Rationale
Retinal imaging non-invasively captures microvascular and neurological features associated with neurocognitive disorders such as Alzheimer’s disease and vascular dementia, as well as systemic conditions like dyslipidemia or hypertension, which are established modifiable risk factors for cognitive decline. Optical coherence tomography (OCT) and color fundus photography (CFP) are widely deployed modalities for monitoring ocular health. Their broad clinical adoption and widespread availability position them as viable tools for early disease detection. By leveraging large unannotated datasets for pretraining, followed by fine-tuning with smaller labeled datasets, vision transformer models coupled with multi-task learning (MTL) can generalize well to unseen data, making them ideal candidates for analyzing retinal images in the context of neurocognitive decline.
Research Question
Based on this context, we propose the following research question: Can retinal images combined with genetic and clinical data, accurately predict the risk of neurocognitive disorders?
Objectives
Our objective is to integrate retinal imaging data, genetic information, and clinical outcomes, to provide a robust, non-invasive tool to assess cognitive impairment. This MTL system will primarily assess neurocognitive disease risk but is hypothesized to perform optimally when concurrently trained to detect modifiable risk factors and monogenic variants associated with these risk factors and neurocognitive disorders.
A 2024 Lancet report (Livingston) indicated that modifying risk factors may prevent or delay nearly half of dementia cases, and as such there is value in presenting individuals with both their neurocognitive disorder risk, alongside their risk of established modifiable risk factors to motivate change. Developing a non-invasive, accessible tool for early risk stratification may empower individuals to proactively manage and reduce their risk of neurocognitive disorders.