A Graph Neural Network-Based Multispectral-View Learning Model for Diabetic Macular Ischemia Detection From Color
Qinghua He1,2,3, Hongyang Jiang1, Danqi Fang1
1Department of Ophthalmology and Visual Sciences, The Chinese University of Hong Kong, Hong Kong Special Administrative Region, China.
Translational Vision Science & Technology
|June 23, 2026
Summary
Artificial intelligence (AI) combined with color fundus photographs (CFPs) can effectively detect diabetic macular ischemia (DMI). This AI model offers a feasible, early, and cost-effective screening method for DMI, improving outcomes for diabetic patients.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Diabetic macular ischemia (DMI) is a vision-impairing condition in diabetic patients due to retinal capillary loss.
- Current diagnostic methods for DMI are limited, and its detection using color fundus photographs (CFPs) and AI is unexplored.
Purpose of the Study:
- To investigate the feasibility of using AI and CFPs for detecting diabetic macular ischemia (DMI).
- To address the skepticism regarding the viability of AI-based DMI detection from CFPs.
Main Methods:
- A graph neural network-based multispectral-view learning (GNN-MSVL) model was developed using 1078 CFPs from diabetic patients.
- The model reconstructs pseudo-multispectral images from CFPs to enhance sensitivity to ischemic changes.
- ResNeXt101 and a customized GNN were employed for feature extraction and cross-spectral relationship learning.
Main Results:
- The GNN-MSVL model achieved 84.7% accuracy and an AUC of 0.900 for DMI detection.
- Performance significantly outperformed baseline CFP-trained models and human experts (P < 0.01).
Conclusions:
- AI-based analysis of CFPs shows significant potential for DMI detection.
- This approach offers a feasible, early, and cost-effective screening method for DMI.
- The study establishes a viable CFP-based screening method for DMI, potentially improving clinical outcomes.
