Joint 2D attention gate and channel-spatial attention network for retinal vessel segmentation of OCT-angiography images

OCT-angiography is a non-invasive visualization imaging technology with high resolution that can more clearly image tiny blood vessels. Certain ophthalmic diseases can be diagnosed by the morphological changes of retinal blood vessels, and the automatic segmentation of blood vessel will definitely r...

Full description

Saved in:
Bibliographic Details
Published in:Signal, image and video processing image and video processing, 2023-06, Vol.17 (4), p.1219-1226
Main Authors: Cui, Baoyu, Qi, Sumin, Meng, Jing, Jiang, Yingjie
Format: Article
Language:English
Subjects:
Angiography
Blood vessels
Computer Imaging
Computer Science
Datasets
Error reduction
Image Processing and Computer Vision
Image resolution
Image segmentation
Medical imaging
Multimedia Information Systems
Original Paper
Pattern Recognition and Graphics
Signal,Image and Speech Processing
Vision
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:OCT-angiography is a non-invasive visualization imaging technology with high resolution that can more clearly image tiny blood vessels. Certain ophthalmic diseases can be diagnosed by the morphological changes of retinal blood vessels, and the automatic segmentation of blood vessel will definitely reduce workload of ophthalmologists. Current segmentation methods result in the loss of important detailed features and the segmenting failure of tiny blood vessels. We propose a joint 2D attention gate and channel-spatial attention network for the retinal vessel segmentation of OCT-angiography images to improve accuracy and reduce error rate. The improved 2D attention gate introduces attention coefficients to increase the weight of the blood vessel area, which preserves the branch structure and edge information of tiny blood vessels to a greater extent by combining context feature information of different levels of units. The channel-spatial attention block is used to enhance channel and spatial channel information, respectively. To enhance the credibility, our network was tested on two public data sets. Compared with channel and spatial attention network, U-Net with residual block and U-Net, the important indicator of accuracy is increased by 1.97%, 2.18% and 3.04%, respectively, in data set I and increased by 0.18%, 0.19% and 0.62%, respectively, in data set II.
ISSN:1863-1703
1863-1711
DOI:10.1007/s11760-022-02329-6