Mining retinal color fundus photographs for Alzheimer's disease biomarker discovery
Approved Research ID: 91706
Approval date: May 9th 2023
Alzheimer's Disease (AD), the most common cause of dementia, has increasing prevalence with vast societal and public health implications. The diagnosis of AD is based on expensive and advanced ionizing neuroimaging or invasive procedures such as lumbar puncture and brain for Alzheimer's pathology. Retinal imaging has multiple advantages, such as safety, non-invasive manner, repeatability, and affordability on a large scale. Hence, retinal imaging has emerged as a powerful biomarker for AD that carries the promise of early AD detection and disease monitoring. The application of artificial intelligence modeling to retinal imaging promises to develop an automated biomarker for AD screening and tracking, it proved the feasibility of our method Based on our previous two experiments, the first one showed that specific amyloid deposits can be biomarkers for AD-related cognitive dysfunction. The second experiment has shown that our deep learning model can distinguish between data from AD patients from Mayo Clinic and cognitively unimpaired subjects from the EyePACS database, the network identifies retinal blood vessel branches with tortuosity change as the potential identifier of AD. In the proposed project, we aim to develop an automated, accurate tool based on non-invasive retinal imaging to diagnose and monitor a devastating and common neurodegenerative disorder. It will be a 3-year project. We hypothesize that our deep learning models will assist early AD diagnosis with retinal color fundus photographs (alone or combined with other neurodegenerative and vascular image biomarkers). This has the potential for routine and affordable application in clinical practice worldwide.