TABLE 2.

Comparison of quantitative indices of the clinical model, radiomics signature, deep learning model and fused model applied to the three cohorts

	Training cohort				Internal validation cohort				Independent test cohort
Methods	CM	RS	DL	FM	CM	RS	DL	FM	CM	RS	DL	FM
AUC	0.7044	0.8223	0.8510	0.8941	0.6264	0.7616	0.8073	0.8301	0.6626	0.7475	0.7513	0.8042
ACC	0.6383	0.7459	0.7735	0.8160	0.6264	0.7177	0.7500	0.7702	0.6150	0.6578	0.6631	0.7273
SENS	0.6157	0.7622	0.8535	0.8535	0.5724	0.6965	0.8137	0.8137	0.6698	0.7453	0.6226	0.7075
SPEC	0.6707	0.7225	0.6585	0.7622	0.5825	0.7476	0.6601	0.7087	0.5432	0.5432	0.7160	0.7531
PPV	0.7286	0.7978	0.7821	0.8375	0.6587	0.7953	0.7712	0.7973	0.6574	0.6810	0.7416	0.7895
NPV	0.5486	0.6790	0.7578	0.7837	0.4918	0.6363	0.7157	0.7300	0.5570	0.6197	0.5918	0.6630
F1 score	0.6036	0.7001	0.7047	0.7728	0.5333	0.6875	0.6868	0.7192	0.5499	0.5789	0.6480	0.7052
Significance level of DeLong test for models compared with FM

	CM	RS	DL
Training cohort	<0.0001	<0.0001	<0.0001
Internal validation cohort	<0.0001	0.0035	0.1083
Independent test cohort	0.0005	0.0295	0.0132

Abbreviations: AUC, area under curve; ACC, accuracy; CM, clinical model; DL, deep learning model; FM, fused model; NPV, negative predictive value; PPV, positive predictive value; RS, radiomics signature; SENS: sensitivity; SPEC, specificity.