TABLE 3.

The performance of “Function,” ”Diffusion,” ”Volumetry,” “Iron,” ”Multimodal,” and “clinical‐multimodal” model in the differential diagnosis

Model	Training dataset						Testing dataset
	AUC (95% CI)	B‐ACC	Sen	Spec	PPV	NPV	AUC (95% CI)	B‐ACC	Sen	Spec	PPV	NPV
Function	0.890 (0.815–0.942)	0.832*	0.833	0.830	0.830	0.833	0.806 (0.661–0.909)	0.845*	0.864	0.826	0.826	0.864
Diffusion	0.713 (0.618–0.797)	0.702*	0.736	0.667	0.684	0.720	0.626 (0.470–0.766)	0.644*	0.636	0.652	0.636	0.652
Volumetry	0.544 (0.445–0.641)	0. 570*	0.509	0.630	0.574	0.567	0.514 (0.360–0.665)	0.579*	0.636	0.522	0.560	0.600
Iron	0.800 (0.712–0.871)	0.740*	0.774	0.704	0.719	0.760	0.741 (0.589–0.860)	0.801*	0.818	0.783	0.783	0.818
Multimodal	0.968 (0.914–0.992)	0.917*	0.981	0.852	0.867	0.979	0.927 (0.809–0.983)	0.911*	0.909	0.913	0.909	0.913
Clinical‐Multimodal	0.986 (0.944–0.999)	0.963*	0.981	0.944	0.945	0.981	0.953 (0.885–0.994)	0.934*	0.955	0.913	0.913	0.955

Abbreviations: AUC, area under the receiver operator curve; B‐ACC, balanced accuracy; Sen, sensitivity; Spec, specificity; PPV, positive predict value; NPV, negative predict value.

^* p < 0.001 under permutation test (1000 times).