Retinal image analytics detects white matter hyperintensities in healthy adults

Area of Science:

• Ophthalmology
• Neurology
• Artificial Intelligence

Background:

• White matter hyperintensities (WMH) are markers of cerebral small vessel disease.
• Early detection of WMH is crucial for preventing stroke and dementia.
• Current detection methods like MRI are costly and not widely accessible for screening.

Purpose of the Study:

• To investigate the efficacy of automatic retinal image analysis (ARIA) using machine learning to detect high WMH burden in asymptomatic older adults.
• To compare ARIA's performance against MRI, the gold standard for WMH assessment.

Main Methods:

• A cross-sectional study involving 180 community-dwelling, stroke- and dementia-free older adults.
• Retinal fundus images were acquired for ARIA analysis.
• Machine learning models were developed to estimate WMH and classify its severity, with MRI serving as the gold standard.

Main Results:

• ARIA demonstrated excellent diagnostic performance for detecting severe WMH (age-related white matter changes grade ≥2), with high sensitivity (0.929) and specificity (0.984).
• A strong correlation (0.897) was observed between WMH volumes measured by MRI and those estimated by ARIA.
• The study included individuals with common vascular risk factors like hypertension, diabetes, and hyperlipidemia.

Conclusions:

• A robust algorithm using ARIA can automatically identify individuals with a high WMH burden from retinal images.
• ARIA shows promise as a non-invasive, accessible tool for early screening of cerebral small vessel disease.
• Further community-based prospective studies are recommended to validate ARIA for population-level screening.