Table 4:
Results of cross-dataset validation as measured by average AUC on per-patient evaluation of models. Evaluation on the same dataset is done via Leave One Patient Out cross-validation. Uncertainties given are the SD over patients.
Model | Trained On | Duke dataset | Helsinki dataset |
---|---|---|---|
DL2 (Class balance) | Duke dataset | 0.970 (0.033) | 0.925 (0.099) |
Pre-trained SVM (SVMt) | Helsinki dataset | 0.826 (0.117) | 0.923 (IQR 0.869–0.990)2 |