Table 1.

Detailed structure of each block of 3D U-Net model, where $N_f$ is the number of filters of corresponding blocks (see Figure 4).

Block Type	Layers	# of Filters	Filter Size	Stride Size	Padding	Activation
Contracting	Convolution	$N_f$	(3,3,3)	(1,1,1)	yes	ReLU
	Batch normalization	-	-	-	-	-
	Convolution	$N_f$	(3,3,3)	(1,1,1)	yes	ReLU
	Batch normalization	-	-	-	-	-
	Max pooling	-	(2,2,2)	-	-	-
Bottleneck	Convolution	$N_f$	(3,3,3)	(1,1,1)	yes	ReLU
	Batch normalization	-	-	-	-	-
	Convolution	$N_f$	(3,3,3)	(1,1,1)	yes	ReLU
	Batch normalization	-	-	-	-	-
Expanding	Up-sampling	$N_f$	(2,2,2)	(2,2,2)	yes	-
	Convolution	$N_f$	(3,3,3)	(1,1,1)	yes	ReLU
	Batch normalization	-	-	-	-	-
	Convolution	$N_f$	(3,3,3)	(1,1,1)	yes	ReLU
	Batch normalization	-	-	-	-	-
Outputs	Convolution	$N_f$	(1,1,1)	(1,1,1)	-	sigmoid