This research aims to develop machine learning (ML) models for early detection of stroke and other neurological diseases (such as Alzheimer’s and Parkinson’s) using retinal fundus images from the UK Biobank. The core hypothesis is that retinal vascular patterns reflect systemic microvascular and neurodegenerative changes, making them effective non-invasive biomarkers.
Research Questions:
1. Can retinal fundus images accurately predict stroke risk using ML?
2. Are there shared retinal biomarkers for neurological diseases like Alzheimer’s and Parkinson’s?
3. How do retinal-based models compare with traditional clinical risk assessments?
Objectives:
1. Design image analysis techniques to extract stroke-related retinal biomarkers.
2. Train and validate ML models for stroke risk prediction.
3. Extend models to detect features linked to Alzheimer’s and Parkinson’s.
4. Benchmark performance using AUC, sensitivity, and specificity.
Scientific Rationale:
The retina offers a non-invasive window into vascular and neurological health. Retinal changes have been associated with both stroke and neurodegeneration. UK Biobank studies suggest retinal features can enhance stroke risk prediction. With advances in image analysis and ML, it is now feasible to extract and analyze these features at scale, enabling clinically relevant and cost-effective early detection tools.
I am a first-year Ph.D. student at the University of Delhi, India, supervised by Dr. Mantosh Biswas (email: [email protected]). I intend to apply for the Global Researcher Access Fund and Platform Credits Programme to support this project.
Data Requirements:
1. Imaging Data: Retinal fundus photographs (Category 100016)
2. Health Outcomes: Stroke (ICD-10: I60-I69), Parkinson’s (G20), Alzheimer’s (G30)
3. Covariates: Age, sex, blood pressure, cholesterol, smoking, diabetes