Results of Performance comparison between the bimodal model and the GAVR model.

TABLE 4 Results of Performance comparison between the bimodal model and the GAVR model.

ModelName	Acc^a	AUC^b	95%CI^c	Sensitivity	Specificity	PPV^d	NPV^e
Training set (n = 553)
GAVR	0.998	1.000	0.9998–1.0000	1.000	0.996	0.997	1.000
2DDL+3DDL	0.998	1.000	0.9999–1.0000	1.000	0.996	0.997	1.000
Rad^f+2DDL	0.993	1.000	0.9996–1.0000	0.990	0.996	0.997	0.989
Rad+3DDL	0.993	1.000	0.9984–1.0000	0.993	0.993	0.993	0.993
Internal test set(n = 237)
GAVR	0.924	0.970	0.9496–0.9866	0.929	0.919	0.912	0.934
2DDL+3DDL	0.869	0.925	0.8904–0.9564	0.920	0.823	0.824	0.919
Rad+2DDL	0.949	0.968	0.9423–0.9887	0.955	0.944	0.939	0.959
Rad+3DDL	0.945	0.965	0.9396–0.9861	0.955	0.936	0.930	0.959
External test set1 (n = 583)
GAVR	0.979	0.987	0.9652–0.9980	0.958	0.982	0.883	0.994
2DDL+3DDL	0.897	0.954	0.9278–0.9751	0.833	0.906	0.556	0.975
Rad+2DDL	0.962	0.973	0.9399–0.9945	0.887	0.972	0.818	0.984
Rad+3DDL	0.967	0.988	0.9799–0.9947	0.930	0.973	0.825	0.990
External test set2 (n = 231)
GAVR	0.921	0.979	0.9626–0.9922	0.876	0.980	0.983	0.860
2DDL+3DDL	0.870	0.927	0.8876–0.9595	0.840	0.910	0.924	0.813
Rad+2DDL	0.904	0.966	0.9424–0.9840	0.845	0.980	0.982	0.831
Rad+3DDL	0.869	0.960	0.9367–0.9795	0.798	0.960	0.963	0.787
Prospective set(n = 200)
GAVR	0.945	0.987	0.9733–0.9969	0.968	0.924	0.920	0.970
2DDL+3DDL	0.925	0.976	0.9542‐00.9930	0.874	0.971	0.965	0.895
Rad+2DDL	0.885	0.933	0.8917–0.9712	0.811	0.952	0.939	0.848
Rad+3DDL	0.895	0.959	0.9326–0.9808	0.821	0.962	0.951	0.856

Acc, Accuracy;

AUC, Area under curve;

CI, Confidence interval;

PPV, Positive predictive value;

NPV, Negative predictive value;

Rad, Radiomics.

John Wiley & Sons, Ltd.