Table 5. Diagnostic performance of the BETNET model and the doctors with different experience levels in the internal test set.

Parameters	BETNET model	Experienced doctor A	Experienced doctor B	Experienced doctor C	Doctor D	Doctor E
Accuracy	91.33%	93.67%	94.33%	85.67%	77.67%	69.33%
Sensitivity	93.39%	93.39%	93.39%	81.94%	72.69%	60.35%
Specificity	84.93%	94.52%	97.26%	97.26%	93.15%	97.26%
AUC	0.951	0.940	0.953	0.896	0.833	0.788
95% CI	0.920–0.972	0.906–0.964	0.923–0.974	0.856–0.928	0.782–0.870	0.737–0.833
P1	–	0.5494	0.8676	0.0059*	0.000*	0.000*
P2	–	0.296	0.188	0.000*	0.000*	0.000*
κ value	0.769	0.836	0.855	0.670	0.521	0.409

AUCs of the BETNET model and doctors were calculated by DeLong et al.’s method. P1: the difference of AUCs between the BETNET model prediction and doctor was compared by Z-test; *, P<0.05. P2: Measures the agreement between the BETNET model prediction and doctors. McNemar’s test was used for the statistical analysis; *, P<0.05. κ value: Measures the agreement between the BETNET model prediction, five doctors, and the pathological result. AUC, area under the ROC curve; CI, confidence interval.