Table 2.
Detailed aortic dissection (AD) detection performance results of the models.
|
AUC (95% CI) |
Accuracy (95% CI) |
Sensitivity (95% CI) |
Specificity (95% CI) |
|
|---|---|---|---|---|
| The 3D DL model | ||||
| Training cohort | 0.997 (0.993–1) |
0.958 (0.922–0.979) |
0.927 (0.851–0.968) |
0.979 (0.935–0.995) |
| Cross validation result | 0.84 (0.7–0.98) |
0.857 (0.69–0.946) |
0.857 (0.562–0.975) |
0.857 (0.626–0.962) |
| Internal testing cohort | 0.803 (0.688–0.917) |
0.794 (0.675–0.879) |
0.655 (0.457–0.814) |
0.897 (0.748–0.967) |
| External testing cohort | 0.814 (0.733–0.895) |
0.757 (0.664–0.831) |
0.891 (0.756–0.959) |
0.662 (0.533–0.771) |
| The deep-MD model | ||||
| Training cohort | 0.961 (0.938–0.985) |
0.919 (0.879–0.948) |
0.936 (0.869–0.972) |
0.908 (0.85–0.946) |
| Cross validation result | 0.953 (0.923–0.983) |
0.916 (0.875–0.945) |
0.936 (0.869–0.972) |
0.902 (0.843–0.941) |
| Internal testing cohort | 0.878* (0.788–0.968) |
0.794 (0.675–0.879) |
0.690 (0.49–0.84) |
0.872 (0.718–0.952) |
| External testing cohort | 0.828 (0.751–0.906) |
0.721 (0.626–0.8) |
0.957 (0.84–0.992) |
0.554 (0.426–0.675) |
| The deep-integrated model | ||||
| Training cohort | 0.962 (0.939–0.986) |
0.919 (0.879–0.948) |
0.9 (0.824–0.947) |
0.933 (0.879–0.964) |
| Cross validation result | 0.956 (0.929–0.983) |
0.919 (0.879–0.948) |
0.909 (0.835–0.953) |
0.926 (0.872–0.96) |
| Internal testing cohort | 0.948** (0.898–0.998) |
0.897 (0.793–0.954) |
0.862 (0.674–0.955) |
0.923 (0.78–0.98) |
| External testing cohort | 0.969*** (0.937–1) |
0.73 (0.636–0.808) |
0.978 (0.87–0.999) |
0.554 (0.426–0.675) |
The p-value was calculated by the DeLong test on the internal testing cohort and the external testing cohort.
*
p < 0.05.
**
p < 0.01.
***
p < 0.001.