Retinal Vessel Segmentation Based on Dynamic Feature Graph Convolutional Network

Linyi Miao; Feng Li

doi:10.3788/CJL240498

Chinese Journal of Lasers, Volume. 51, Issue 15, 1507208(2024)

Retinal Vessel Segmentation Based on Dynamic Feature Graph Convolutional Network

Linyi Miao and Feng Li^*

School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(12)

Fig. 1. Overall structure of DF-Net

Download full size

Fig. 2. Vanilla convolution-based SCDB and double dilated convolution block

Download full size

Fig. 3. Dynamic-feature graph convolutional module

Download full size

Fig. 4. Visualization of segmentation results for each method on the Fives dataset. (a) Original image; (b) ground truth; (c) U-Net;

Download full size

Fig. 5. Visualization of segmentation results for each method on the HRF dataset. (a) Original image; (b) ground truth; (c) U-Net;

Download full size

Fig. 6. Ablation experiment visualization results

Download full size

Table 1. Comparison of experimental results of different methods on the Fives dataset

View table

Table 1. Comparison of experimental results of different methods on the Fives dataset

Model	Year	A_cc	S_e	S_p	AUC	F₁	C_MCC
U-Net^［7］	2015	0.9846	0.8858	0.9837	0.9922	0.8889	0.8810
Deeplab V3+^［33］	2018	0.9845	0.8905	0.9923	0.9934	0.8894	0.8874
Attention U-Net^［10］	2018	0.9834	0.8991	0.9899	0.9928	0.8923	0.8866
U-Net++^［34］	2020	0.9844	0.8977	0.9911	0.9932	0.8915	0.8832
OCE-Net^［10］	2022	0.9846	0.8974	0.9929	0.9936	0.8906	0.8824
SGAT-Net^［35］	2023	0.9857	0.8943	0.9942	0.9926	0.8959	0.8959
G-CASCADE^［36］	2023	0.9869	0.8984	0.9925	0.9932	0.9023	0.8963
DF-Net	2024	0.9876	0.9088	0.9936	0.9950	0.9125	0.9059

Table 2. Comparison of experimental results of different methods on the HRF dataset

View table

Table 2. Comparison of experimental results of different methods on the HRF dataset

Model	Year	A_cc	S_e	S_p	AUC	F₁	C_MCC
U-Net^［7］	2015	0.9673	0.7781	0.9809	0.9786	0.7798	0.7726
Deeplab V3+^［33］	2018	0.9649	0.7810	0.9837	0.9813	0.7845	0.7785
Attention U-Net^［10］	2018	0.9635	0.7880	0.9801	0.9809	0.7828	0.7789
U-Net++^［34］	2020	0.9654	0.7939	0.9836	0.9823	0.7993	0.7980
OCE-Net^［11］	2022	0.9687	0.8357	0.9828	0.9849	0.8227	0.8111
SGAT-Net^［35］	2023	0.9680	0.8203	0.9840	0.9838	0.8223	0.8110
G-CASCADE^［36］	2023	0.9682	0.8189	0.9832	0.9825	0.8187	0.8089
DF-Net	2024	0.9733	0.8322	0.9837	0.9856	0.8318	0.8202

Table 3. Comparison of experimental results of different methods on the DRIVE dataset

View table

Table 3. Comparison of experimental results of different methods on the DRIVE dataset

Model	Year	A_cc	S_e	S_p	AUC	F₁	C_MCC
U-Net^［7］	2015	0.9534	0.7843	0.9810	0.9768	0.8112	0.8086
Deeplab V3+^［33］	2018	0.9518	0.7820	0.9800	0.9758	0.7983	0.7846
Attention U-Net^［10］	2018	0.9580	0.8163	0.9826	0.9842	0.8049	0.7860
U-Net++^［34］	2020	0.9586	0.8256	0.9823	0.9854	0.8124	0.8043
OCE-Net^［11］	2022	0.9611	0.8164	0.9804	0.9831	0.8289	0.8118
SGAT-Net^［35］	2023	0.9662	0.8026	0.9786	0.9800	0.8329	0.8176
G-CASCADE^［36］	2023	0.9642	0.8100	0.9824	0.9813	0.8280	0.8154
DF-Net	2024	0.9700	0.8081	0.9865	0.9861	0.8260	0.8081

Table 4. Comparison of experimental results of different methods on the STARE dataset

View table

Table 4. Comparison of experimental results of different methods on the STARE dataset

Model	Year	A_cc	S_e	S_p	AUC	F₁	C_MCC
U-Net^［7］	2015	0.9712	0.8167	0.9838	0.9857	0.8112	0.7707
Deeplab V3+^［33］	2018	0.9714	0.8078	0.9829	0.9849	0.8118	0.7898
Attention U-Net^［10］	2018	0.9717	0.8153	0.9844	0.9862	0.8127	0.7975
U-Net++^［34］	2020	0.9733	0.8264	0.9851	0.9883	0.8240	0.8097
OCE-Net^［11］	2022	0.9742	0.8212	0.9865	0.9876	0.8341	0.8164
SGAT-Net^［35］	2023	0.9768	0.8328	0.9872	0.9934	0.8445	0.8206
G-CASCADE^［36］	2023	0.9746	0.8232	0.9868	0.9870	0.8348	0.8132
DF-Net	2024	0.9764	0.8279	0.9874	0.9918	0.8403	0.8161

Table 5. Comparison of experimental results of different methods on the CHASE_DB1 dataset

View table

Table 5. Comparison of experimental results of different methods on the CHASE_DB1 dataset

Model	Year	A_cc	S_e	S_p	AUC	F₁	C_MCC
U-Net^［7］	2015	0.9676	0.7650	0.9784	0.9783	0.7698	0.7432
Deeplab V3+^［33］	2018	0.9723	0.7980	0.9827	0.980	0.7855	0.7667
Attention U-Net^［10］	2018	0.9741	0.8125	0.9843	0.9847	0.7998	0.7863
U-Net++^［34］	2020	0.9753	0.8217	0.9850	0.9861	0.8094	0.7965
OCE-Net^［11］	2022	0.9778	0.8338	0.9824	0.9872	0.8186	0.7980
SGAT-Net^［35］	2023	0.9772	0.8378	0.9864	0.9896	0.8236	0.8012
G-CASCADE^［36］	2023	0.9770	0.8355	0.9848	0.9868	0.8226	0.7989
DF-Net	2024	0.9760	0.8396	0.9859	0.9901	0.8254	0.8042

Table 6. Comparison of ablation experimental results

View table

Table 6. Comparison of ablation experimental results

Model	A_cc	S_e	S_p	AUC	F₁	C_MCC
U-Net（Baseline）	0.9789	0.8758	0.9887	0.9722	0.8823	0.8710
U-Net+SDCB	0.9843	0.8783	0.9902	0.9829	0.8917	0.8835
U-Net+dilated_SDCB	0.9849	0.8831	0.9929	0.9861	0.8946	0.8878
U-Net+DDCB	0.9852	0.8958	0.9937	0.9921	0.8989	0.8932
U-Net+DFGCM	0.9866	0.8952	0.9936	0.9944	0.9052	0.8981
U-Net+DDCB+DFGCM （ours）	0.9876	0.9088	0.9936	0.9949	0.9126	0.9059

Tools

Get Citation

Copy Citation Text

Linyi Miao, Feng Li. Retinal Vessel Segmentation Based on Dynamic Feature Graph Convolutional Network[J]. Chinese Journal of Lasers, 2024, 51(15): 1507208

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Optical Diagnostics and Therapy

Received: Jan. 15, 2024

Accepted: Mar. 6, 2024

Published Online: Jul. 16, 2024

The Author Email: Li Feng (lifenggold@163.com)

DOI:10.3788/CJL240498

CSTR:32183.14.CJL240498

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology