 1.26 | 98.70 ± 0.88 | 98.65 ± 0.97 | 98.07 ± 1.47 |
| EfficientNetB3 | 3-Fold | 97.81 ± 0.10 | 98.09 ± 0.10 | 98.23 ± 0.23 | 98.24 ± 0.21 | 98.58 ± 0.15 |
| EfficientNetB3 | 5-Fold | 99.27 ± 0.74 | 99.36 ± 0.69 | 99.43 ± 0.61 | 99.41 ± 0.63 | 99.54 ± 0.60 |
| EfficientNetB3 | 5-Fold | 96.39 ± 1.85 | 96.49 ± 1.42 | 97.02 ± 1.59 | 96.88 ± 1.56 | 97.37 ± 1.12 |
| EfficientNetB3 | 5-Fold | 99.70 ± 1.07 | 99.70 ± 1.03 | 99.72 ± 0.87 | 99.63 ± 0.88 | 99.31 ± 0.78 |