Table 3.
Performance of two radiologists and the deep learning model on validation cohorts.
| Validation cohort | Reader | Accuracy(95%CI) | Sensitivity(95%CI) | Specificity(95%CI) |
|---|---|---|---|---|
| Internal | Reader 1 | 0.685 (0.564, 0.786) |
0.933 (0.660, 0,.997) |
0.621 (0.483, 0.742) |
| Reader 2 | 0.585 (0.454, 0.688) |
0.800 (0.514, 0.947) |
0.517 (0.383, 0.649) |
|
| Model | 0.795 (0.681, 0.877) |
0.733 (0.448, 0.911) |
0.810 (0.682, 0.897) |
|
| Reader 1 | 0.747 (0.631, 0.837) |
0.774 (0.585, 0.897) |
0.727 (0.570, 0.845) |
|
| External | Reader 2 | 0.573 (0.454, 0.685) |
0.839 (0.655, 0.939) |
0.386 (0.247, 0.545) |
| Model | 0.747 (0.631, 0.837) |
0.710 (0.518, 0.851) |
0.773 (0.618, 0.880) |
AUC, area under the receiver operating characteristics curve; CI, confidence interval.