Table 1.

Comparison of the performance of five deep learning models.

Model Name	Acc	AUC	95% Cl	SENS	SPE	PPV	NPV	Precision	Recall	F1	Threshold	Cohort
Densenet121	0.839	0.912	0.877–0.948	0.805	0.873	0.864	0.817	0.864	0.805	0.833	0.519	TRAIN
	0.722	0.768	0.661–0.874	0.821	0.667	0.575	0.872	0.575	0.821	0.676	0.313	Val
	0.65	0.637	0.460–0.813	0.737	0.571	0.609	0.706	0.609	0.737	0.667	0.313	TEST
Resnet18	0.919	0.973	0.956–0.990	0.924	0.915	0.916	0.923	0.916	0.924	0.92	0.494	TRAIN
	0.873	0.886	0.801–0.972	0.714	0.961	0.909	0.86	0.909	0.714	0.8	0.513	Val
	0.85	0.876	0.766–0.986	0.737	0.952	0.933	0.8	0.933	0.737	0.824	0.513	TEST
SimpleViT	0.513	0.443	0.370–0.517	0.178	0.855	0.538	0.508	0.538	0.178	0.268	0.56	TRAIN
	0.62	0.591	0.463–0.719	0.786	0.529	0.478	0.818	0.478	0.786	0.595	0.36	Val
	0.65	0.497	0.305–0.691	0.842	0.5	0.593	0.769	0.593	0.842	0.696	0.36	TEST
Resnet50	0.924	0.907	0.861–0.954	0.907	0.941	0.939	0.91	0.939	0.907	0.922	0.501	TRAIN
	0.65	0.622	0.435–0.808	0.842	0.476	0.593	0.769	0.593	0.842	0.696	0.379	VAL
	0.722	0.661	0.533–0.789	0.321	0.941	0.75	0.716	0.75	0.321	0.45	0.481	TEST
Resnet101	0.919	0.973	0.957–0.990	0.915	0.924	0.923	0.916	0.923	0.915	0.919	0.498	TRAIN
	0.675	0.638	0.456–0.820	0.842	0.524	0.615	0.786	0.615	0.842	0.711	0.142	VAL
	0.595	0.587	0.460–0.714	0.857	0.451	0.462	0.852	0.462	0.857	0.6	0.128	TEST