Table 2 Boundary position errors (in pixels) for each of the semantic segmentation methods with comparison to the baseline.
From: Automatic choroidal segmentation in OCT images using supervised deep learning methods
Method | ILM | RPE | CSI | |||
|---|---|---|---|---|---|---|
ME | MAE | ME | MAE | ME | MAE | |
Standard | −0.07 (0.22) | 0.51 (0.10) | −0.10 (0.19) | 0.45 (0.12) | 0.85 (2.45) | 2.86 (2.01) |
Standard [CE] | 0.07 (0.25) | 0.51 (0.11) | −0.03 (0.20) | 0.45 (0.11) | 0.19 (2.24) | 2.58 (1.65) |
Residuals | 0.05 (0.23) | 0.50 (0.11) | −0.14 (0.19) | 0.45 (0.11) | 0.02 (2.31) | 2.59 (1.58) |
Residuals [CE] | −0.21 (0.24) | 0.53 (0.09) | −0.09 (0.29) | 0.46 (0.22) | 0.68 (2.05) | 2.53 (1.52) |
RNN | −0.25 (0.22) | 0.55 (0.09) | −0.27 (0.19) | 0.49 (0.13) | 1.05 (2.35) | 2.56 (1.89) |
RNN [CE] | −0.11 (0.24) | 0.51 (0.09) | −0.25 (0.19) | 0.48 (0.12) | 0.42 (2.42) | 2.59 (1.92) |
cSE | 0.02 (0.68) | 0.54 (0.63) | −0.08 (0.18) | 0.44 (0.11) | 0.59 (2.53) | 2.73 (1.97) |
cSE [CE] | −0.15 (0.28) | 0.52 (0.16) | −0.21 (0.21) | 0.47 (0.12) | 0.41 (2.10) | 2.57 (1.57) |
sSE | −0.03 (0.22) | 0.51 (0.08) | −0.16 (0.18) | 0.46 (0.12) | 0.02 (2.61) | 2.84 (1.83) |
sSE [CE] | −0.08 (0.36) | 0.52 (0.26) | −0.27 (0.21) | 0.50 (0.13) | 0.81 (2.33) | 2.72 (1.84) |
scSE | −0.04 (0.33) | 0.53 (0.24) | −0.16 (0.20) | 0.46 (0.12) | 0.34 (2.30) | 2.60 (1.68) |
scSE [CE] | −0.16 (0.23) | 0.52 (0.09) | 0.13 (0.20) | 0.46 (0.11) | 0.19 (2.20) | 2.60 (1.51) |
Combined | 0.06 (0.22) | 0.50 (0.09) | −0.07 (0.20) | 0.44 (0.12) | 1.10 (2.71) | 2.69 (2.14) |
Combined [CE] | −0.03 (0.24) | 0.51 (0.09) | −0.19 (0.20) | 0.47 (0.12) | 1.25 (2.06) | 2.53 (1.52) |
Baseline37 | −0.27 (0.41) | 0.58 (0.36) | −1.14 (0.65) | 1.23 (0.60) | −3.64 (8.62) | 5.82 (7.77) |
