Table 2 Diagnostic performance of radiomic model, clinical model, and three radiologists for predicting the differentiation grade of INMA.
AUC (95%CI) | Accuracy (95%CI) | Specificity (95%CI) | Sensitivity (95%CI) | |
|---|---|---|---|---|
Internal test cohort | ||||
The clinical model | 0.875 (0.811–0.938) | 0.766 (0.682–0.837) | 0.753 (0.650–0.838) | 0.800 (0.631–0.916) |
The intratumor radiomic model | 0.882 (0.807–0.957) | 0.839 (0.759–0.896) | 0.876 (0.786–0.934) | 0.743 (0.564–0.868) |
The combined_3mm radiomic model | 0.907 (0.844–0.969) | 0.855 (0.778–0.909) | 0.876 (0.786–0.934) | 0.800 (0.625–0.909) |
The combined_5mm radiomic model | 0.858 (0.783–0.933) | 0.790 (0.706–0.856) | 0.831 (0.734-0.900) | 0.686 (0.506–0.826) |
Without AI assistance | ||||
Junior radiologist 1 | 0.666 (0.558–0.773) | 0.669 (0.579–0.751) | 0.674 (0.567–0.770) | 0.657 (0.478–0.809) |
Junior radiologist 2 | 0.694 (0.588-0.800) | 0.710 (0.621–0.788) | 0.730 (0.626–0.819) | 0.657 (0.478–0.809) |
Senior radiologist 3 | 0.765 (0.671–0.859) | 0.750 (0.664–0.823) | 0.730 (0.626–0.819) | 0.800 (0.631–0.916) |
With AI assistance | ||||
Junior radiologist 1 | 0.821 (0.733–0.910)* | 0.831 (0.753–0.892) | 0.843 (0.750–0.911) | 0.800 (0.631–0.916) |
Junior radiologist 2 | 0.827 (0.739–0.915)* | 0.839 (0.762–0.899) | 0.854 (0.763–0.920) | 0.800 (0.631–0.916) |
Senior radiologist 3 | 0.850 (0.770–0.930)* | 0.847 (0.771–0.905) | 0.843 (0.750–0.911) | 0.857 (0.697–0.952) |
External test cohort | ||||
The clinical model | 0.760 (0.603–0.916) | 0.737 (0.569–0.866) | 0.692 (0.482–0.857) | 0.833 (0.516–0.979) |
The intratumor radiomic model | 0.760 (0.580–0.939) | 0.789 (0.622–0.899) | 0.923 (0.734–0.987) | 0.500 (0.223–0.777) |
The combined_3mm radiomic model | 0.772 (0.593–0.952) | 0.816 (0.651–0.917) | 0.961 (0.784–0.998) | 0.500 (0.223–0.777) |
The combined_5mm radiomic model | 0.766 (0.593–0.952) | 0.763 (0.594–0.880) | 0.885 (0.687–0.970) | 0.500 (0.223–0.777) |
