Deep membrane systems for multitask segmentation in diabetic retinopathy

Automatic segmentation of microaneurysms (MAs), hard exudates (EXs) and optic disc (OD) are crucial to the diagnostic assessment of diabetic retinopathy (DR). However, the small sizes of MAs and EXs, as well as the large variations in the locations and shapes of MAs and EXs make these segmentation t...

Full description

Saved in:
Bibliographic Details
Published in:Knowledge-based systems 2019-11, Vol.183, p.104887, Article 104887
Main Authors: Xue, Jie, Yan, Shuo, Qu, Jianhua, Qi, Feng, Qiu, Chenggong, Zhang, Hongyan, Chen, Meirong, Liu, Tingting, Li, Dengwang, Liu, Xiyu
Format: Article
Language:English
Subjects:
Artificial neural networks
Deep convolutional neural networks
Diabetes
Diabetic retinopathy
Diagnostic systems
Exudation
Hybrid structures
Hybrid systems
Membrane structures
Membranes
Multitask segmentation
New membrane systems
Segmentation
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Automatic segmentation of microaneurysms (MAs), hard exudates (EXs) and optic disc (OD) are crucial to the diagnostic assessment of diabetic retinopathy (DR). However, the small sizes of MAs and EXs, as well as the large variations in the locations and shapes of MAs and EXs make these segmentation tasks challenging. To alleviate these challenges, in this paper, we propose a novel and automatic multitask segmentation method based on a new membrane system named a dynamic membrane system with hybrid structures. Three new types of rules in the new membrane system are designed to solve complex real applications in parallel. In membrane structures, efficient convolutional neural networks (CNNs) are implemented to perform pixel-wise segmentations of MAs, EXs and OD in DR. Evaluations on three public datasets demonstrate the robustness of our proposed method for correctly segmenting MAs, EXs and OD in various settings. Our experimental results outperform the state-of-the-art methods.
ISSN:0950-7051
1872-7409
DOI:10.1016/j.knosys.2019.104887