Table 3.
Performance of different PFS prediction models in the validation group.
| Models | CNNTL-EGFR | CNNTL-BM | CNNScratch |
|---|---|---|---|
| AUC | 0.744 | 0.701 | 0.668 |
| 95% CI | 0.645 to 0.843 | 0.598 to 0.805 | 0.559 to 0.776 |
| Threshold | 0.449 | 0.379 | 0.490 |
| Accuracy | 72.0% | 68.0% | 68.0% |
| Sensitivity | 75.0% | 77.3% | 54.5% |
| Specificity | 69.6% | 60.7% | 78.6% |
| Models | CNNTL-EGFR and Clinical | CNNTL-BM and Clinical | CNNScratch and Clinical |
| AUC | 0.771 | 0.756 | 0.715 |
| 95% CI | 0.676 to 0.866 | 0.659 to 0.854 | 0.614 to 0.816 |
| Threshold | 0.575 | 0.615 | 0.496 |
| Accuracy | 74.0% | 75.0% | 70.0% |
| Sensitivity | 56.8% | 52.3% | 56.8% |
| Specificity | 87.5% | 92.9% | 80.4% |
CNN, convolutional neural network; AUC, area under receiver operating characteristic curve; threshold, threshold at the optimal decision point; CI, confidence interval.