Related Experiment Video
Updated: Jul 18, 2026
09:17
Using Retinal Imaging to Study Dementia
Published on: November 6, 2017
21.6K
DiaNet v2 deep learning based method for diabetes diagnosis using retinal images
Hamada R H Al-Absi1, Anant Pai2, Usman Naeem2
1College of Science and Engineering, Hamad Bin Khalifa University, Doha, Qatar.
Scientific Reports
|January 18, 2024
Summary
A new deep learning model, DiaNet v2, uses retinal images for accurate diabetes mellitus diagnosis. This non-invasive method shows over 92% accuracy, offering a promising alternative to traditional tests, especially in the MENA region.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Diabetes mellitus (DM) is a widespread chronic metabolic disorder with significant morbidity and mortality.
- Undiagnosed diabetes cases are prevalent, particularly in the Middle East North Africa (MENA) region, necessitating improved diagnostic tools.
- Current diagnostic methods like FPG, OGTT, RPG, and HbA1c have limitations, including potential misclassification and patient discomfort.
Purpose of the Study:
- To enhance the accuracy of diabetes diagnosis by developing an advanced predictive model.
- To address the limitations of current diagnostic methods by utilizing retinal images.
- To create a more accessible and non-invasive approach for diabetes detection.
Main Methods:
- Development of the DiaNet v2 model, an enhanced deep learning system for diabetes detection based on retinal images.
- Utilizing a large dataset comprising 5545 participants (2540 diabetic, 3005 control) from Qatar Biobank (QBB) and Hamad Medical Corporation (HMC).
- Training and validation of the model on retinal images covering a wide range of pathologies.
Main Results:
- DiaNet v2 achieved an accuracy exceeding 92% in distinguishing diabetic patients from controls.
- The model demonstrated high sensitivity (93%) and specificity (91%).
- The study successfully leveraged a comprehensive retinal image dataset and deep learning for accurate diabetes diagnosis.
Conclusions:
- DiaNet v2 offers a highly accurate, non-invasive method for diabetes diagnosis using retinal images.
- This deep learning approach has the potential to revolutionize early diabetes detection and intervention planning.
- The model provides a valuable tool, particularly for regions like MENA with high diabetes prevalence and diagnostic challenges.
