Table 2.
Performance of deep convolutional networks (DCNNs) for radiograph orientation
| DCNN | Testing set | AUC | Sensitivity | Specificity | PPV | NPV | Accuracy (%) |
|---|---|---|---|---|---|---|---|
| General | NIH CXR14 | 1 | 1 | 0.99 | 1 | 0.99 | 99.6 |
| Shenzhen, China and Montgomery County, USA | 0.999 | N/A | 0.99 | N/A | N/A | 99.3 | |
| Peds NIH CXR14 | 0.999 | 0.99 | 0.99 | 1 | 0.99 | 99.4 | |
| Johns Hopkins Hospital (Baltimore, MD) | 0.985 | 0.98 | 0.96 | 0.96 | 0.98 | 96.0 | |
| Pediatrics | Peds NIH CXR14 | 0.997 | 0.98 | 0.98 | 0.98 | 0.98 | 98.0 |
AUC area under the receiver operating characteristic (ROC) curve, PPV positive predictive value, NPV negative predictive value. Positive refers to AP radiographs while negative refers to PA radiographs