Enhancing diabetic retinopathy detection through federated convolutional neural networks: Exploring different stages of progression
Diabetic Retinopathy (DR) is a usual diabetes consequence as well as the world's leading cause of vision loss. Proper identification and treatment are critical for preventing visual loss. This paper presents a comprehensive solution for DR identification utilizing deep learning algorithms. To r...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-05-01
|
| Series: | Alexandria Engineering Journal |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1110016825001930 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Diabetic Retinopathy (DR) is a usual diabetes consequence as well as the world's leading cause of vision loss. Proper identification and treatment are critical for preventing visual loss. This paper presents a comprehensive solution for DR identification utilizing deep learning algorithms. To reduce class imbalance and strengthen this model, we first prepare unbalanced data utilizing the Synthetic Minority Oversampling Technique (SMOTE). In addition, Weiner and median filtration(weinmed) reduce turbulence in retinal images, improving the image quality for future investigations. The VGG was created primarily for segmentation and classification and is well-known for its ability to gather fine visual details. The obtained characteristics are incorporated into several CNN (convolutional neural network) models, which generate discriminatory patterns to determine DR severity levels. Federated learning within this plan allows shared instruction across many medical institutions while protecting patient privacy and security. The combination of SMOTE technology data pre-processing, filtering approaches, VGG-based segmentation, federated learning, and CNN (FedCNN) categorization produces encouraging results, demonstrating the feasibility of scalable and secure DR detection systems. The VGG was created primarily for feature extraction and is well-known for its ability to gather fine visual details. The obtained characteristics are incorporated into several CNN (convolutional neural network) models, which generate discriminatory patterns to determine DR severity levels. Also included in this plan is federated learning, allowing shared instruction across many medical institutions while protecting patient privacy and security. |
|---|---|
| ISSN: | 1110-0168 |