A joint Singapore-UK-Hong Kong research team has published details of an AI-based deep learning model that will help the early detection of Alzheimer’s disease using retinal photographs. The team trained a model by enrolling 648 people with Alzheimer’s and 3,240 without, taking over 12,000 retinal photographs. The model achieved an accuracy of 83.6%, 93.2% sensitivity, 82.0% specificity, and an area under the receiver-operating-characteristic curve of 0.93 (perfect score being 1).
Model Performance and Applications
In conclusion, the researchers determined that they had achieved proof of concept in relation to their model, which they note is the first deep learning model to detect Alzheimer’s disease from retinal photographs alone. They hope that the technology will help to address current issues of under-diagnosis surrounding Alzheimer’s disease, which is currently screened for using a complex series of cognitive tests, clinical assessments, and supportive evidence from neuroimaging (e.g., PET), and cerebrospinal fluid biomarker evidence.
Notably, the deep learning model was able to differentiate between those who were amyloid β positive from those who were amyloid β negative, performed well in patients with other eye diseases such as age-related macular degeneration, and could also be used based on retinal photographs from just one eye, which is necessary in patients with glaucoma, for example. The expectation is that the technology could be applied during screening in optometry and ophthalmology settings and primary healthcare facilities to improve rates of Alzheimer’s diagnosis.
Research Team and Publication
Researchers hailed from the Chinese University of Hong Kong, Raffles Neuroscience Centre in Singapore, Singapore’s National University Health System, the Royal Victoria Hospital, and the Queen’s University Belfast, UK, among others. The findings were published in The Lancet Digital Health in a report titled “A deep learning model for detection of Alzheimer’s disease based on retinal photographs: a retrospective, multicentre case-control study.”-Fineline Info & Tech
