Table 3.

Summary of confusion matrix elements and performance metrics of both models tested by real OCT data (Figure 11) with two binarization schemes. bin-1 indicates simple average with threshold binarization; bin-2 represents counting repeated signal binarization, with results of 1, 2, and 3 repeated signals shown.

Model	TN	FP	FN	TP	A	P	R	F1
Model 1 bin-1	96.86%	3.14%	57.49%	42.51%	0.697	0.931	0.425	0.584
Model 2 bin-1	95.71%	4.29%	49.24%	50.76%	0.732	0.922	0.508	0.655
Model 1 bin-2, N rep. 1	81.85%	18.15%	26.01%	73.99%	0.779	0.803	0.740	0.770
Model 1 bin-2, N rep. 2	89.95%	10.05%	39.02%	60.98%	0.755	0.859	0.610	0.713
Model 1 bin-2, N rep. 3	94.30%	5.70%	46.68%	53.32%	0.738	0.903	0.533	0.671
Model 2 bin-2, N rep. 1	70.83%	29.17%	11.32%	88.68%	0.798	0.752	0.887	0.814
Model 2 bin-2, N rep. 2	81.16%	18.84%	23.44%	76.56%	0.789	0.804	0.766	0.784
Model 2 bin-2, N rep. 3	88.17%	11.83%	32.95%	67.05%	0.776	0.850	0.670	0.750