Table 1.

Mean absolute distance (MAD) and standard deviation (Std. Dev.) in μm evaluated on 20 manually delineated scans of 9 surfaces, comparing AURA toolkit [13], R-Net [9], ReLayNet [18], SP (shortest path on our surface map), and our proposed method. Depth resolution is 3.9 μm. Numbers in bold are the best in that row.

Boundary	MAD (Sth. Dev.)
	AURA	R-Net	ReLayNet	SP	Our’s
ILM	2.37 (0.36)	2.38 (0.36)	3.17 (0.61)	2.70 (0.39)	2.41 (0.40)
RNFL-GCL	3.09 (0.64)	3.10 (0.55)	3.75 (0.84)	3.38 (0.68)	2.96 (0.71)
IPL-INL	3.43 (0.53)	2.89 (0.42)	3.42 (0.45)	3.11 (0.34)	2.87 (0.46)
INL-OPL	3.25 (0.48)	3.15 (0.56)	3.65 (0.34)	3.58 (0.32)	3.19 (0.53)
OPL-ONL	2.96 (0.55)	2.76 (0.59)	3.28 (0.63)	3.07 (0.53)	2.72 (0.61)
ELM	2.69 (0.44)	2.65 (0.66)	3.04 (0.43)	2.86 (0.41)	2.65 (0.73)
IS-OS	2.07 (0.81)	2.10 (0.75)	2.73 (0.45)	2.45 (0.31)	2.01 (0.57)
OS-RPE	3.77 (0.94)	3.81 (1.17)	4.22 (1.48)	4.10 (1.42)	3.55 (1.02)
BM	2.89 (2.18)	3.71 (2.27)	3.09 (1.35)	3.23 (1.36)	3.10 (2.02)
Overall	2.95 (1.04)	2.95 (1.10)	3.37 (0.92)	3.16 (0.88)	2.83 (0.99)