Table 3.
Predictive performance of EGFR mutation status using different models compared with conventional PET parameters and clinical feature.
| Model/Parameters | Training set | 10-fold cross validation using SVM algorithm | ||||||
|---|---|---|---|---|---|---|---|---|
| AUC | Sensitivity (%) | Specificity (%) | Accuracy (%) | AUC | Sensitivity (%) | Specificity (%) | Accuracy (%) | |
| Combined Model | 0.866 | 82.60% | 81.20% | 81.80% | 0.827 | 73.74% | 76.07% | 75.29% |
| PET/CT Radiomics Model | 0.868 | 92.80% | 66.30% | 77.10% | 0.769 | 67.11% | 67.03% | 67.06% |
| CT Radiomics Model | 0.792 | 58.00% | 87.10% | 75.30% | 0.754 | 64.22% | 69.87% | 67.65% |
| PET Radiomics Model | 0.738 | 55.10% | 82.20% | 71.20% | 0.750 | 60.29% | 69.69% | 67.06% |
| Gender | 0.664 | 53.60% | 79.20% | 68.80% | / | / | / | / |
| SUVmax | 0.683 | 84.10% | 49.50% | 63.50% | / | / | / | / |
| TLG | 0.662 | 66.70% | 64.40% | 65.30% | / | / | / | / |
SVM, support vector machine.