The retinal microvasculature provides a direct, noninvasive representation of systemic vascular and neural integrity. Morphological and perfusion changes observed in fundus imagery reflect pathophysiological processes shared across ocular, cardiovascular, and neurodegenerative diseases. With the growing availability of large-scale retinal image datasets, machine learning offers a scalable means of quantifying these subtle alterations and identifying early biomarkers of systemic dysfunction.
The primary research question of this study is whether retinal fundus imaging can be reliably utilized to automatically detect common ocular pathologies and extract imaging biomarkers predictive of systemic and cardiovascular disease risk.
The research objectives are threefold:
Algorithmic development: Design and validate deep learning models for automated detection of diabetic retinopathy, glaucoma, and age-related macular degeneration from fundus images.
Biomarker extraction: Quantify microvascular and neurodegenerative imaging features-such as vessel caliber, tortuosity, and fractal dimension-and analyze their associations with systemic and cardiovascular health indicators.
Predictive evaluation: Assess the prognostic value of retinal biomarkers for future cardiovascular, cerebrovascular, and neurodegenerative outcomes, and compare model predictions against established clinical risk metrics.
To achieve these aims, we will curate a multimodal dataset linking retinal imaging with relevant clinical and metabolic records, implement optimized model architectures for disease detection, and conduct cross-sectional and longitudinal analyses to quantify predictive associations. The overarching rationale is to enable scalable, accessible, and early-stage diagnostics that integrate ocular imaging into precision medicine workflows.