Table 3.

Performance measurements generated by clinical model, DL models and radiomics models trained on different sequences in the internal test dataset.

Models	AUC (95% CI)	P-value	Accuracy	Specificity	Sensitivity
Clinical Variables	0.840 (0.774-0.973)	0.047*	0.905	0.914	0.857
Radiomics_T1WI	0.773 (0.686-0.930)	0.036*	0.833	0.882	0.625
Radiomics_T2WI	0.786 (0.686-0.930)	0.038*	0.833	0.906	0.600
Radiomics_FLAIR	0.823 (0.632-0.897)	0.004*	0.786	0.929	0.500
Radiomics_DWI	0.803 (0.686-0.930)	0.014*	0.833	0.906	0.600
Radiomics_Combined	0.889 (0.715-0.946)	0.029*	0.857	0.936	0.636
DL_T1WI	0.721 (0.659-0.914)	0.035*	0.810	0.838	0.600
DL_T2WI	0.747 (0.560-0.861)	0.032*	0.738	0.806	0.333
DL_FLAIR	0.771 (0.659-0.914)	0.014*	0.810	0.903	0.546
DL_DWI	0.805 (0.686-0.930)	0.011*	0.833	0.861	0.667
DL_Combined	0.845 (0.715-0.946)	0.019*	0.857	0.886	0.714
Fusion	0.963 (0.874-0.999)	Na	0.976	1.000	0.900

*P < 0.05. The paired Student t-test was used to compare the prediction performance of the prognosis in patients with anti-NMDAR encephalitis between the fusion model and all the other models (The fusion model is the reference).

AUC, area under the receiver operating characteristic curve; CI, confidence interval; DL, deep learning; anti-NMDAR, anti-N-methyl-D-aspartate receptor; Na, Not available.