AI-based Alzheimer’s Detection via Retinal OCT Imaging
Original Title: Biomarkers
Journal: Alzheimer's & dementia : the journal of the Alzheimer's Association
Overview
Alzheimer's disease presents a significant global health challenge, with early detection being a priority for effective intervention. This study investigates the use of retinal optical coherence tomography (OCT) as a non-invasive biomarker for Alzheimer's disease. The researchers developed deep learning models designed to analyze both en face images and conventional analysis reports, including retinal nerve fiber layer (RNFL), macular thickness, and ganglion cell-inner plexiform layer (GCIPL) data. The primary dataset for training and internal validation consisted of 3,228 paired OCT reports and images from 1,239 subjects, comprising individuals with Alzheimer's dementia and cognitively normal controls. Performance was evaluated using internal validation and two external datasets from Hong Kong and Singapore. The results indicated that the "Ensemble Model" achieved high performance, with accuracies of 90.5% in internal validation, 80.3% in the first external test, and 74.2% in the second external test. These findings suggest that retinal imaging combined with artificial intelligence can assist in identifying cognitive decline.
Novelty
The technical approach introduces a multi-stage fusion network architecture that incorporates "Feature Extraction", "Feature Fusion", and "Feature Reconstruction" modules. Unlike previous methods that might focus on a single type of retinal image, this methodology integrates multiple inputs from a single eye, specifically combining optic nerve head-centered and macula-centered data. The implementation of an ensemble learning technique to synthesize these diverse inputs into a unified model represents a progression in the application of artificial intelligence to ophthalmic imaging for neurodegenerative disease. By leveraging both raw en face images and processed thickness and deviation maps from conventional OCT reports, the system captures a broader spectrum of Alzheimer's-related retinal changes than models relying on localized data alone. This comprehensive integration of standard clinical outputs with raw image data allows the model to utilize a variety of spatial and quantitative features simultaneously.
Potential Clinical / Research Applications
The potential clinical and research applications of this technology are centered on the development of accessible screening tools for cognitive impairment. In a clinical setting, this AI-driven approach could serve as a non-invasive preliminary test to identify individuals who require more intensive diagnostic procedures, such as positron emission tomography or cerebrospinal fluid analysis. Researchers could utilize these models to monitor disease progression or evaluate the efficacy of therapeutic interventions in clinical trials. Furthermore, because OCT is widely available in optometry and ophthalmology clinics, this method could be integrated into routine eye exams to facilitate earlier identification of at-risk individuals. The ability to detect Alzheimer's-related changes through the eye offers a cost-effective alternative to traditional neuroimaging, potentially broadening the reach of diagnostic services in diverse healthcare environments and providing a scalable solution for population-level screening.