Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Hyperspectral retinal imaging to detect Alzheimer’s disease in a memory clinic setting

Dallora, Ana Luiza ; Alexander, Jan ; Palesetti, Pushpa Priyanka ; Guenot, Diego ; Selvander, Madeleine LU ; Berglund, Johan Sanmartin and Behrens, Anders (2025) In Alzheimer's Research and Therapy 17(1).
Abstract

Background: Previous literature indicate retinal hyperspectral imaging as a non-invasive method with the potential for identifying amyloid-beta (Aβ) protein deposits. Current diagnostic methods, such as cerebrospinal fluid analysis or positron emission tomography, are costly, invasive, and non-scalable. Hyperspectral imaging offers a potentially accessible alternative for early detection of Alzheimer’s disease. The aim of this study is to investigate the potential of retinal hyperspectral imaging in identifying Aβ-positive patients within a clinical cohort from a memory clinic. Methods: A prospective cross-sectional cohort study was conducted between January 2023 and May 2024 at a single memory clinic in Sweden. The study recruited 57... (More)

Background: Previous literature indicate retinal hyperspectral imaging as a non-invasive method with the potential for identifying amyloid-beta (Aβ) protein deposits. Current diagnostic methods, such as cerebrospinal fluid analysis or positron emission tomography, are costly, invasive, and non-scalable. Hyperspectral imaging offers a potentially accessible alternative for early detection of Alzheimer’s disease. The aim of this study is to investigate the potential of retinal hyperspectral imaging in identifying Aβ-positive patients within a clinical cohort from a memory clinic. Methods: A prospective cross-sectional cohort study was conducted between January 2023 and May 2024 at a single memory clinic in Sweden. The study recruited 57 patients (35 Aβ-positive and 22 Aβ-negative) who underwent lumbar puncture as part of their diagnostic workup for cognitive complaints. Retinal hyperspectral images were captured from all participants at the time of their lumbar puncture and again 2–4 weeks later. Data was collected from five anatomical regions of the retina (Superior 1, Superior 2, Inferior 1, Inferior 2, and the center of the Fovea).The main outcome was the Aβ status (Aβ-positive or Aβ-negative). Catboost machine learning models were trained on hyperspectral imaging data to predict Aβ status. A nested cross-validation approach was used to train and evaluate classification models. Performance metrics included area under the curve (AUC), accuracy, sensitivity, and specificity. Results: The best-performing model used the combination of regions Superior 1, Superior 2, and center of the fovea, achieving a mean AUC of 0.77 (0.05), mean accuracy of 0.66 (0.03), and mean sensitivity of 0.73 (0.13) and mean specificity of 0.55 (0.12). Performance was consistent across outer folds. Models using all five regions or less-informative combinations yielded lower and more variable results. Conclusions: Retinal hyperspectral imaging combined with the Catboost algorithm demonstrated significant potential as a non-invasive biomarker for detecting Alzheimer’s disease in a consecutive clinical cohort. Further studies should validate these findings in larger, more diverse populations and explore the integration of hyperspectral imaging with other diagnostic modalities. Limited sample size and imaging constraints highlight the need for validation in diverse clinical settings. Trial registration: ClinicalTrials.gov, ID: NCT05604183 (registration date: 2022-10-27).

(Less)
Please use this url to cite or link to this publication:
author
; ; ; ; ; and
organization
publishing date
type
Contribution to journal
publication status
published
subject
keywords
Alzheimer’s disease, Amyloid-beta (Aβ), Biomarker, Catboost, Cerebrospinal fluid, Cognitive impairment, Hyperspectral imaging, Machine learning, Memory clinic, Retina
in
Alzheimer's Research and Therapy
volume
17
issue
1
article number
232
publisher
BioMed Central (BMC)
external identifiers
  • scopus:105020324403
  • pmid:41153055
ISSN
1758-9193
DOI
10.1186/s13195-025-01887-4
language
English
LU publication?
yes
id
210944b9-d6ed-4b3d-a561-0bae89501ec5
date added to LUP
2025-12-12 15:08:10
date last changed
2025-12-13 03:00:05
@article{210944b9-d6ed-4b3d-a561-0bae89501ec5,
  abstract     = {{<p>Background: Previous literature indicate retinal hyperspectral imaging as a non-invasive method with the potential for identifying amyloid-beta (Aβ) protein deposits. Current diagnostic methods, such as cerebrospinal fluid analysis or positron emission tomography, are costly, invasive, and non-scalable. Hyperspectral imaging offers a potentially accessible alternative for early detection of Alzheimer’s disease. The aim of this study is to investigate the potential of retinal hyperspectral imaging in identifying Aβ-positive patients within a clinical cohort from a memory clinic. Methods: A prospective cross-sectional cohort study was conducted between January 2023 and May 2024 at a single memory clinic in Sweden. The study recruited 57 patients (35 Aβ-positive and 22 Aβ-negative) who underwent lumbar puncture as part of their diagnostic workup for cognitive complaints. Retinal hyperspectral images were captured from all participants at the time of their lumbar puncture and again 2–4 weeks later. Data was collected from five anatomical regions of the retina (Superior 1, Superior 2, Inferior 1, Inferior 2, and the center of the Fovea).The main outcome was the Aβ status (Aβ-positive or Aβ-negative). Catboost machine learning models were trained on hyperspectral imaging data to predict Aβ status. A nested cross-validation approach was used to train and evaluate classification models. Performance metrics included area under the curve (AUC), accuracy, sensitivity, and specificity. Results: The best-performing model used the combination of regions Superior 1, Superior 2, and center of the fovea, achieving a mean AUC of 0.77 (0.05), mean accuracy of 0.66 (0.03), and mean sensitivity of 0.73 (0.13) and mean specificity of 0.55 (0.12). Performance was consistent across outer folds. Models using all five regions or less-informative combinations yielded lower and more variable results. Conclusions: Retinal hyperspectral imaging combined with the Catboost algorithm demonstrated significant potential as a non-invasive biomarker for detecting Alzheimer’s disease in a consecutive clinical cohort. Further studies should validate these findings in larger, more diverse populations and explore the integration of hyperspectral imaging with other diagnostic modalities. Limited sample size and imaging constraints highlight the need for validation in diverse clinical settings. Trial registration: ClinicalTrials.gov, ID: NCT05604183 (registration date: 2022-10-27).</p>}},
  author       = {{Dallora, Ana Luiza and Alexander, Jan and Palesetti, Pushpa Priyanka and Guenot, Diego and Selvander, Madeleine and Berglund, Johan Sanmartin and Behrens, Anders}},
  issn         = {{1758-9193}},
  keywords     = {{Alzheimer’s disease; Amyloid-beta (Aβ); Biomarker; Catboost; Cerebrospinal fluid; Cognitive impairment; Hyperspectral imaging; Machine learning; Memory clinic; Retina}},
  language     = {{eng}},
  number       = {{1}},
  publisher    = {{BioMed Central (BMC)}},
  series       = {{Alzheimer's Research and Therapy}},
  title        = {{Hyperspectral retinal imaging to detect Alzheimer’s disease in a memory clinic setting}},
  url          = {{http://dx.doi.org/10.1186/s13195-025-01887-4}},
  doi          = {{10.1186/s13195-025-01887-4}},
  volume       = {{17}},
  year         = {{2025}},
}