Non-Invasive AI-Based Retinal Inflammation Detection and Severity Estimation Using OCT B-Scans

Abstract

Researchers at the University of California, Davis have developed a machine learning system that accurately detects and estimates retinal inflammation severity in uveitis patients using non-invasive OCT B-scan images.

Full Description

This technology utilizes a deep learning model trained on OCT B-scan images aligned with fluorescein angiography (FA) to detect presence and severity of retinal inflammation related to uveitis. Leveraging a novel annotation pipeline and explainable AI techniques such as Grad-CAM, it offers a rapid, non-invasive, and cost-effective alternative to invasive FA procedures, providing clinician-level diagnostic support for improved and accessible patient care.

Applications

• Clinical ophthalmology decision-support tools for uveitis diagnosis and monitoring. 
• Integration in OCT imaging platforms and cloud-based diagnostic software. 
• Telemedicine platforms enabling remote inflammation assessment. 
• Pharmaceutical clinical trials requiring standardized inflammatory severity measures. 
• Screening programs in low-resource and pediatric healthcare settings.

Features/Benefits

• Detects retinal inflammation non-invasively without fluorescein dye injection. 
• Provides rapid, automated analysis from standard OCT scans. 
• Delivers high diagnostic accuracy, outperforming fellowship-trained clinicians. 
• Quantifies severity with strong correlation to ground truth standards. 
• Enhances trust and interpretability with explainable AI insights (e.g., Grad-CAM heatmaps). 
• Enables scalable adoption in telemedicine and low-resource settings. 
• Problems Solved Eliminates the invasiveness, cost, and limited accessibility of fluorescein angiography.
• Identifies subtle signs of retinal inflammation that are difficult to detect clinically. 
• Standardizes and quantifies inflammation severity grading. 
• Expands diagnostics where fluorescein angiography is contraindicated or unavailable, including for pediatric and remote care. 
• Accelerates diagnosis and treatment decisions in uveitis management.

Patent Status

Patent Pending