Related Experiment Video
Updated: Jul 26, 2026
07:35
In Vivo Vascular Injury Readouts in Mouse Retina to Promote Reproducibility
Published on: April 21, 2022
2.2K
Using Deep Learning to Segment Retinal Vascular Leakage and Occlusion in Retinal Vasculitis
Dhanach Dhirachaikulpanich1,2,3, Jianyang Xie1, Xiuju Chen4
1Department of Eye & Vision Sciences, University of Liverpool, Liverpool, UK.
Ocular Immunology and Inflammation
|January 23, 2024
Summary
Deep learning models can reliably segment retinal vascular leakage and occlusion in fluorescein angiography (FA) images for patients with retinal vasculitis (RV). This advancement aids in diagnosing and managing RV by accurately identifying key pathological features.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Retinal vasculitis (RV) presents with retinal vascular leakage or occlusion, detectable via fluorescein angiography (FA).
- Current methods lack a standardized approach for segmenting these RV features in FA images.
Purpose of the Study:
- To develop and evaluate a deep learning model for the segmentation of retinal vascular leakage and occlusion in RV patients.
Main Methods:
- A deep learning model was developed using 463 FA images from 82 RV patients.
- The dataset was split into training (60%), validation (20%), and testing (20%) sets.
- Various deep learning architectures (DeeplabV3+, UNet++, UNet) were assessed for optimal binary segmentation of leakage and occlusion, with Dice score used for reliability evaluation.
Main Results:
- The best model for vascular leakage achieved a Dice score of 0.6279 (95% CI 0.5584-0.6974).
- The optimal model for occlusion yielded a Dice score of 0.6992 (95% CI 0.6109-0.7874).
Conclusions:
- The developed deep learning models demonstrate reliable performance in segmenting retinal vascular leakage and occlusion in FA images of RV patients.
- This technology offers a potential tool for improved diagnosis and monitoring of retinal vasculitis.
