Automated Ventricular System Segmentation in Paediatric Patients Treated for Hydrocephalus Using Deep Learning Methods

Table 2

Parameters of neural network architecture that were used consistently during training.

Network architecture		Training		Optimization
Encoder	ResNet34	Learning rate	1e-4	Optimizer	Adam
Image size	512 x 512	Number of epochs	4	Learning rate policy	1cycle
Self-attention	True	Batch size	10	Loss	Generalized dice loss
Precision	FP16	Weight decay	1e-7