. 2020 Nov 6;34(1):116–123. doi: 10.1007/s10278-020-00394-2

Table 2.

Candidate values for each hyperparameter

Hyperparameter	Values
Number of convolutional layers	− 4~ + 4
Number of max-pooling layers	− 3~ + 3
Number of fully connected layers	− 2~ + 1
Filter size	3~9
Number of filters	48~576
Number of units in fully connected layer	2048~6144
Minibatch size	4~20
Dropout	0.25~0.75
Initial learning rates	0.5e⁻⁴~1.5e⁻⁴
Momentum	0.8~0.95
L2 Regularization	1e⁻¹⁰~1e⁻²