Using high-resolution retinal images, a new AI model detects Alzheimer’s disease and mild cognitive impairment early, offering hope for timely, non-invasive and affordable dementia care.
Study: Early detection of dementia through retinal imaging and reliable AI. Image credits: bfk / Shutterstock
This is evident from a recent study published in the journal NPJ Digital Medicineresearchers have developed a deep learning algorithm that can analyze high-resolution images of the retinal vasculature to detect Alzheimer’s disease (AD) and mild cognitive impairment (MCI) at an early stage. This innovative approach could allow doctors to identify patients at high risk of dementia, allowing timely pharmacological intervention to slow the progression of the disease and its damaging effects.
The study, which used 5,751 images from 1,671 participants, found that this new method significantly outperformed conventional AD and MCI detection techniques. The method is not only accurate, but also cheap, fast and completely non-invasive, making it ideal for large-scale screening. Researchers suggest that further validation of this approach could revolutionize dementia care, enabling early detection and treatment at scale.
Background
Dementia is a progressive condition that primarily affects cognitive functions, including memory, information processing and reasoning, leading to a decrease in the ability to perform everyday tasks. Recent advances in medicine have extended human lifespan, contributing to a global rise in the number of dementia cases, which now affects more than 50 million people and is expected to increase in the coming years.
Despite extensive research, existing methods for detecting dementia, including biochemical testing and magnetic resonance imaging (MRI), are limited by high cost, invasiveness and the need for specialized facilities. Furthermore, these techniques often fail to detect early-onset Alzheimer’s disease (EOAD), limiting the opportunity for early intervention in high-risk individuals.
Recent research has shown a possible link between the retina – often called the ‘window to the brain’ – and neurodegenerative diseases such as AD. Clinical studies and histopathological reports have demonstrated unique retinal microvascular changes in AD patients. These findings have sparked interest in using advanced ophthalmic imaging techniques, such as optical coherence tomography angiography (OCTA), to identify retinal biomarkers associated with cognitive decline.
OCTA is an advanced imaging technique that allows rapid and non-invasive imaging of the retinal microvasculature, including even the smallest capillaries with a resolution of 5–6 μm. This technology provides detailed insight into the microvascular network and structure of the foveal avascular zone across different retinal layers, as well as the choroid.
By using OCTA data in combination with artificial intelligence (AI), researchers aim to create a cost-effective and scalable solution to identify individuals at risk of dementia, promote healthy aging and enable timely interventions.
About the study
In this study, the researchers developed and tested a new deep learning model called ‘Eye-AD’. The model is specifically designed to analyze OCTA images and identify patients with early Alzheimer’s disease or mild cognitive impairment. The model processes high-resolution data from different layers of the retina, including the superficial vascular complex (SVC), deep vascular complex (DVC) and choriocapillaris (CC), to detect patterns associated with cognitive decline.
The study data were collected from two primary cohorts – ROMCI (Retinal OCTA-based MCI detection) and ROAD (Retinal OCTA-based EOAD detection) – which consisted of retinal images centered on the fovea. The images captured information about the vascular structures of the retina and provided the dataset for model training and evaluation.
The Eye-AD model consists of two main parts: a convolutional neural network (CNN) for feature extraction and a graph neural network (GNN) for final predictions. These components work together to analyze the intricate relationships between the layers of the retina, allowing for a more comprehensive evaluation of cognitive function. The model was tested with multiple encoders, including ResNet and ConvNeXt, and ConvNeXt was ultimately chosen for its speed and consistent performance.
Study findings
The results of the study indicate that the Eye-AD model is very accurate and reliable. The performance exceeded that of traditional biochemical and MRI-based detection methods, with an AUC (Area Under the Curve) of 0.9355 for detecting EOAD and 0.8630 for MCI. The model’s strength lies in its accuracy, cost-effectiveness and ability to process data quickly, making it a practical option for large-scale screening.
The model interpretability analysis also highlighted that the deep vascular complex (DVC) played a more crucial role in detecting EOAD and MCI than the other retinal layers, suggesting that deeper retinal structures may be more affected by cognitive decline. Specifically, the DVC was found to contribute more to the model’s predictions, with an average importance score of 40% and 49% for EOAD and MCI cases, respectively. These findings suggest that changes in the deeper layers of the retina may provide insight into the mechanisms underlying dementia and its progression.
Furthermore, the study showed that differences in retinal structure were more pronounced in patients with EOAD than in patients with MCI, which likely reflects the more severe impact of Alzheimer’s disease on the retinal vasculature. These findings support the hypothesis that dementia-related changes are more noticeable in the deeper layers of the retina.
Conclusion
The Eye-AD model represents a significant advance in the early detection of dementia. The ability to non-invasively screen large populations using only high-resolution retinal images makes it an ideal tool for widespread cognitive health assessments. The model is not only accurate, but also widely applicable, making it possible to identify individuals with early-stage Alzheimer’s disease or MCI and enable timely interventions.
Although the model has shown promise, the researchers emphasize the need for further research to validate its performance in more diverse populations. Integrating other modalities, such as blood tests or cognitive assessments, could also increase the diagnostic power of the model. With continued development, the Eye-AD model has the potential to become a valuable resource for dementia screening and monitoring in the future, promoting healthier aging and improving patient outcomes.