Related Experiment Video
Updated: Jul 9, 2026
09:17
Using Retinal Imaging to Study Dementia
Published on: November 6, 2017
22.4K
Enhancing fundus image analysis for diabetic retinopathy using CheXNet with CBAM and Grad-CAM visualization
Wedad Al-Dolat1, Salem Alhatamleh2, Noor Alqudah3
1Department of Ophthalmology, Faculty of Medicine, Yarmouk University, Irbid, Jordan.
Frontiers in Medicine
|March 13, 2026
Summary
A new deep learning model significantly improves diabetic retinopathy (DR) classification accuracy from fundus images. This AI framework enhances early detection, aiding in timely clinical management for diabetic eye disease.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Diabetic retinopathy (DR) is a primary cause of vision loss in diabetic patients.
- Early DR detection and grading are crucial for effective clinical management.
- Automated fundus image analysis for DR faces challenges due to image quality and lesion variability.
Purpose of the Study:
- To develop a deep learning framework for automated diabetic retinopathy classification.
- To enhance feature representation in fundus images using attention mechanisms.
- To provide visual explanations for model predictions.
Main Methods:
- A DenseNet121 backbone initialized with CheXNet weights was utilized.
- A Convolutional Block Attention Module (CBAM) was integrated for improved feature learning.
- Gradient-weighted Class Activation Mapping (Grad-CAM) was employed for model interpretability.
Main Results:
- The proposed CheXNet_CBAM model achieved 96.12% accuracy on the APTOS 2019 dataset.
- The model attained 96.33% accuracy on the DDR dataset, outperforming other CNN architectures.
- Integration of CBAM enhanced discriminative feature learning.
Conclusions:
- The deep learning framework with CBAM shows high performance in DR classification.
- The model's interpretability is enhanced through Grad-CAM visualization.
- Further validation is needed for real-world clinical application.
