Table 3.
Sensitivity and specificity of the CNNa models when matched to the average performance of the teledermatologists.
|
|
Specificity when matched to sensitivity, % (95% CI) | P value | Sensitivity when matched to specificity, % (95% CI) | P value |
| Teledermatologists (average) | 79.2 (74.82-82.91) | Reference | 82.9 (76.1-88.4) | Reference |
| CNN-Sb | 72 (67.4-76.3) | .053 | 74.7 (67.2-81.3) | .10 |
| CNN-S2c | 65.2 (60.4-69.8) | .03 | 71.5 (63.8-78.4) | .07 |
| CNN-NSd | 46.7 (41.8-51.7) | .01 | 56.3 (48.2-64.2) | .03 |
aCNN: convolutional neural network.
bCNN-S: CNN standardized.
cCNN-S2: CNN standardized number 2.
dCNN-NS: CNN nonstandardized.