Table 3. The results of model analysis.

Models	Training cohort					Validation cohort				Cut-off values
	AUC	ACC	SEN	SPE		AUC	ACC	SEN	SPE
Clinical	0.893	0.848	0.804	0.872		0.858	0.772	0.700	0.811	0.540
Liver imaging	0.623	0.614	0.565	0.640		0.671	0.667	0.650	0.676	0.665
Radiomics	0.800	0.742	0.739	0.744		0.789	0.754	0.700	0.784	0.590
RC	0.931	0.894	0.848	0.919		0.911	0.842	0.700	0.919	0.546
RL	0.852	0.818	0.804	0.826		0.841	0.789	0.750	0.811	0.572
LC	0.904	0.879	0.739	0.953		0.896	0.789	0.650	0.865	0.409
RCL	0.942	0.886	0.870	0.895		0.942	0.877	0.850	0.892	0.661

Lesser than cut-off value means combined hepatocellular and cholangiocarcinoma; ROC analysis demonstrated that the AUC of RCL model was statistically significant compared to the other models except for the RC model. RC, radiomics-clinical; RL, radiomics-liver imaging; LC, liver imaging-clinical; RCL, radiomics-clinical-liver imaging; AUC, area under the curve; ACC, accuracy; SEN, sensitivity; SPE, specificity.