An international team led by The Chinese University of Hong Kong (CUHK)’s Faculty of Medicine (CU Medicine) has successfully developed the world’s first artificial intelligence (AI) model that can detect Alzheimer’s disease solely through fundus photographs or images of the retina. The model is more than 80% accurate after validation.
Fundus photography is widely accessible, non-invasive and cost-effective. This means that the AI model incorporated with fundus photography is expected to become an important tool for screening people at high risk of Alzheimer’s disease in the community. Details have been published in The Lancet Digital Health under the international journal The Lancet.
Limitations of Alzheimer’s disease current detection methods
In Hong Kong, 1 in 10 people aged 70 or above suffers from dementia, with more than half of those cases attributed to Alzheimer’s disease. This disease is associated with an excessive accumulation of abnormal amyloid plaque and neurofibrillary tangles in the brain, leading to the death of brain cells and resulting in progressive cognitive decline.
The Clinical Professional Consultant of the Division of Neurology in CU Medicine’s Department of Medicine and Therapeutics stated that memory complaints are common among middle-aged and elderly people, and are often considered a sign of Alzheimer’s disease.
It is sometimes difficult to make an accurate diagnosis of Alzheimer’s disease based on cognitive tests and structural brain imaging. However, methods to detect Alzheimer’s pathology, such as an amyloid-PET scan or testing of cerebrospinal fluid collected via lumber puncture, are invasive and less accessible.
To address the current clinical gap, CU Medicine has led several medical centres and institutions from Singapore, the United Kingdom and the United States to successfully develop an AI model using state-of-the-art technologies which can detect Alzheimer’s disease using fundus photographs alone.
Studying disorders of the central nervous system via the retina
The S.H. Ho Professor of Ophthalmology and Visual Sciences and Chairman of CU Medicine’s Department of Ophthalmology and Visual Sciences explained that the retina is an extension of the brain in terms of embryology, anatomy and physiology. In the entire central nervous system, only the blood vessels and nerves in the retina allow direct visualisation and analysis.
Thus, it is widely considered a window through which disorders in the central nervous system can be studied. Through non-invasive fundus photography, a range of changes in the blood vessels and nerves of the retina that are associated with Alzheimer’s disease can be detected.
The team developed and validated their AI model using nearly 13,000 fundus photographs from 648 Alzheimer’s disease patients (including patients from the Prince of Wales Hospital) and 3,240 cognitively normal subjects. Upon validation, the model showed 84% accuracy, 93% sensitivity and 82% specificity in detecting Alzheimer’s disease. In the multi-ethnic, multi-country datasets, the AI model achieved accuracies ranging from 80% to 92%.
Accessibility, non-invasiveness and high cost-effectiveness of the AI model using fundus photography help the detection of Alzheimer’s cases both in the clinic and the community
A Professor of Medicine and Director of the Therese Pei Fong Chow Research Centre for Prevention of Dementia at CU Medicine stated that in addition to its accessibility and non-invasiveness, the accuracy of the new AI model is comparable to imaging tests such as magnetic resonance imaging (MRI).
It shows the potential to become not only a diagnostic test in clinics but also a screening tool for Alzheimer’s disease in community settings. Looking ahead, the team aims to validate its efficacy in identifying high-risk cases of the disease hidden in the community, so that various preventive treatments such as anti-amyloid drugs can be initiated early to slow down cognitive decline and brain damage.
The Associate Professor in the Department of Ophthalmology and Visual Sciences at CU Medicine said that in addition to applying novel AI technologies in the model, the team also tested it in different scenarios. Notably, their AI model retained a robust ability to differentiate between subjects with and without Alzheimer’s disease, even in the presence of concomitant eye diseases like macular degeneration and glaucoma which are common in city-dwellers and the older population.
Their results further support the hypothesis that the team’s AI analysis of fundus photographs is an excellent tool for the detection of memory-depriving Alzheimer’s disease. To move this research towards clinical application, the team is developing an integrated, AI-based platform to combine information from both blood vessels and nerves of the retina captured by fundus photography and optical coherence tomography for the detection of Alzheimer’s disease. Their findings should provide more evidence to move AI from code to the real world.