Table 3.

Comparison of Segmentation of Layer Boundary Lines by the CNN Models to Human Graders

		Central 6 mm of B-Scan, Pixels			Full B-Scan Width (9 mm), Pixels
Bland–Altman Analysis		Mean Difference	SE	CoR (1.96 * SD)	Mean Difference	SE	CoR (1.96 * SD)
U-Net vs. graders	ILM	0.311	0.012	8.107	0.499	0.013	9.827
	dINL	−0.020	0.010	6.545	−0.161	0.008	6.184
	EZ	−0.130	0.006	2.891	−0.131	0.006	2.870
	pRPE	0.433	0.003	2.058	0.444	0.003	2.150
	BM	0.372	0.003	1.786	0.343	0.003	1.937
SW vs. graders	ILM	0.102	0.003	2.241	0.136	0.006	4.777
	dINL	−0.378	0.010	6.720	−0.451	0.008	6.146
	EZ	−0.578	0.007	3.327	−0.599	0.007	3.333
	pRPE	0.369	0.003	1.938	0.365	0.003	1.948
	BM	0.226	0.003	1.721	0.278	0.003	2.591
Hybrid model vs. graders	ILM	0.135	0.003	1.663	0.146	0.003	2.260
	dINL	0.030	0.009	6.168	−0.096	0.008	5.763
	EZ	−0.143	0.006	2.629	−0.152	0.006	2.631
	pRPE	0.451	0.003	1.881	0.469	0.003	1.954
	BM	0.392	0.002	1.611	0.374	0.002	1.748
Grader 2 vs. grader 1	ILM	−0.021	0.002	1.106	−0.019	0.002	1.300
	dINL	−0.427	0.010	6.266	−0.379	0.008	5.746
	EZ	0.213	0.006	2.861	0.201	0.006	2.896
	pRPE	−0.769	0.004	2.476	−0.747	0.003	2.491
	BM	−0.408	0.003	1.992	−0.389	0.003	2.052