Table 4.
Receiver operating characteristic curve (ROC) analyses of 7 deep learning radiomics (DLR) models for training and validation sets.
| Model | Accuracy | AUC (95% CI) | Sensitivity | Specificity | PPV | NPV |
|---|---|---|---|---|---|---|
| Training dataset | ||||||
| SVM | 0.829 | 0.89 (0.8407-0.9399) | 0.866 | 0.804 | 0.753 | 0.897 |
| ExtraTrees | 0.787 | 0.892 (0.8458-0.9388) | 0.851 | 0.742 | 0.695 | 0.878 |
| XGBoost | 0.823 | 0.913 (0.8721-0.9544) | 0.821 | 0.825 | 0.764 | 0.87 |
| Light GBM | 0.805 | 0.878 (0.8271-0.9279) | 0.866 | 0.763 | 0.716 | 0.892 |
| MLP | 0.799 | 0.876 (0.8253-0.9276) | 0.821 | 0.784 | 0.724 | 0.864 |
| LR | 0.823 | 0.90 (0.8548-0.9454) | 0.881 | 0.784 | 0.737 | 0.905 |
| RF | 0.841 | 0.894 (0.8453-0.9419) | 0.881 | 0.814 | 0.766 | 0.908 |
| Validation dataset | ||||||
| SVM | 0.704 | 0.776 (0.6659-0.8863) | 0.778 | 0.659 | 0.583 | 0.829 |
| ExtraTrees | 0.718 | 0.829 (0.7347-0.9235) | 0.926 | 0.591 | 0.581 | 0.929 |
| XGBoost | 0.746 | 0.832 (0.7351-0.9290) | 0.815 | 0.705 | 0.629 | 0.861 |
| Light GBM | 0.789 | 0.815 (0.7120-0.9184) | 0.704 | 0.841 | 0.731 | 0.822 |
| MLP | 0.789 | 0.859 (0.7743-0.9446) | 0.741 | 0.818 | 0.714 | 0.837 |
| LR | 0.69 | 0.783 (0.6729-0.8927) | 0.778 | 0.636 | 0.568 | 0.824 |
| RF | 0.69 | 0.79 (0.6846-0.8945) | 0.852 | 0.591 | 0.561 | 0.867 |
AUC, area under the curve; SVM, support vector machine; GBM, gradient boosting machine; MLP, multilayer perceptron; LR, logistic regression; RF, random forest; PPV, positive predictive value; NPV, negative predictive value.