Summary: Researchers have developed a promising, non-invasive retinal imaging method that uses polarized light and artificial intelligence to distinguish between Alzheimer’s disease and TDP-43–related neurodegenerative disorders such as Amyotrophic Lateral Sclerosis (ALS) and Frontotemporal Lobar Degeneration (FTLD-TDP). By analyzing how protein deposits in the retina scatter polarized light, the team can reliably identify whether deposits are amyloid beta—linked to Alzheimer’s—or TDP-43, which is characteristic of ALS and FTLD. The technique, combined with machine learning, reached diagnostic accuracies up to 96% in their tests.
This approach promises a fast, affordable screening tool for early detection of neurodegenerative disease biomarkers in the eye, potentially enabling diagnosis years before clinical symptoms appear. Because the retina is an accessible extension of the central nervous system, retinal imaging using polarized light could expand access to advanced diagnostics in primary care and underserved communities.
Key Facts
- Protein fingerprinting with polarized light: The method distinguishes amyloid beta (Alzheimer’s) from TDP-43 (ALS/FTLD) by measuring distinct polarized light scattering patterns from protein deposits in retinal tissue.
- AI-powered classification: The study tested two machine learning approaches: Random Forest, which reached about 86% accuracy, and convolutional neural networks (CNNs), which achieved approximately 96% accuracy in classifying deposit type.
- Clinical relevance: Beyond identifying the protein type, the polarized-light signature correlated with the severity of brain pathology, indicating potential for both early detection and disease staging.
Source: University of Waterloo
A retinal image could provide a rapid and non-invasive way for clinicians to distinguish among similar neurodegenerative diseases, including Alzheimer’s disease, ALS, and frontotemporal lobar degeneration, with high accuracy.
A research team at the University of Waterloo has advanced a diagnostic approach that images donated retinal samples with polarized light to reveal unique optical signatures produced by protein deposits. These signatures allow automated systems to separate amyloid beta deposits—commonly associated with Alzheimer’s disease—from deposits of the protein TDP-43, which are linked to ALS and FTLD-TDP. The technique could give clinicians a practical tool for screening and differential diagnosis that does not rely on invasive testing or expensive PET imaging.
Early and accurate diagnosis can enable timely interventions that may slow progression and support development of targeted therapies. Detecting disease-associated protein deposits in the retina could change clinical pathways by identifying at-risk individuals before substantial brain damage occurs.
“This is a major step toward earlier and more accurate diagnosis,” said Dr. Melanie Campbell, professor emeritus of physics and optometry. “Currently FTLD and ALS are often only identified after symptoms appear, when disease may be advanced. Detecting these conditions earlier could fundamentally change how we manage and treat them.”
The research team imaged flat-mounted retinas taken post-mortem from individuals diagnosed with Alzheimer’s, FTLD-TDP, and ALS. Using a polarimeter, they recorded multiple polarized light interaction properties across hundreds of presumed protein deposits—270 presumed amyloid beta deposits and 138 presumed TDP-43 deposits—then analyzed those properties directly and as images with machine learning models.
Distinct differences in nine measured polarimetric properties allowed the Random Forest classifier to reach about 86.5% accuracy using a subset of those features. Convolutional neural networks trained on images of three polarimetric property distributions exceeded 96% classification accuracy, demonstrating strong potential of image-based AI for deposit-type identification.
The team envisions adapting this technology into a clinical screening test over the coming years. Because it relies on light-based imaging and AI rather than invasive sampling or high-cost scans, the approach could be deployed in eye clinics and community health settings, increasing diagnostic reach.
Key Questions Answered:
A: The retina is part of the central nervous system and can reflect pathological changes that occur in the brain and spinal cord. Protein aggregates that form in neurodegenerative disease can appear in retinal tissue, offering a visible biomarker for brain pathology.
A: Proteins like amyloid beta and TDP-43 interact differently with polarized light because of differences in structure and how they scatter light. These interactions create reproducible optical signatures that AI models can detect and learn to classify.
A: Researchers aim to refine and validate this approach further and hope it could become a practical clinical screening tool within a few years. Its low cost and non-invasive nature make it suitable for broader deployment in primary care settings once validated in larger clinical studies.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- The full journal paper was reviewed for accuracy.
- Additional explanatory context was added by staff to clarify methods and implications.
About this visual neuroscience and neurodegeneration research news
Author: Pamela Smyth
Source: University of Waterloo
Contact: Pamela Smyth, University of Waterloo
Image: Image credited to Neuroscience News
Original Research: Retinal Deposits of TDP-43 and Amyloid Beta and Associated Neurodegenerative Diseases are Accurately Classified using Measured Interactions with Polarized Light in Machine Learning Algorithms. Authors: Melanie CW Campbell, Lyndsy Acheson, Erik L Mason, Tanya Hareesha Shetty, Laura Emptage, Rachel Redekop, Monika Kitor, Ian R MacKenzie, Naomi C Futhey, Veronica Hirsch-Reinshagen, Ging-Yuek Robin Hsiung. Journal: Alzheimer’s & Dementia. DOI: 10.1002/alz70861_108465. (Open access)
Abstract
Retinal Deposits of TDP-43 and Amyloid Beta and Associated Neurodegenerative Diseases are Accurately Classified using Measured Interactions with Polarized Light in Machine Learning Algorithms
Background
This study demonstrates that polarized-light interactions differ consistently between retinal amyloid beta deposits (linked to Alzheimer’s disease) and retinal deposits of TDP-43 (found in FTLD and ALS). Leveraging polarimetric imaging of the retina offers a potential non-invasive differential diagnostic for these neurodegenerative diseases. The polarized-light signatures were evaluated with machine learning to classify deposit types.
Method
Post-mortem eyes and brain tissue came from participants with ALS (including one with concurrent FTLD) and multiple FTLD cases, alongside individuals with brain amyloid and tau consistent with Alzheimer’s disease. Researchers imaged flat-mounted retinas with a polarimeter and used thioflavin fluorescence to support deposit identification. They imaged 270 presumed amyloid beta deposits and 138 presumed TDP-43 deposits, then quantified nine polarimetric properties. Machine learning classifiers—Random Forest and convolutional neural networks—were trained to distinguish the deposit types using both statistical property values and images representing polarimetric distributions.
Result
Significant differences were found across multiple polarized-light interaction measures between presumed TDP-43 and amyloid beta deposits. Using borderline-SMOTE augmentation, Random Forest classification achieved approximately 86.5% accuracy with six selected polarimetric features. CNNs trained on images of three polarimetric distributions achieved greater than 96% classification accuracy.
Conclusion
Machine learning applied to polarized-light measurements of retinal deposits can differentiate Alzheimer’s-associated amyloid beta from TDP-43 deposits characteristic of ALS and FTLD with high accuracy. This non-invasive, early, and cost-effective approach has the potential to become a practical differential diagnostic tool, improving access to screening and early intervention for diverse patient populations.