Table 3.

Performances of the integrative image-based RA and DL models.

Model	RA model			DL model
	Image-kinetic RA model	Image-molecular RA model	Image-kinetic-molecular RA model	Image-kinetic DL model	Image-molecular DL model	Image-kinetic-molecular DL model
AUROC	0.629	0.755	0.781	0.707	0.79	0.832
	(0.595, 0.663)	(0.708, 0.802)	(0.735, 0.828)	(0.654, 0.761)	(0.768, 0.812)	(0.816, 0.847)
Accuracy	0.619	0.695	0.731	0.661	0.752	0.772
	(0.571, 0.668)	(0.638, 0.753)	(0.678, 0.784)	(0.596, 0.725)	(0.715, 0.788)	(0.724, 0.821)
Sensitivity	0.647	0.778	0.795	0.692	0.797	0.781
	(0.559, 0.735)	(0.669, 0.887)	(0.703, 0.887)	(0.579, 0.806)	(0.723, 0.869)	(0.696, 0.867)
Specificity	0.611	0.671	0.712	0.65	0.739	0.769
	(0.537, 0.685)	(0.58, 0.762)	(0.634, 0.791)	(0.54, 0.761)	(0.681, 0.797)	(0.69, 0.849)
PPV	0.329	0.413	0.451	0.368	0.473	0.497
	(0.267, 0.391)	(0.333, 0.493)	(0.367, 0.536)	(0.318, 0.417)	(0.401, 0.546)	(0.408, 0.587)
NPV	0.855	0.911	0.922	0.88	0.925	0.924
	(0.816, 0.894)	(0.872, 0.951)	(0.888, 0.956)	(0.859, 0.902))	(0.897, 0.953)	(0.896, 0.953)
P *	<0.001	0.118	–	<0.001	<0.001	–
P #	–	–	–	<0.001	<0.001	<0.001

Note: Data in parentheses are 95% confidence intervals. RA, radiomics analysis; DL, deep learning; AUROC, area under the receiver operating characteristics curve; PPV, positive predictive value; NPV, negative predictive value.

*P value of the comparison inside the RA models and DL models, respectively.

# P value of the comparison between the RA models and DL models, respectively.