Table 4.
Dataset summary and prediction performance of deep learning systems on HdH Dataset (training, development and test) and TCIA Dataset
| Dataset | #Patients | EGFR+ a | AUC (w/mixup, ensemble) (%) | AUC (w/mixup, vanilla) (%) | AUC (w/o mixup, ensemble) (%) | AUC (w/o mixup, vanilla) (%) |
|---|---|---|---|---|---|---|
| HdH training dataset | 348 | 185 (53.2%) | 76.7 | 76.0 | 71.0 | 70.1 |
| HdH development dataset | 116 | 61 (52.6%) | 74.1 | 74.6 | 69.2 | 70.4 |
| HdH test dataset | 115 | 62 (53.9%) | 75.8 | 76.8 | 67.9 | 67.9 |
| TCIA dataset | 37 | 9 (24.3%) | 75.0 | 68.3 | 70.6 | 71.4 |
a
Shown as the number of cases (percentage).