Research Questions: 1. Can deep learning (DL) models detect systemic diseases by using retinal images, demographic data, and risk factors? 2. Can DL models predict the future risk of developing systemic diseases using baseline retinal images and risk factors? 3. What specific retinal biomarkers are associated with systemic diseases?
Objectives: Aim 1. To develop DL models capable of detecting systemic diseases and predicting future disease risk by integrating retinal image-derived biomarkers with established risk factors. Aim 2. To investigate associations between retinal biomarkers extracted from retinal images and a broad spectrum of systemic diseases.
Scientific Rationale: The retina offers a unique, non-invasive window into systemic health. Sharing embryological origins and microvascular architecture with vital organs, retinal alterations often precede clinical manifestations of systemic diseases by years. Fundus photography (FP) captures vascular morphology changes linked to hypertension, diabetes, and stroke risk. Optical coherence tomography (OCT) reveals neuronal layer thinning associated with preclinical neurodegenerative diseases. These modalities provide highly accessible, automatable, and cost-effective tools for large-scale screening and risk stratification.
With the global burden of chronic diseases rising, innovative approaches for early detection and risk stratification are critical, especially since many conditions remain asymptomatic until irreversible damage occurs. Deep learning (DL) excels at identifying subtle, complex patterns beyond human capability directly from images.
This project will harness advanced AI techniques, designed to integrate multimodal data, and the advantages of UK Biobank with its unprecedented scale, longitudinal design, and rich linked health data to learn complex patterns indicative of systemic pathologies directly from retinal images.