As part of my MSc in Computer Science, my research focuses on deep learning applications in retinal imaging for the early detection and preventative screening of Alzheimer’s Disease (AD). Alzheimer’s, the most common cause of dementia, affects millions globally. Traditional diagnostic methods, like MRI and PET scans, are often costly and inaccessible, particularly in under-resourced areas. Given this challenge, retinal imaging has emerged as a promising, non-invasive alternative due to the retina’s anatomical and physiological connection to the brain.
Project Overview
The purpose of my research is to conduct a systematic review of current deep learning models used in retinal imaging to detect Alzheimer’s-related retinal biomarkers. These biomarkers include retinal nerve fibre layer (RNFL) thinning, retinal vascular changes, and amyloid-beta deposition — all indicators potentially linked to AD. Through this review, I aim to evaluate the performance of these models, identify existing research gaps, and provide recommendations for standardising methodologies in this emerging field.
Literature Insights and Research Gaps
Deep learning, particularly Convolutional Neural Networks (CNNs) and hybrid models, has shown potential in detecting AD-related changes within retinal images. However, significant challenges remain:
- Data Standardisation: Variability in imaging quality across devices and patient demographics impacts the generalisability of deep learning models.
- Data Availability: Limited access to labelled datasets specific to AD-related retinal changes constrains model training and validation.
- Model Interpretability: Many models act as “black boxes,” where the decision-making process is unclear, which poses issues for clinical integration and acceptance.
Identifying these challenges has led to an exploration of future directions, such as longitudinal studies, multi-modal approaches, and collaborative databases that could enhance diagnostic accuracy and clinical applicability.
Research Questions and Methodology
To bridge the gap between current advancements and clinical application, my research addresses questions such as:
- What deep learning models are most effective for AD detection via retinal imaging?
- How do these models’ performance metrics, like accuracy and specificity, compare?
- What challenges and ethical considerations impact the practical deployment of these models?
Using a systematic review framework, I will synthesise findings across studies to evaluate these models comprehensively, following established PRISMA guidelines to ensure rigour and reproducibility. The goal is to propose guidelines for methodological standardisation and examine ethical, social, and regulatory considerations relevant to deploying AI-based diagnostic tools in real-world settings.
Conclusion
Ultimately, my research seeks to enhance the feasibility of using deep learning in retinal imaging for Alzheimer’s screening, providing insights that could contribute to the development of accessible and reliable early diagnostic tools. The findings aim to support both the clinical and ethical integration of AI in dementia diagnosis, potentially benefiting patients, caregivers, and healthcare systems worldwide.