Table 2.
Experimental results for PVTAD framework in comparison with single and parallel ResNet-18 and ViT-Tiny models for AD vs. CN classification on ADNI dataset based on five-fold cross validation.
| Study | Modality | Model | ACC | SEN | SPE | PRE | F1 | AUC |
|---|---|---|---|---|---|---|---|---|
| [10] | sMRI (Coronal View) | ResNet-18 | 81.25 | - | - | - | - | - |
| [12] | sMRI (Coronal View) | ResNet-18 | 88.5 | 100 | - | 75.9 | 86.3 | - |
| ViT-Tiny | 95.3 | 94.4 | - | 90 | 93.2 | - | ||
| This paper | sMRI (WM Coronal View) | ResNet-18 | 71.06 | 62.91 | 78.7 | 73.7 | 67.61 | 72 |
| ViT-Tiny | 96.23 | 95.91 | 96.53 | 96.3 | 96.1 | 95 | ||
| Parallel ResNet-18 and ViT-Tiny | 96.87 | 96.62 | 97.19 | 96.92 | 96.76 | 96 | ||
| PVTAD model | 97.7 | 97.15 | 98.16 | 98.02 | 97.6 | 98 |
ACC: Accuracy; SEN: Sensitivity; SPE: Specificity; PRE: Precision; F1: F1-score; AUC: Area under the curve.