Table 3.

Predictive performance of radiomics models in non-contrast-enhanced and different contrast-enhanced phases

Dataset	Metric	Non-contrast	Arterial	Venous	Delayed
Training	AUC	0.792	0.803	0.791	0.801
	Accuracy	0.739	0.757	0.752	0.743
	Sensitivity	0.741	0.667	0.667	0.704
	Specificity	0.738	0.770	0.764	0.749
Validation 1	AUC	0.822	0.814	0.807	0.834
	Accuracy	0.708	0.760	0.771	0.688
	Sensitivity	0.889	0.667	0.667	0.889
	Specificity	0.690	0.770	0.782	0.667
Validation 2	AUC	0.711	0.725	0.737	0.709
	Accuracy	0.667	0.696	0.696	0.609
	Sensitivity	0.667	0.500	0.500	0.500
	Specificity	0.667	0.737	0.737	0.632

Best AUC values in each validation dataset were marked with bold text. AUC, area under the receiver operating characteristic curve