Hybrid Deep Feature Fusion of 2D CNN and 3D CNN for Vestibule Segmentation from CT Images

Table 1

Parameter setting of each module of 3D DenseUNet.

Block	Feature size	Convolution layer
Input		—
Convolution 1		conv
Pooling		max pooling
Dense block 1
Transition layer 1		conv; average pooling
Dense block 2
Transition layer 2		conv; average pooling
Dense block 3		(
Transition layer 3		conv; average pooling
Dense block 4
Upsampling layer 1		upconv
Sum with dense block 3		—
Upsampling layer 2		upconv
Sum with dense block 2		—
Upsampling layer 3		upconv
Sum with dense block 1		—
Upsampling layer 4		upconv
Sum with convolution 1		—
Upsampling layer 5		upconv
Output		conv