Skip to main content
. 2025 Oct 6;15:34776. doi: 10.1038/s41598-025-18570-1

Table 6.

Ablation study results showing the incremental impact of quantitative evaluation of model components on classification performance.

Model variant Accuracy (%) AUC (%) Sensitivity (%) Specificity (%)
Pre-trained DenseNet169 89.8 76.03 66.29 97.31
Enhanced ViT 93.1 83.91 69.4 98.44
Fusion without SDEB 96 91.5 88 99.1
Full Hybrid (EViT-Dens169 + SDEB) 97.1 95.17 90.83 99.29