Table 2 Classification metrics of CRCNet at the patient level.
From: Diagnostic evaluation of a deep learning model for optical diagnosis of colorectal cancer
The performance of CRCNet across three test sets | |||
|---|---|---|---|
Performance metrics | Tianjin Cancer Hospital (n = 363) | Tianjin First Central Hospital (n = 430) | Tianjin General Hospital (n = 1470) |
Accuracy (95% CI) | 0.873 (0.835–0.906) | 0.916 (0.886–0.941) | 0.980 (0.972–0.987) |
Recall rate (95% CI) | 0.904 (0.844–0.947) | 0.789 (0.690–0.868) | 0.746 (0.629–0.842) |
Specificity (95% CI) | 0.853 (0.798–0.897) | 0.950 (0.921–0.971) | 0.992 (0.986–0.996) |
Precision (95% CI) | 0.805 (0.736–0.863) | 0.807 (0.709–0.883) | 0.828 (0.713–0.911) |
Negative predicted value (95% CI) | 0.930 (0.885–0.961) | 0.944 (0.915–0.966) | 0.987 (0.980–0.992) |
Kappaa | 0.742 | 0.745 | 0.775 |
F1b | 0.852 | 0.798 | 0.785 |
