Table 6.

Dice score and Mean Absolute Error (with Standard Deviation) results of 4 class pixel segmentation with various neural network models after removing $3\%$ of anomalous data samples and expanding data set with 4 augmentation techniques. The best results are highlighted in bold.

Model	Average Dice Score (SD)				MAE (SD) [px]
	Vitreous	Preretinal Space	Retina	Region b. Retina	PCV	ILM	RPE
UNet	0.9968 (0.008)	0.9591 (0.074)	0.9944 (0.002)	0.9984 (0.002)	1.33 (2.88)	0.50 (0.28)	0.57 (0.29)
LFUNet	0.9973 (0.006)	0.9590 (0.072)	0.9942 (0.002)	0.9985 (0.001)	1.50 (3.53)	0.50 (0.12)	0.57 (0.24)
Attention UNet	0.9850 (0.030)	0.8802 (0.186)	0.9926 (0.004)	0.9956 (0.010)	4.02 (6.54)	0.60 (0.26)	0.80 (0.73)
ReLayNet	0.9603 (0.054)	0.8015 (0.240)	0.9918 (0.004)	0.9859 (0.025)	12.63 (13.84)	0.71 (0.59)	0.96 (0.88)