Multi-Scale Retinal Vessel Segmentation Based on Fully Convolutional Neural Network

Tingyue Zheng; Chen Tang; Zhenkun Lei

doi:10.3788/AOS201939.0211002

Acta Optica Sinica, Volume. 39, Issue 2, 0211002(2019)

Multi-Scale Retinal Vessel Segmentation Based on Fully Convolutional Neural Network

Tingyue Zheng^1、*, Chen Tang^1、*, and Zhenkun Lei²

Author Affiliations

¹ School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China

² State Key Laboratory of Structural Analysis for Industrial Equipment, Dalian University of Technology, Dalian, Liaoning 116024, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(10)

Fig. 1. Structural diagram of residual block

Download full size

Fig. 2. Schematic of multi-scale ASPP module

Download full size

Fig. 3. Structural diagram of network for retinal vessel segmentation

Download full size

Fig. 4. Image preprocessing. (a) Typical fundus image of DRIVE dataset; (b) fundus image after preprocessing

Download full size

Fig. 5. Segmentation test results based on DRIVE dataset. (a) Original fundus images; (b) segmentation standard images; (c) segmentation results of images

Download full size

Fig. 6. Segmentation test results based on STARE dataset. (a) Original fundus images; (b) segmentation standard images; (c) segmentation results of images

Download full size

Fig. 7. Segmentation results in local areas. (a)(b) Original fundus images; (c)-(f) local fundus images; (g)-(j) segmentation standard images; (k)-(n) segmentation results of images

Download full size

Table 1. Average performance evaluation results based on DRIVE and STARE datasets
View table
Table 1. Average performance evaluation results based on DRIVE and STARE datasets
Dataset Method R_Se /% R_Sp /% R_Acc /% R_AUC /%
DRIVE 2nd human observer 77.60 97.24 94.72 —
Proposed method 80.53 97.67 95.46 97.71
STARE 2nd human observer 89.52 93.84 93.49 —
Proposed method 82.99 97.94 96.84 98.17

Table 2. Performance comparison of the proposed and other methods based on DRIVE dataset

View table

Table 2. Performance comparison of the proposed and other methods based on DRIVE dataset

Type	Method	Year	R_Se /%	R_Sp /%	R_Acc /%	R_AUC /%
	Ref.[3]	2011	74.10	97.51	94.34	—
Unsupervised	Ref. [5]	2014	62.80	98.40	93.80	—
methods	Ref. [2]	2015	76.55	97.04	94.42	96.14
	Ref. [7]	2015	74.20	98.20	95.40	86.20
	Ref. [9]	2012	74.06	98.07	94.80	97.47
	Ref. [13]	2015	81.73	97.33	97.67	94.75
Supervised	Ref. [12]	2016	77.63	97.68	94.95	97.20
methods	Ref. [14]	2016	75.69	98.16	95.27	97.38
	Ref. [15]	2016	76.03	—	95.23	—
	Ref. [17]	2017	75.01	97.95	94.99	—
	Proposed method	2018	80.53	97.67	95.46	97.71

Table 3. Performance comparison of the proposed and other methods based on STARE dataset

View table

Table 3. Performance comparison of the proposed and other methods based on STARE dataset

Type	Methods	Year	R_Se /%	R_Sp /%	R_Acc /%	R_AUC /%
	Ref. [3]	2011	72.60	97.56	94.97	—
Unsupervised	Ref. [5]	2014	58.60	98.70	94.48	—
methods	Ref. [2]	2015	77.16	97.01	95.63	94.97
	Ref. [7]	2015	78.00	97.80	95.60	87.40
	Ref. [9]	2012	75.48	97.63	95.34	97.68
	Ref. [13]	2015	81.04	97.91	98.13	97.51
Supervised	Ref. [12]	2016	78.67	97.54	95.66	97.85
methods	Ref. [14]	2016	77.26	98.44	96.28	98.79
	Ref. [15]	2016	74.12	—	95.85	—
	Proposed method	2018	82.99	97.94	96.84	98.17

Tools

Get Citation

Copy Citation Text

Tingyue Zheng, Chen Tang, Zhenkun Lei. Multi-Scale Retinal Vessel Segmentation Based on Fully Convolutional Neural Network[J]. Acta Optica Sinica, 2019, 39(2): 0211002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Imaging Systems

Received: Aug. 2, 2018

Accepted: Sep. 25, 2018

Published Online: May. 10, 2019

The Author Email:

DOI:10.3788/AOS201939.0211002

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Average performance evaluation results based on DRIVE and STARE datasets

Table 1. Average performance evaluation results based on DRIVE and STARE datasets

Table 2. Performance comparison of the proposed and other methods based on DRIVE dataset

Table 2. Performance comparison of the proposed and other methods based on DRIVE dataset

Table 3. Performance comparison of the proposed and other methods based on STARE dataset

Table 3. Performance comparison of the proposed and other methods based on STARE dataset