Interpretable model based on multisequence magnetic resonance imaging radiomics for predicting the pathological grades of hepatocellular carcinomas

Table 5 Predictive performance of each model

Group	Model	AUC value (95%CI)	Accuracy	Sensitivity	Specificity	Positive predictive value	Negative predictive value
Training group	FS-T2WI	0.761 (0.562-0.857)	0.862	0.214	0.986	0.750	0.867
	AP	0.870 (0.714-0.918)	0.897	0.500	0.973	0.778	0.910
	PVP	0.868 (0.714-0.959)	0.908	0.500	0.986	0.875	0.911
	Radiomics	0.917 (0.857-0.959)	0.943	0.714	0.986	0.909	0.947
	Clinical	0.869 (0.643-0.973)	0.885	0.429	0.973	0.750	0.899
	Combined	0.941 (0.857-0.945)	0.942	0.714	0.986	0.909	0.947
Validation group	FS-T2WI	0.724 (0.625-0.833)	0.816	0.333	0.906	0.400	0.879
	AP	0.802 (0.686-1.000)	0.763	0.333	0.844	0.286	0.871
	PVP	0.797 (0.688-1.000)	0.763	0.167	0.875	0.200	0.848
	Radiomics	0.901 (0.833-0.906)	0.895	0.667	0.938	0.667	0.938
	Clinical	0.865 (0.594-1.000)	0.895	0.500	0.969	0.750	0.912
	Combined	0.932 (0.812-1.000)	0.895	0.667	0.938	0.667	0.938

CI: Confidence interval; AUC: Area under the receiver operating characteristic curve; FS-T2WI: Fat-suppressed T2-weighted image; AP: Arterial phase; PVP: Portal venous phase.