Related Experiment Video
Updated: Jun 19, 2026
07:44
In vivo Structural Assessments of Ocular Disease in Rodent Models using Optical Coherence Tomography
Published on: July 24, 2020
2.8K
Distributed training of foundation models for ophthalmic diagnosis
Sina Gholami1, Fatema-E Jannat1, Atalie Carina Thompson2
1Department of Electrical and Computer Engineering, University of North Carolina at Charlotte, Charlotte, NC, USA.
Communications Engineering
|January 22, 2025
Summary
A new deep learning framework improves early detection of eye diseases like diabetic retinopathy using federated learning. This privacy-preserving method enhances diagnostic accuracy across diverse populations without sharing patient data.
Area of Science:
- Ophthalmology
- Artificial Intelligence
- Medical Imaging
Background:
- Vision impairment affects 2.2 billion globally, with many cases preventable through early diagnosis.
- Diabetic retinopathy and age-related macular degeneration require scalable detection methods.
- Optical coherence tomography (OCT) is crucial for diagnosing retinal diseases.
Purpose of the Study:
- To develop a distributed deep learning framework for enhanced eye disease detection from OCT images.
- To integrate self-supervised and domain-adaptive federated learning for improved diagnostic accuracy and generalization.
- To enable collaborative model development while preserving data privacy.
Main Methods:
- Utilized a self-supervised, mask-based pre-training strategy for a robust foundation encoder.
- Trained the encoder on seven diverse OCT datasets.
- Compared performance across local, centralized, and federated learning settings, incorporating domain adaptation.
Main Results:
- Self-supervised methods (centralized and federated) improved the area under the curve by at least 10% compared to local models.
- Domain-adaptive federated learning enhanced performance and generalization across different populations and imaging conditions.
- The framework demonstrated effective collaborative model development without direct data sharing.
Conclusions:
- The proposed distributed deep learning framework offers a scalable, privacy-preserving solution for retinal disease screening.
- Self-supervised and federated learning significantly improve the detection of eye diseases from OCT images.
- Domain adaptation further boosts the model's ability to generalize in diverse clinical settings.
