Table 3.
Ablation study on ILN-GNet for PD Detection using Dataset 2.
| Metrics | ILN-GNet | Model with conventional LinkNet | Model with conventional LinkNet and Ghostnet | Model with conventional PHOG | Model with conventional WF | Model without Feature Extraction | Model without MDSCM | Model without WAP-BN |
|---|---|---|---|---|---|---|---|---|
| Accuracy | 95.20% | 92.30% | 94.40% | 89.10% | 85.10% | 91.10% | 88.10% | 89.60% |
| Sensitivity | 97.40% | 94.80% | 94.80% | 89.60% | 87.80% | 90.40% | 89.10% | 89.80% |
| Specificity | 93.20% | 90.20% | 94.00% | 88.70% | 82.70% | 91.70% | 87.20% | 89.50% |
| Precision | 92.60% | 89.30% | 93.20% | 87.30% | 81.50% | 90.40% | 85.90% | 88.20% |
| F-measure | 94.90% | 92.00% | 94.00% | 88.40% | 84.50% | 90.40% | 87.50% | 89.00% |
| MCC | 90.40% | 84.80% | 88.70% | 78.20% | 70.40% | 82.20% | 76.30% | 79.20% |
| NPV | 97.60% | 95.20% | 95.40% | 90.80% | 88.70% | 91.70% | 90.20% | 91.00% |
| FPR | 6.80% | 9.80% | 6.00% | 11.30% | 17.30% | 8.30% | 12.80% | 10.50% |
| FNR | 2.60% | 5.20% | 5.20% | 10.40% | 12.20% | 9.60% | 10.90% | 10.20% |
| FDR | 7.40% | 10.70% | 6.80% | 12.70% | 18.60% | 9.60% | 14.10% | 11.80% |