Gland and Colonoscopy Segmentation Method Combining Self-Attention and Convolutional Neural Network

Jiabao Zhang; Zhiyong Xiao

doi:10.3788/LOP212696

Journals >Laser & Optoelectronics Progress >Volume 60 >Issue 2 >Page 0217002 > Article

Laser & Optoelectronics Progress
Vol. 60, Issue 2, 0217002 (2023)

Gland and Colonoscopy Segmentation Method Combining Self-Attention and Convolutional Neural Network

Jiabao Zhang and Zhiyong Xiao^*

Author Affiliations

School of Artificial Intelligence and Computer Science, Jiangnan University, Wuxi 214122, Jiangsu , China

show less

DOI: 10.3788/LOP212696 Cite this Article Set citation alerts

Jiabao Zhang, Zhiyong Xiao. Gland and Colonoscopy Segmentation Method Combining Self-Attention and Convolutional Neural Network[J]. Laser & Optoelectronics Progress, 2023, 60(2): 0217002 Copy Citation Text

show less

Fig. 1. Local UNet encoding structure

Download full size

Fig. 2. Cross fusion block structure

Download full size

Fig. 3. LG UNet structure diagram

Download full size

Fig. 4. Comparison of output results of each network

Download full size

Model	Dice		IOU
Model	Test A	Test B	Test A	Test B
U-Net	0.8826	0.8269	0.7934	0.7216
Local UNet	0.9133	0.8746	0.8447	0.7868
LG-v1	0.9267	0.8822	0.8663	0.8010
LG-v2	0.9302	0.8834	0.8711	0.8028
LG-v3	0.9362	0.8844	0.8817	0.8049

Table 1. Results of ablation experiment

Model	Dice		IOU
Model	Test A	Test B	Test A	Test B
U-Net -2	0.8826	0.8269	0.7934	0.7216
U-Net -1	0.8819	0.8539	0.7956	0.7606
LG UNet -2	0.9362	0.8844	0.8817	0.8049
LG UNet -1	0.9100	0.8663	0.8380	0.7858

Table 2. Double channel and single channel output results are compared

Model	Dice			IOU
Model	Test A	Test B	Mean	Test A	Test B	Mean
U-Net	0.8826	0.8269	0.8547	0.7934	0.7216	0.7575
ResUNet	0.8940	0.8459	0.8699	0.8146	0.7461	0.7803
KiU-Net^［28］	0.8898	0.8527	0.8712	0.8065	0.7565	0.7815
MedT	0.8674	0.8051	0.8362	0.7711	0.6889	0.7300
Rota-Net^［29］	0.919	0.849	0.884
U-Node^［30］	0.8930	0.8420	0.8675
LG UNet	0.9362	0.8844	0.9103	0.8817	0.8049	0.8433

Table 3. Comparison results with other algorithms on Glas dataset

Model	Dice	IOU
Model	Test A	Test A
U-Net	0.8110	0.7250
KiU-Net^［28］	0.7724	0.6689
ResUNet	0.8306	0.7477
ResUNet++	0.8508	0.7699
ColonSegNet^［31］	0.8206
MedT	0.8039	0.7116
LG UNet	0.8563	0.7782

Table 4. Comparison results with other algorithms on Kvasir-SEG dataset

Model	ClinicDB		NoMuSeg
Model	Dice	IOU	Dice	IOU
U-Net	0.7926	0.7115	0.7930	0.7188
LG UNet	0.8607	0.7946	0.8317	0.7126

Table 5. Comparisons on other data sets

Jiabao Zhang, Zhiyong Xiao. Gland and Colonoscopy Segmentation Method Combining Self-Attention and Convolutional Neural Network[J]. Laser & Optoelectronics Progress, 2023, 60(2): 0217002

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information

微信扫一扫：分享

微信扫一扫：分享