Table 1:

Average performance of 3D U-Net augmentation methods and normalization across 5 repetitions

Flip AugmentationNo Normalization
None3 DirectionLeft-RightLeft-Right
Training
 Cross-entropy0.0461 (SD, 0.0017)0.0596 (SD, 0.0006)0.0502 (SD, 0.0008)0.03504 (SD, 0.0010)
 Dice score0.9380 (SD, 0.0016)0.9140 (SD, 0.0004)0.9315 (SD, 0.0012)0.9284 (SD, 0.0008)
Validation
 Cross-entropy0.0726 (SD, 0.0041)0.07472 (SD, 0.0050)0.06771 (SD, 0.0039)0.04062 (SD, 0.0023)
 Dice score0.8698 (SD, 0.0136)0.8518 (SD, 0.0163)0.8819 (SD, 0.0122)0.8419 (SD, 0.0279)