Retinal Vessel Segmentation of Prematurity Infants Based on FDMU-net

Liang Wang; Chunxiao Chen; Xue Fu; Lin Wang

doi:10.3788/LOP202158.1417003

Journals >Laser & Optoelectronics Progress >Volume 58 >Issue 14 >Page 1417003 > Article

Laser & Optoelectronics Progress
Vol. 58, Issue 14, 1417003 (2021)

Retinal Vessel Segmentation of Prematurity Infants Based on FDMU-net

Liang Wang¹, Chunxiao Chen^1、*, Xue Fu¹, and Lin Wang²

Author Affiliations

¹College of Automation Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing, Jiangsu 211106, China

²Shanghai Shengwei Medical Technology Co., Ltd., Shanghai 201321, China

show less

DOI: 10.3788/LOP202158.1417003 Cite this Article Set citation alerts

Liang Wang, Chunxiao Chen, Xue Fu, Lin Wang. Retinal Vessel Segmentation of Prematurity Infants Based on FDMU-net[J]. Laser & Optoelectronics Progress, 2021, 58(14): 1417003 Copy Citation Text

show less

Fig. 1. Network architecture FDMU-net

Download full size

Fig. 2. Dense block

Download full size

Fig. 3. Multi-convolution kernel extraction unit

Download full size

Fig. 4. Different models for ROC curve on each dataset. (a) DRIVE; (b) STARE; (c) Shengwei

Download full size

Fig. 5. Comparison of different models for segmentation of small blood vessels of different models. (a) Original fundus image; (b) segmentation of U-net; (c) segmentation of FDMU-net

Download full size

Fig. 6. Comparison of anti-interference ability of different models. (a) Original fundus image; (b) segmentation of U-net; (c) segmentation of FDMU-net

Download full size

Fig. 7. Different models for segmentation on each dataset. (a) Fundus images of premature infants; (b) manual labeling;(c) segmentation results of Ref.[17]; (d) segmentation results of Ref.[18]; (e) segmentation results of FMDU-net

Download full size

Dataset	No. of original images	No. of images	Original size	Input size
DRIVE	32	320	565×584	565×584
STARE	16	160	605×700	605×700
Shengwei	64	640	1200×1600	480×640

Table 1. Parameters of training set for each dataset

Dataset	Method	Accuracy	Sensitivity	Specificity	Dice	AUC
DRIVE	U-net	0.9618	0.7709	0.9812	0.7871	0.9731
	DU-net	0.9670	0.7881	0.9818	0.8043	0.9701
	DMU-net	0.9671	0.8122	0.9819	0.8177	0.9764
	FDMU-net	0.9675	0.8152	0.9826	0.8191	0.9808
STARE	U-net	0.9627	0.8215	0.9755	0.7894	0.9731
	DU-net	0.9678	0.8217	0.9782	0.7961	0.9757
	DMU-net	0.9675	0.8391	0.9784	0.8003	0.9837
	FDMU-net	0.9685	0.8484	0.9783	0.8011	0.9892
Shengwei	U-net	0.9690	0.6203	0.9812	0.6064	0.9446
	DU-net	0.9720	0.6400	0.9817	0.6232	0.9477
	DMU-net	0.9727	0.6594	0.9833	0.6279	0.9590
	FDMU-net	0.9728	0.6633	0.9840	0.6302	0.9595

Table 2. Experimental comparison results on different datasets

Dataset	Method	Year	Accuracy	Sensitivity	Specificity	AUC
DRIVE	Ref. [17]	2019	0.9546	0.8053	0.9767	0.9771
	Ref. [18]	2020	0.9562	0.7823	0.9815	0.9793
	FDMU-net		0.9675	0.8152	0.9826	0.9808
STARE	Ref. [17]	2019	0.9684	0.8299	0.9794	0.9817
	Ref. [18]	2020	0.9617	0.8217	0.9766	0.9854
	FDMU-net		0.9685	0.8484	0.9783	0.9892
Shengwei	Ref. [17]	2019	0.9697	0.6255	0.9831	0.9447
	Ref. [18]	2020	0.9717	0.6337	0.9786	0.9503
	FDMU-net		0.9728	0.6633	0.9840	0.9595

Table 3. Performance comparison of FDMU-net and other methods

Liang Wang, Chunxiao Chen, Xue Fu, Lin Wang. Retinal Vessel Segmentation of Prematurity Infants Based on FDMU-net[J]. Laser & Optoelectronics Progress, 2021, 58(14): 1417003

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information