Table 2. Accuracy, sensitivity, specificity, and AUC values of the DLR models for recurrence prediction (56 patients from Hospital I and 18 patients from Hospital II).

Models	Hospital I (internal data set)						Hospital II (independent data set)
	ACC	SEN	SPC	AUC	P		ACC	SEN	SPC	AUC	P
DLR-A	0.71	0.90	0.67	0.80	0.003		0.61	0.55	0.66	0.77	0.058
DLR-V	0.73	0.60	0.76	0.58	0.429		0.44	0.22	0.67	0.48	0.895
DLR-A&V	0.71	0.80	0.70	0.72	0.034		0.61	0.44	0.78	0.64	0.310

The threshold of the predictive probability used to calculate ACC, SEN, and SPC was the highest Youden index of the cross-validation ROC curves for the internal data set. A P value indicates the significance level of the comparison between an AUC with that of a random case (AUC =0.5). AUC, area under the curve; DLR, deep learning radiomics; ACC, accuracy; SEN, sensitivity; SPC, specificity; A, arterial; V, venous; A&V, arterial & venous.