Table 2.
Performance metrics for the top three machine learning algorithms for predicting whether patients survive longer than the group median in the EGFR, ALK, and KRAS mutation-positive groups using radiomic features only.
| Mutation | Classifier | Accuracy | AUC* | Sensitivity | Specificity |
|---|---|---|---|---|---|
| EGFR | Ada Boost Classifier | 84.30% | 0.905 | 86.00% | 82.00% |
| Bagging Classifier | 84.00% | 0.915 | 90.00% | 79.00% | |
| Gradient Boosting Classifier | 88.10% | 0.95 | 90.00% | 87.00% | |
| ALK | Gradient Boosting Classifier | 85.70% | 0.92 | 88.00% | 83.00% |
| Random Forest Classifier | 77.80% | 0.93 | 95.00% | 68.00% | |
| Extra Trees Classifier | 85.70% | 0.936 | 90.00% | 81.00% | |
| KRAS | Extra Trees Classifier | 78.70% | 0.913 | 84.00% | 75.00% |
| Gradient Boosting Classifier | 85.10% | 0.955 | 83.00% | 87.00% | |
| Ada Boost Classifier | 95.70% | 0.957 | 100.00% | 92.00% |
*
AUC, area under the receiver operating characteristic curve. EGFR, epidermal growth factor receptor; ALK, anaplastic lymphoma kinase; KRAS, Kirsten rat sarcoma viral oncogene homolog.