Table 3 Classification metrics of endoscopists versus CRCNet.
From: Diagnostic evaluation of a deep learning model for optical diagnosis of colorectal cancer
The performance of endoscopists and CRCNet | ||||||
|---|---|---|---|---|---|---|
Tianjin Cancer Hospital (n = 363) | Tianjin First Central Hospital (n = 290) | Tianjin General Hospital (n = 271) | ||||
Performance metrics | Endoscopista | CRCNet | Endoscopista | CRCNet | Endoscopista | CRCNet |
Accuracy (95% CI) | 0.824 (0.781–0.861) | 0.873 (0.835–0.906) | 0.928 (0.891–0.955) | 0.903 (0.863–0.935) | 0.934 (0.897–0.960) | 0.963 (0.933–0.982) |
Recall rate (95% CI) | 0.849 (0.781–0.903) | 0.904 (0.844–0.947) | 0.867 (0.779–0.929) | 0.933 (0.861–0.975) | 0.900 (0.805–0.959) | 0.914 (0.823–0.968) |
Specificity (95% CI) | 0.912 (0.867–0.946) | 0.853 (0.798–0.897) | 0.920 (0.873–0.954) | 0.890 (0.838–0.930) | 0.940 (0.898–0.969) | 0.980 (0.950–0.995) |
Precision (95% CI) | 0.842 (0.764–0.902) | 0.805 (0.736–0.863) | 0.838 (0.751–0.905) | 0.792 (0.703–0.865) | 0.844 (0.744–0.917) | 0.941 (0.856–0.984) |
Negative predicted value (95% CI) | 0.880 (0.825–0.924) | 0.930 (0.885–0.961) | 0.941 (0.899–0.969) | 0.967 (0.930–0.988) | 0.964 (0.928–0.986) | 0.970 (0.937–0.989) |
Kappab | 0.622 | 0.742 | 0.82 | 0.785 | 0.828 | 0.903 |
F1c | 0.768 | 0.852 | 0.878 | 0.857 | 0.873 | 0.928 |
