Table 9 State-of-the-art comparison.
Study | Model | Accuracy | Sensitivity | Specificity | F1-Score | Explainability Tools |
|---|---|---|---|---|---|---|
Marmolejo-Saucedo et al. [14] | CNN with numGrad-CAM | 97.11% | 95.58% | 96.81% | 96.12% | numGrad-CAM (IoU 90.11%) |
Ramy A. Zeineldin et al. [41] | CNN with NeuroXAI | 98.62% | NA | NA | NA | Grad-CAM, SHAP, LIME, Information Flow Maps |
Weina Jin et al. [42] | CNN with SmoothGrad | 88.00% | NA | NA | NA | SmoothGrad |
Srirupa Guha et al. [43] | Custom ResNet50 | 97.50% | NA | NA | NA | Saliency Map, SHAP, Occlusion, LIME, Grad-CAM |
Yibin Wang et al. [44] | GMGENet | 92.20% | NA | NA | NA | Grad-CAM, ECE Heatmaps |
Proposed Model | EfficientNet-B5 + ResNet-50 + CNN | 99.40% | 98.75% | 98.42% | 98.55% | Grad-CAM, SHAP, SmoothGrad, Guided Grad-CAM, LIME |
