Quantitative Analysis of Retinal Fluid by a Deep Learning Model in Uveitic Macular Edema

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Uveitic macular edema (UME) poses a significant threat to visual acuity.
• Accurate prognostic markers are crucial for managing UME.
• Current methods like central macular thickness (CMT) may not fully capture disease complexity.

Purpose of the Study:

• To evaluate the prognostic value of AI-derived fluid volumes in UME.
• To compare the predictive performance of AI fluid measurements against CMT alone.
• To assess the utility of AI in quantifying intraretinal fluid (IRF) and subretinal fluid (SRF) in UME.

Main Methods:

• A deep learning model (2D U-Net) was trained to segment IRF and SRF in OCT scans.
• The model was applied to patients with UME from the FAST clinical trial.
• Statistical models assessed the association of baseline fluid volumes and CMT with visual acuity (VA) change.

Main Results:

• AI model achieved good segmentation accuracy for IRF and SRF.
• Baseline IRF volume showed a significant interaction with treatment assignment, impacting VA change.
• AI-derived fluid volumes (IRF and SRF) improved prognostic models compared to CMT alone.

Conclusions:

• AI-based fluid segmentation provides quantitative measurements for UME.
• AI-derived fluid volumes offer additional prognostic information for visual outcomes in UME.
• Integrating AI fluid analysis into clinical workflows and trials is recommended.