Table 2.

Result comparisons of the proposed model and other models. Params means the number of parameters. To compute p-values for different two-view models, the model proposed in paper [16] is used as the reference base.

Model	Params	CC	MLO	Accuracy (%)	Precision	Recall	Specificity	F1 Score	AUC	p-Value
Single-view DeiT-tiny	5.5 M	✓		69.6 ± 1.4	0.708 ± 0.035	0.662 ± 0.040	0.728 ± 0.058	0.683 ± 0.013	0.724 ± 0.013	<0.001
Single-view DeiT-tiny	5.5 M		✓	72.5 ± 2.8	0.728 ± 0.037	0.713 ± 0.027	0.737 ± 0.045	0.720 ± 0.028	0.769 ± 0.036	0.376
Two-view DeiT-tiny (proposed)	5.5 M	✓	✓	77.0 ± 1.2	0.797 ± 0.039	0.726 ± 0.063	0.814 ± 0.057	0.757 ± 0.022	0.814 ± 0.026	0.031
Two-view DeiT-small (proposed)	21.7 M	✓	✓	76.3 ± 2.8	0.799 ± 0.071	0.706 ± 0.033	0.818 ± 0.081	0.747 ± 0.018	0.818 ± 0.039	0.009
CNN feature concatenation [16]	44.7 M	✓	✓	73.9 ± 2.4	0.761 ± 0.05	0.696 ± 0.041	0.781 ± 0.071	0.725 ± 0.019	0.784 ± 0.016	-
View-wise CNN [17]	22.4 M	✓	✓	71.7 ± 2.1	0.735 ± 0.04	0.677 ± 0.051	0.756 ± 0.069	0.702 ± 0.021	0.759 ± 0.023	N/A