Polyp image segmentation based on multi-scale ResNeSt-50 aggregation network and message passing

Table 1. Five datasets used in this paper
View table
View in Article
Table 1. Five datasets used in this paper
数据集 Kvasir-SEG^［34］ CVC-ClinicDB^［33］ ColonDB^［35］ ETIS-LaribPolypDB ^［1］ CVC-300^［36］
数量 1 000 612 380 196 60
图像大小 500⁺×500⁺ 384×288 574×500 1 225×966 574×500

Table 2. Allocation of training dataset and test dataset of experiment

View table

View in Article

Table 2. Allocation of training dataset and test dataset of experiment

	训练集	测试集
实验A	Kvasir-SEG（900），CVC-ClinicDB（550）	Kvasir-SEG（100）
实验B	Kvasir-SEG（900），CVC-ClinicDB（550）	CVC-ClinicDB（62）
实验C	Kvasir-SEG（900），CVC-ClinicDB（550）	ColonDB（380）
实验D	Kvasir-SEG（900），CVC-ClinicDB（550）	ETIS-LaribPolypDB （196）
实验E	Kvasir-SEG（900），CVC-ClinicDB（550）	CVC-300（60）

Table 3. Ablation experiment of ResNeSt-50 module（training on Kvasir-SEG and ClinicDB dataset）

View table

View in Article

Table 3. Ablation experiment of ResNeSt-50 module（training on Kvasir-SEG and ClinicDB dataset）

模型	测试集	mDice	mIoU	Smeasure	MAE
模型a	ColonDB	0.512	0.444	0.712	0.061
模型b		0.608	0.523	0.760	0.052
模型c		0.676	0.606	0.799	0.043
模型d		0.757	0.675	n/a	n/a
模型a	ETIS- LaribPolypDB	0.398	0.335	0.684	0.036
模型b		0.522	0.447	0.738	0.026
模型c		0.614	0.547	0.781	0.028
模型d		0.706	0.628	n/a	n/a

Table 4. Effect of multi-scale training on model segmentation
View table
View in Article
Table 4. Effect of multi-scale training on model segmentation
mDice mIoU Smeasure MAE
未采用多尺度训练 0.888 0.827 0.908 0.031
使用多尺度训练 0.913 0.858 0.925 0.023

Table 5. Effect of RAdam optimizer on model segmentation
View table
View in Article
Table 5. Effect of RAdam optimizer on model segmentation
Epoch mDice mIoU Smeasure MAE
Adam 20 0.904 0.851 0.920 0.027
RAdam 30 0.913 0.858 0.925 0.023

Table 6. Trained on CVC-ClinicDB^［33］，Kvasir-SEG^［34］dataset， compare the test results on Kvasir-SEG^［34］dataset
View table
View in Article
Table 6. Trained on CVC-ClinicDB^［33］，Kvasir-SEG^［34］dataset， compare the test results on Kvasir-SEG^［34］dataset
mDice mIoU Smeasure MAE
本文编码-解码器 0.913 0.858 0.925 0.023
本文编码-解码器+TTA 0.915 0.861 0.925 0.023
本文模型 0.916 0.863 0.921 0.023

Table 7. Trained on the CVC-ClinicDB^［33］，Kvasir-SEG^［34］dataset， respectively compare the test results on ColonDB^［35］，ETIS-LaribPolypDB^［1］，CVC-300^［36］

View table

View in Article

Table 7. Trained on the CVC-ClinicDB^［33］，Kvasir-SEG^［34］dataset， respectively compare the test results on ColonDB^［35］，ETIS-LaribPolypDB^［1］，CVC-300^［36］

	ColonDB^［35］		ETIS-LaribPolypDB^［1］		CVC-300^［36］
	mDice	mIoU	mDice	mIoU	mDice	mIoU
本文编码-解码器	0.757	0.675	0.706	0.628	0.858	0.781
本文编码-解码器 +TTA	0.783	0.700	0.764	0.686	0.869	0.797
本文模型	0.786	0.703	0.770	0.691	0.871	0.798

Table 8. Trained on CVC-ClinicDB^［33］，Kvasir-SEG^［34］dataset， compare the test results on Kvasir-SEG^［34］dataset

View table

View in Article

Table 8. Trained on CVC-ClinicDB^［33］，Kvasir-SEG^［34］dataset， compare the test results on Kvasir-SEG^［34］dataset

	mDice	mIoU	Smeasure	MAE
U-Net^［9］	0.818	0.746	0.858	0.055
U-Net++^［10］	0.821	0.743	0.862	0.048
ResUnet++^［11］	0.734	0.639	n/a	n/a
SFA^［38］	0.723	0.611	0.782	0.075
PraNet^［12］	0.898	0.840	0.915	0.030
本文方法	0.916	0.863	0.921	0.023

Table 9. Trained on CVC-ClinicDB，Kvasir-SEGdataset， compare the test results on ETIS-LaribPolypDBdataset

View table

View in Article

Table 9. Trained on CVC-ClinicDB，Kvasir-SEGdataset， compare the test results on ETIS-LaribPolypDBdataset

	mDice	mIoU	Smeasure	MAE
U-Net^［9］	0.055	0.046	0.517	0.021
U-Net++^［10］	0.093	0.078	0.545	0.018
SFA^［38］	0.012	0.006	0.425	0.130
ResUnet++^［11］	0.016	0.009	0.500	0.011
PraNet^［12］	0.099	0.073	0.515	0.074
本文方法	0.487	0.401	0.734	0.003

Table 10. Trained on the CVC-ClinicDB^［33］，Kvasir-SEG^［34］dataset， respectively compare the test results on ColonDB^［35］，ETIS-LaribPolypDB^［1］，CVC-300^［36］dataset

View table

View in Article

Table 10. Trained on the CVC-ClinicDB^［33］，Kvasir-SEG^［34］dataset， respectively compare the test results on ColonDB^［35］，ETIS-LaribPolypDB^［1］，CVC-300^［36］dataset

	ColonDB^［35］		ETIS-LaribPolypDB^［1］		CVC-300^［36］
	mDice	MIoU	mDice	MIoU	mDice	MIoU
U-Net^［9］	0.512	0.444	0.398	0.335	0.71	0.627
U-Net++^［10］	0.483	0.410	0.401	0.344	0.707	0.624
ResUnet++^［11］	n/a	n/a	n/a	n/a	n/a	n/a
SFA^［38］	0.469	0.347	0.297	0.217	0.467	0.329
PraNet^［12］	0.709	0.640	0.628	0.567	0.871	0.797
本文方法	0.786	0.703	0.770	0.691	0.871	0.798

Tools

Get Citation

Copy Citation Text

Ping XIA, Guangyi ZHANG, Bangjun LEI, Yaobing ZOU, Tinglong TANG. Polyp image segmentation based on multi-scale ResNeSt-50 aggregation network and message passing[J]. Optics and Precision Engineering, 2023, 31(18): 2765

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Information Sciences

Received: Mar. 30, 2023

Accepted: --

Published Online: Oct. 12, 2023

The Author Email: LEI Bangjun (Bangjun.Lei@ieee.org)

DOI:10.37188/OPE.20233118.2765

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Five datasets used in this paper

Table 1. Five datasets used in this paper

Table 2. Allocation of training dataset and test dataset of experiment

Table 2. Allocation of training dataset and test dataset of experiment

Table 3. Ablation experiment of ResNeSt-50 module（training on Kvasir-SEG and ClinicDB dataset）

Table 3. Ablation experiment of ResNeSt-50 module（training on Kvasir-SEG and ClinicDB dataset）

Table 4. Effect of multi-scale training on model segmentation

Table 4. Effect of multi-scale training on model segmentation

Table 5. Effect of RAdam optimizer on model segmentation

Table 5. Effect of RAdam optimizer on model segmentation

Table 6. Trained on CVC-ClinicDB［33］，Kvasir-SEG［34］ dataset， compare the test results on Kvasir-SEG［34］ dataset

Table 6. Trained on CVC-ClinicDB［33］，Kvasir-SEG［34］ dataset， compare the test results on Kvasir-SEG［34］ dataset

Table 7. Trained on the CVC-ClinicDB［33］，Kvasir-SEG［34］dataset， respectively compare the test results on ColonDB［35］，ETIS-LaribPolypDB［1］，CVC-300［36］

Table 7. Trained on the CVC-ClinicDB［33］，Kvasir-SEG［34］dataset， respectively compare the test results on ColonDB［35］，ETIS-LaribPolypDB［1］，CVC-300［36］

Table 8. Trained on CVC-ClinicDB［33］，Kvasir-SEG［34］ dataset， compare the test results on Kvasir-SEG［34］ dataset

Table 8. Trained on CVC-ClinicDB［33］，Kvasir-SEG［34］ dataset， compare the test results on Kvasir-SEG［34］ dataset

Table 9. Trained on CVC-ClinicDB，Kvasir-SEGdataset， compare the test results on ETIS-LaribPolypDBdataset

Table 9. Trained on CVC-ClinicDB，Kvasir-SEGdataset， compare the test results on ETIS-LaribPolypDBdataset

Table 10. Trained on the CVC-ClinicDB［33］，Kvasir-SEG［34］dataset， respectively compare the test results on ColonDB［35］，ETIS-LaribPolypDB［1］，CVC-300［36］dataset

Table 10. Trained on the CVC-ClinicDB［33］，Kvasir-SEG［34］dataset， respectively compare the test results on ColonDB［35］，ETIS-LaribPolypDB［1］，CVC-300［36］dataset

微信扫一扫：分享

Table 6. Trained on CVC-ClinicDB^［33］，Kvasir-SEG^［34］dataset， compare the test results on Kvasir-SEG^［34］dataset

Table 6. Trained on CVC-ClinicDB^［33］，Kvasir-SEG^［34］dataset， compare the test results on Kvasir-SEG^［34］dataset

Table 7. Trained on the CVC-ClinicDB^［33］，Kvasir-SEG^［34］dataset， respectively compare the test results on ColonDB^［35］，ETIS-LaribPolypDB^［1］，CVC-300^［36］

Table 7. Trained on the CVC-ClinicDB^［33］，Kvasir-SEG^［34］dataset， respectively compare the test results on ColonDB^［35］，ETIS-LaribPolypDB^［1］，CVC-300^［36］

Table 8. Trained on CVC-ClinicDB^［33］，Kvasir-SEG^［34］dataset， compare the test results on Kvasir-SEG^［34］dataset

Table 8. Trained on CVC-ClinicDB^［33］，Kvasir-SEG^［34］dataset， compare the test results on Kvasir-SEG^［34］dataset

Table 10. Trained on the CVC-ClinicDB^［33］，Kvasir-SEG^［34］dataset， respectively compare the test results on ColonDB^［35］，ETIS-LaribPolypDB^［1］，CVC-300^［36］dataset

Table 10. Trained on the CVC-ClinicDB^［33］，Kvasir-SEG^［34］dataset， respectively compare the test results on ColonDB^［35］，ETIS-LaribPolypDB^［1］，CVC-300^［36］dataset