Fundus Microvascular Image Segmentation Method Based on Parallel U‐Net Model

Xinjuan Liu; Xu Han; Erxi Fang

doi:10.3788/CJL241041

Chinese Journal of Lasers, Volume. 51, Issue 21, 2107108(2024)

Fundus Microvascular Image Segmentation Method Based on Parallel U‐Net Model

Xinjuan Liu, Xu Han, and Erxi Fang^*

Author Affiliations

School of Electronic and Information, Soochow University, Suzhou 215006, Jiangsu , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(14)

Fig. 1. Overall network structure of MPU-Net

Download full size

Fig. 2. Fundus vascular and microvascular labels. (a) Ground truth; (b) microvascular label

Download full size

Fig. 3. Example of a morphological structure operator

Download full size

Fig. 4. Schematic diagram of the multi-scale feature mixing and fusion module (MSF)

Download full size

Fig. 5. Vascular segmentation dataset

Download full size

Fig. 6. Heat map examples showing the probabilistic features output by the network before and after adding MSF

Download full size

Fig. 7. Heat map examples showing the probability features output by the network before and after adding Mic-Net

Download full size

Fig. 8. Vascular segmentation examples of ablation experiment on the DRIVE test set

Download full size

Fig. 9. Vascular segmentation examples of ablation experiment on the CHASE_DB1 test set

Download full size

Fig. 10. Vascular segmentation examples of ablation experiment on the STARE test set

Download full size

Table 1. Experimental results comparison of MPU-Net with existing state-of-the-art methods on the DRIVE test set

View table

Table 1. Experimental results comparison of MPU-Net with existing state-of-the-art methods on the DRIVE test set

Method	Year	Accuracy	Sensitivity	Specificity	AUC
U-Net^［6］	2015	0.9531	0.7537	0.9820	0.9755
Yan et al.^［20］	2019	0.9538	0.7631	0.9820	0.9750
DG-Net^［10］	2020	0.9604	0.7614	0.9837	0.9846
CS2-Net^［21］	2021	0.9553	0.8154	0.9757	0.9784
ACCA-MLA-D-U-Net^［12］	2022	0.9581	0.8046	0.9850	0.9827
Mao et al.^［14］	2023		0.8105	0.9785	0.9812
TDCAU-Net^［16］	2024	0.9556	0.8187	0.9756	0.9795
MPU-Net （ours）	2024	0.9710	0.8243	0.9853	0.9889

Table 2. Experimental results comparison of MPU-Net with existing state-of-the-art methods on the CHASE_DB1 test set

View table

Table 2. Experimental results comparison of MPU-Net with existing state-of-the-art methods on the CHASE_DB1 test set

Method	Year	Accuracy	Sensitivity	Specificity	AUC
U-Net^［6］	2015	0.9578	0.8228	0.9701	0.9772
Vessel-Net^［22］	2019	0.9661	0.8132	0.9814	0.9860
CS2-Net^［21］	2021	0.9651	0.8329	0.9784	0.9851
ACCA-MLA-D-U-Net^［12］	2022	0.9673	0.8402	0.9801	0.9874
WA-Net^［23］	2022	0.9653	0.8042	0.9826	0.9841
Mao et al.^［14］	2023		0.8241	0.9850	0.9893
TDCAU-Net^［16］	2024	0.9738	0.8243	0.9836	0.9878
MPU-Net （ours）	2024	0.9764	0.8593	0.9844	0.9913

Table 3. Experimental results comparison of MPU-Net with existing state-of-the-art methods on the STARE test set

View table

Table 3. Experimental results comparison of MPU-Net with existing state-of-the-art methods on the STARE test set

Method	Year	Accuracy	Sensitivity	Specificity	AUC
Yan et al.^［24］	2018	0.9612	0.7581	0.9846	0.9801
Yan et al.^［20］	2019	0.9638	0.7735	0.9857	0.9833
CS2-Net^［21］	2021	0.9670	0.8396	0.9813	0.9875
ACCA-MLA-D-U-Net^［12］	2022	0.9665	0.7914	0.9870	0.9864
LUVS-Net^［25］	2023	0.9753	0.8133	0.9861	0.8187
MPU-Net （ours）	2024	0.9768	0.7844	0.9907	0.9905

Table 4. Comparison in ablation experimental results (mean ± std)

View table

Table 4. Comparison in ablation experimental results (mean ± std)

Dataset	Method	MSF	Mic-Net	Accuracy	Sensitivity	Specificity	Dice coefficient	AUC	Sen_Mic
DRIVE	1	×	×	0.9672± 0.0002	0.7928± 0.0073	0.9842± 0.0009	0.8082±0.0008	0.9816± 0.0010	0.6815± 0.0127
	2	√	×	0.9708 ±0.0001	0.8234 ±0.0052	0.9851± 0.0006	0.8309±0.0002	0.9887± 0.0001	0.7153± 0.0066
	3	√	√	0.9710± 0.0001	0.8243± 0.0012	0.9853± 0.0002	0.8314±0.0002	0.9889± 0.0001	0.7199± 0.0013
Dataset	Method	MSF	Mic-Net	Accuracy	Sensitivity	Specificity	Dice coefficient	AUC	Sen_Mic
CHASE_DB1	1	×	×	0.9757± 0.0001	0.8296± 0.0037	0.9855± 0.0003	0.8112±0.0006	0.9878± 0.0006	0.7528± 0.0050
	2	√	×	0.9762± 0.0002	0.8563± 0.0045	0.9843± 0.0005	0.8193±0.0042	0.9912± 0.0001	0.7833± 0.0064
	3	√	√	0.9764± 0.0002	0.8593± 0.0034	0.9844± 0.0005	0.8212±0.0032	0.9913± 0.0001	0.7874± 0.0052
Dataset	Method	MSF	Mic-Net	Accuracy	Sensitivity	Specificity	Dice coefficient	AUC	Sen_Mic
STARE	1	×	×	0.9766± 0.0005	0.7413± 0.0100	0.9931± 0.0003	0.7979±0.0056	0.9831± 0.0013	0.6066± 0.0184
	2	√	×	0.9764± 0.0003	0.7726± 0.0081	0.9911± 0.0005	0.8031±0.0030	0.9829± 0.0011	0.6345± 0.0090
	3	√	√	0.9768± 0.0002	0.7844± 0.0059	0.9907± 0.0005	0.8089±0.0016	0.9905± 0.0005	0.6388± 0.0042

Tools

Get Citation

Copy Citation Text

Xinjuan Liu, Xu Han, Erxi Fang. Fundus Microvascular Image Segmentation Method Based on Parallel U‐Net Model[J]. Chinese Journal of Lasers, 2024, 51(21): 2107108

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Biomedical Optical Imaging

Received: Jul. 9, 2024

Accepted: Aug. 26, 2024

Published Online: Oct. 31, 2024

The Author Email: Erxi Fang (fangerxi@suda.edu.cn)

DOI:10.3788/CJL241041

CSTR:32183.14.CJL241041

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology