Summary: Researchers are transforming the “window to the soul” into a window for diagnosis. By using polarized light to image protein deposits in the retina, the team can now distinguish between Alzheimer’s, ALS, and Frontotemporal Lobular Dementia (FTLD-TDP) with startling accuracy. While Alzheimer’s is marked by amyloid beta, ALS and FTLD are characterized by TDP-43 deposits.
This non-invasive eye test uses AI models to identify unique light patterns reflected by these proteins, achieving up to 96% accuracy. This breakthrough offers a fast, affordable way to detect these diseases years before traditional symptoms appear, potentially bringing high-level diagnostics to underserved communities.
Key Facts
- Protein Fingerprinting: The tool uses polarized light to differentiate between amyloid beta (Alzheimer’s) and TDP-43 (ALS/FTLD) based on how the deposits scatter light.
- AI Precision: Two AI models were tested: Random Forest predicted the correct disease with 86% accuracy, while convolutional neural networks (CNN) reached a staggering 96%.
- Predictive Power: Beyond just naming the disease, the light patterns in the retina were able to predict the severity of protein deposits in the patient’s brain.
Source: University of Waterloo
A retinal image could help doctors quickly distinguish between similar neurodegenerative diseases such as ALS and Alzheimer’s disease, and with remarkable accuracy, according to new research.
Researchers from the University of Waterloo are leading the development of a fast, non-invasive, and affordable diagnostic tool. There is currently no objective diagnostic test for ALS or Frontotemporal Lobular Dementia (FTLD-TDP), in which the protein TDP-43 forms deposits in the spinal cord and brain, respectively.
The new tool could help with earlier diagnosis, enabling timely interventions known to slow disease progression and supporting more targeted treatment development that could improve patient outcomes.
“This is a major step toward earlier and more accurate diagnosis,” said Dr. Melanie Campbell, professor emeritus of physics and optometry. “Right now, FTLD and ALS are diagnosed only after symptoms appear, which often means the disease is already advanced. Being able to detect these conditions earlier could transform how we treat them.”
The researchers used polarized light to image protein deposits in donated retinal samples from patients with Alzheimer’s and compared them to samples from patients with Frontotemporal Lobular Dementia with TDP-43 (FTLD-TDP) and Amyotrophic Lateral Sclerosis (ALS).
By analyzing the light patterns, the team was able to accurately differentiate between the deposits of amyloid beta, associated with Alzheimer’s, from those with TDP-43, associated with FTLD and ALS, and predict the severity of the disease in the brain.
They also uploaded data on the light interaction into two AI models, Random Forest, an ensemble learning method, and convolutional neural networks, an image-based method, to see if they could learn to differentiate amyloid beta from TDP-43 deposits.
The differences between the two deposit types were strong enough that Random Forest predicted the right disease 86 per cent of the time, and convolutional neural networks were even higher at 96 per cent of the time.
“We hope that within a few years, this technology will evolve into a simple eye test capable of detecting and distinguishing multiple brain diseases, giving patients in smaller, underserved communities access to this type of testing,” Campbell said. “A fast, accessible diagnostic tool could make a profound difference for patients and families.”
Key Questions Answered:
A: The retina is actually an extension of the central nervous system. When proteins like TDP-43 or amyloid beta start gunking up the brain, they often show up in the eye first. It’s like a “biological preview” of what’s happening in the spinal cord and brain.
A: Different proteins are shaped differently. Amyloid beta and TDP-43 reflect polarized light in distinct patterns—almost like a unique optical signature. The AI is trained to recognize these “light fingerprints” that the human eye can’t see.
A: Researchers hope to evolve this into a standard eye test within a few years. Because it uses light and AI rather than expensive PET scans or invasive spinal taps, it could eventually be available at a local optometrist’s office.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- Journal paper reviewed in full.
- Additional context added by our staff.
About this visual neuroscience and neurodegeneration research news
Author: Pamela Smyth
Source: University of Waterloo
Contact: Pamela Smyth – University of Waterloo
Image: The image is credited to Neuroscience News
Original Research: Open access.
“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” by 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. Alzheimer’s & Dementia
DOI:10.1002/alz70861_108465
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
We have shown that interactions with polarized light differed significantly between retinal amyloid beta deposits, associated with Alzheimer’s disease, and deposits of TDP-43 found in the neurodegenerative diseases Frontotemporal Lobular Dementia (FTLD) and Amyotrophic Lateral Sclerosis (ALS). Our non-invasive retinal imaging could be the first differential diagnostic of these neurodegenerative diseases. Here, the deposits’ polarized light interactions are used in machine learning to classify the deposits.
Method
Post-mortem eyes and brains were obtained from 2 individuals with ALS, 1 of whom also had FTLD, and 4 individuals with FTLD, including 1 with Type C. Brain TDP-43 was present in the FTLD cases and some had age-related tau. Flat-mounted retinas were imaged using a polarimeter and then in thioflavin fluorescence. 270 presumed amyloid beta deposits in 10 individuals who had brain amyloid beta and tau and a moderate to high likelihood of AD, and 138 presumed TDP-43 deposits in those with FTLD and/or ALS were imaged.
In 1 individual with concurrent low values of brain amyloid and 1 with FTLD-Type C, only thioflavin negative deposits were classed as potential TDP-43 deposits. Interactions of polarized light with deposits were analyzed. Random forest (RF), an ensemble learning method and convolutional neural networks (CNN), were then used to differentiate amyloid beta from TDP-43 deposits.
Result
The deposit means and/or standard deviations of nine different polarized light interactions were significantly different between the presumed TDP-43 retinal deposits, found in ALS and FTLD, and amyloid deposits found in AD. With borderline SMOTE augmentation of the data, we achieved a classification accuracy of 86.5± 0.6% using random forest, utilizing 6 of these interactions. CNN achieved a classification accuracy of >96% accuracy using images of the distributions of 3 polarimetric properties.
Conclusion
Machine learning, using the averages and standard deviations of polarized light properties in RF or images showing the distribution of these properties across the deposits (CNN) can differentiate retinal deposits associated with Alzheimer’s disease from those associated with ALS and FTLD, with a relatively high accuracy. This first differential diagnostic of Alzheimer’s disease from TDP-43 related diseases, is early, non-invasive and inexpensive and would reach underserved populations.
