Computational and Mathematical Methods in Medicine

Research Article

SERR-U-Net: Squeeze-and-Excitation Residual and Recurrent Block-Based U-Net for Automatic Vessel Segmentation in Retinal Image

Table 1

Details of SERR-U-Net.


Block name	Layer (type)	Output size	Params

	Input (inputlayer)	(48, 48, 1)	0
Encoder block (1)	Conv2d_1 (Conv2D)	(48, 48, 16)	2320
Encoder block (1)	B_N(Batch Normalization)	(48, 48, 16)	64
SE_block_1 (SE-ResNet)
Max_pooling2d (maxpooling2D)
Encoder block (2)	Conv2d_2 (Conv2D)	(24, 24, 32)	9248
Encoder block (2)	B_N(Batch Normalization)	(24, 24, 32)	128
SE_block_2 (SE-ResNet)
Max_pooling2d (maxpooling2D)
Encoder block (3)	Conv2d_3 (Conv2D)	(12, 12, 64)	36928
Encoder block (3)	B_N(Batch Normalization)	(12, 12, 64)	256
SE_block_3 (SE-ResNet)
Max_pooling2d (maxpooling2D)
Encoder block (4)	Conv2d_4 (Conv2D)	(6, 6, 128)	147712
Encoder block (4)	B_N(Batch Normalization)	(6, 6, 128)	512
SE_block_4 (SE-ResNet)
Encoder block (5)	Conv2d_5 (Conv2D)	(6, 6, 128)	147712
Encoder block (5)	B_N(Batch Normalization)	(6, 6, 128)	512
Up_sampling2d (upsampling2D)
Decoder block (6)	Conv2d_6 (Conv2D)	(12, 12, 64)	36928
Decoder block (6)	B_N(Batch Normalization)	(12, 12, 64)	256
Up_sampling2d (upsampling2D)
Decoder block (7)	Conv2d_7 (Conv2D)	(24, 24, 32)	9248
Decoder block (7)	B_N(Batch Normalization)	(24, 24, 32)	128
Up_sampling2d (upsampling2D)
Decoder block (8)	Conv2d_8 (Conv2D)	(48, 48, 16)	2320
Decoder block (8)	B_N(Batch Normalization)	(48, 48, 16)	64
	Conv2d_9 (Conv2D)	(48, 48, 1)	0