Related Experiment Video
Updated: Aug 7, 2026
09:17
Using Retinal Imaging to Study Dementia
Published on: November 6, 2017
Deep Learning-based Automated Segmentation of Ultra-Widefield Retinal Vasculature for Cardiometabolic Disease
IEEE Journal of Biomedical and Health Informatics
|August 5, 2026
Summary
A new deep learning framework, ECS-Net, accurately segments retinal vessels in ultra-widefield images. This method reveals disease-specific vascular patterns, aiding systemic health assessment.
Area of Science:
- Ophthalmology and Medical Imaging
- Artificial Intelligence in Healthcare
- Cardiovascular and Metabolic Diseases
Background:
- Ultra-widefield (UWF) retinal imaging offers insights into systemic microvascular health.
- Accurate retinal vessel segmentation in UWF images is challenging due to image complexity and reduced contrast.
- Existing methods struggle with the unique characteristics of true-color UWF retinal images.
Purpose of the Study:
- To develop a deep learning (DL) framework, ECS-Net, for precise retinal vessel segmentation in true-color UWF images.
- To quantify region-specific vascular parameters (VD, FD, TC, MC) from segmented UWF retinal images.
- To investigate the association between these vascular parameters and cardiometabolic diseases.
Main Methods:
- Developed ECS-Net, a DL framework with an enhanced encoder-decoder architecture incorporating a dual-domain context enhancement module (DCEM) and atrous spatial pyramid pooling (ASPP).
- Evaluated ECS-Net's performance using the Dice coefficient, comparing it against state-of-the-art algorithms.
- Quantified region-specific vascular parameters (central, peripheral, global zones) and performed multivariable logistic regression analysis in 4,618 participants.
Main Results:
- ECS-Net achieved a superior Dice coefficient of 0.8349, outperforming existing methods (0.7001-0.8136).
- Hypertension was inversely associated with vessel density (VD) and fractal dimension (FD) across all retinal zones.
- Diabetes showed regional specificity, with altered tortuosity (TC), mean curvature (MC), and FD predominantly in peripheral and global zones.
Conclusions:
- ECS-Net is the first DL framework tailored for retinal vessel segmentation in true-color UWF images.
- The study identified disease-specific regional vascular patterns in UWF images, highlighting limitations of conventional fundus photography.
- UWF imaging combined with advanced segmentation offers valuable insights for systemic disease assessment and developing interpretable AI models.