Table 4.
A comparison of performances of our ensemble model and various deep learning models.
| Model | Methods | Specificity | Recall | F-score | Accuracy |
|---|---|---|---|---|---|
| Our ensemble method (InceptionV3+MobileNetV2) | 98.97 ± 0.22 | 96.89 ± 0.65 | 96.90 ± 0.65 | 98.45 ± 0.32 | |
| Ensemble method [66] (Vgg16+Densenet201) | • 91.35 ± 0.42 | • 74.05 ± 0.96 | • 65.34 ± 1.10 | • 87.04 ± 0.57 | |
| AlexNet | Trained on ODS-NPTW | 76.15 ± 0.37 | 28.48 ± 1.46 | 24.18 ± 1.82 | 64.36 ± 0.64 |
| FT on ODS-NPTW | 92.69 ± 0.70 | 78.11 ± 2.20 | 77.77 ± 2.17 | 89.03 ± 1.03 | |
| FT on ODS-PTW | 97.41 ± 0.43 | 92.21 ± 1.25 | 92.20 ± 1.33 | 96.13 ± 0.62 | |
| FT on ADS-ALUF | • 97.67 ± 0.46 | • 93.05 ± 1.08 | • 92.96 ± 1.13 | • 96.50 ± 0.64 | |
| SqueezeNet | Trained on ODS-NPTW | 75.28 ± 0.31 | 25.85 ± 0.95 | 14.00 ± 1.32 | 62.91 ± 0.69 |
| FT on ODS-NPTW | 92.46 ± 0.49 | 77.30 ± 1.13 | 77.35 ± 1.15 | 88.69 ± 0.65 | |
| FT on ODS-PTW | 97.24 ± 0.38 | 91.77 ± 0.87 | 91.78 ± 0.95 | 95.87 ± 0.54 | |
| FT on ADS-ALUF | • 97.73 ± 0.25 | • 93.21 ± 0.47 | • 93.19 ± 0.47 | • 96.60 ± 0.33 | |
| Densenet201 | Trained on ODS-NPTW | 74.64 ± 1.43 | 23.73 ± 4.03 | 22.20 ± 4.68 | 61.99 ± 2.14 |
| FT on ODS-NPTW | 87.46 ± 0.74 | 62.36 ± 2.49 | 62.55 ± 2.33 | 81.21 ± 1.01 | |
| FT on ODS-PTW | 97.36 ± 0.43 | 92.08 ± 1.35 | 92.04 ± 1.41 | 96.05 ± 0.65 | |
| FT on ADS-ALUF | • 97.91 ± 0.38 | • 93.67 ± 1.35 | • 93.67 ± 1.33 | • 96.86 ± 0.32 | |
| MobileNetV2 | Trained on ODS-NPTW | 74.48 ± 0.57 | 23.46 ± 1.73 | 14.29 ± 0.85 | 61.74 ± 1.06 |
| FT on ODS-NPTW | 92.33 ± 0.54 | 77.04 ± 1.06 | 76.68 ± 1.13 | 88.50 ± 0.74 | |
| FT on ODS-PTW | 97.91 ± 0.21 | 93.76 ± 0.63 | 93.71 ± 0.66 | 96.86 ± 0.32 | |
| FT on ADS-ALUF | • 98.24 ± 0.27 | • 94.71 ± 1.01 | • 94.68 ± 1.01 | • 97.36 ± 0.42 | |
| InceptionV3 | Trained on ODS-NPTW | 72.16 ± 0.58 | 16.57 ± 1.33 | 16.31 ± 1.00 | 58.24 ± 0.59 |
| FT on ODS-NPTW | 85.01 ± 0.41 | 55.02 ± 1.09 | 55.29 ± 1.17 | 77.53 ± 0.52 | |
| FT on ODS-PTW | 98.08 ± 0.45 | 94.19 ± 1.63 | 94.21 ± 1.60 | 97.11 ± 0.70 | |
| FT on ADS-ALUF | • 98.49 ± 0.38 | • 95.42 ± 1.31 | • 95.43 ± 1.28 | • 97.74 ± 0.58 |
ODS, ADS, NPTW, PTW, ALUF, FT stands for original dataset, augmented dataset, no pre-trained weights, pre-trained weights, all layers un-frozen, fine-tuned. A • denotes that our deep learning ensemble model is statistically better than its competing model.