Table 14 Diagnostic performance across modern architectures with statistical measures.
Architecture | Without enhancement (mean ± SD) | 95% CI | With DentoSMART-LDM (mean ± SD) | 95% CI | absolute gap | Relative improvement | Effect size (Cohen’s d) | t-test p-value |
|---|---|---|---|---|---|---|---|---|
EfficientNet-B5 | 82.1 ± 1.2% | [81.3, 82.9] | 97.3 ± 0.6% | [96.9, 97.7] | 15.2% | 18.5% | 3.24 | p < 0.001 |
ResNet-152 | 79.7 ± 1.4% | [78.8, 80.6] | 96.4 ± 0.7% | [95.9, 96.9] | 16.7% | 21.0% | 3.41 | p < 0.001 |
Vision transformer | 83.4 ± 1.1% | [82.7, 84.1] | 97.8 ± 0.5% | [97.5, 98.1] | 14.4% | 17.3% | 3.18 | p < 0.001 |
ConvNeXt-Large | 84.2 ± 1.0% | [83.6, 84.8] | 97.1 ± 0.6% | [96.7, 97.5] | 12.9% | 15.3% | 2.97 | p < 0.001 |
Swin transformer | 85.1 ± 0.9% | [84.6, 85.6] | 98.2 ± 0.4% | [97.9, 98.5] | 13.1% | 15.4% | 3.05 | p < 0.001 |
DenseNet-201 | 80.8 ± 1.3% | [80.0, 81.6] | 95.9 ± 0.8% | [95.4, 96.4] | 15.1% | 18.7% | 3.29 | p < 0.001 |
RegNet-Y | 81.5 ± 1.2% | [80.7, 82.3] | 96.7 ± 0.7% | [96.2, 97.2] | 15.2% | 18.7% | 3.26 | p < 0.001 |
EfficientNetV2 | 84.7 ± 1.0% | [84.1, 85.3] | 97.5 ± 0.5% | [97.2, 97.8] | 12.8% | 15.1% | 2.94 | p < 0.001 |
Average | 82.7 ± 1.1% | [82.2, 83.2] | 97.1 ± 0.6% | [96.8, 97.4] | 14.4% | 17.5% | 3.17 | p < 0.001 |
