Deep Learning-Based Quantification of Vitreous Hyperreflective Foci as a Biomarker for Intraocular Inflammation

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Vitreous hyperreflective foci (vHRF) are indicators of intraocular inflammation (IOI).
• Accurate quantification of vHRF is crucial for assessing IOI.
• Current methods for vHRF quantification can be time-consuming and subjective.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI)-driven pipeline for quantifying vHRF from Optical Coherence Tomography (OCT) images.
• To assess the association between AI-quantified vHRF metrics and intraocular inflammation (IOI).

Main Methods:

• A deep learning segmentation model (U-Net Transformer with Vision Transformer backbone) was trained on OCT B-scans with expert vHRF annotations.
• The model was applied to a clinical analysis cohort from the GALLEGO trial (NCT03972709).
• Quantitative vHRF metrics were evaluated for their association with clinically diagnosed concurrent IOI using various statistical analyses.

Main Results:

• The AI model demonstrated strong segmentation performance for vHRF.
• Eyes with IOI showed significantly elevated vHRF metrics (P < 0.001).
• vHRF volume density was the strongest biomarker for IOI detection (AUC = 0.84), with significant associations found in logistic regression models.

Conclusions:

• The AI-driven pipeline accurately quantifies vHRF from OCT images.
• vHRF volume density is a significant imaging biomarker for concurrent IOI.
• Automated vHRF quantification shows promise for inflammation assessment, warranting further validation.