Evaluation of Multimodal Algorithms for the Segmentation of Multiparametric MRI Prostate Images

Table 1

Optimal hyperparameter configurations for DenseVNet, ScaleNet, and HighRes3DNet.

Hyperparameters	DenseVNet	ScaleNet	HighRes3DNet
Activation function	ReLU	Leaky ReLU	Leaky ReLU
Optimizers	Adam	Nesterov momentum	Adam
Batch size	8	4	4
Interpolation	Linear	B-spline	B-spline
Learning rate	10^-3	10^-2	10^-2
Loss function	Dice	Dice no square	Dice plus cross-entropy
Whitening	True	True	False
Normalization	False	False	True
Regularization type	L1	L2	L2
Sample per volume	1	10	8
Volume padding size	16	24	16
Window sampling	Resize	Resize	Resize
Spatial window size	(56, 56, 56)	(40, 40, 48)	(96, 96, 48)
No. of iterations	2000	1000	1000