TABLE 2.

Diagnostic performances of the image-based deep learning models with the hard and soft voting strategies.

Voting strategy	Dataset	AUC (95% CI)	ACC	SEN	SPE	NPV	PPV
Hard voting	Training set	0.712 (0.625–0.797)	0.786	0.936	0.487	0.792	0.785
	Test set	0.689 (0.554–0.826)	0.717	0.821	0.5556	0.6667	0.742
Soft voting	Training set	0.705 (0.615–0.794)	0.778	0.923	0.487	0.760	0.783
	Test set	0.633 (0.497–0.767)	0.674	0.821	0.444	0.615	0.697

ACC, accuracy; AUC, area under the curve; CI, confidence interval; NPV, negative predictive value; PPV, positive predictive value; SEN, sensitivity; SPE, specificity.