Machine learning and novel ophthalmologic biomarkers for Alzheimer’s disease screening: Systematic Review

¹ Biostatistics and Informatics Unit, Department of Epidemiology, Clinical Research and Community Health, Faculty of Medicine and Pharmacy of Fez, Sidi Mohamed Ben Abdellah University, Fez, Morocco
² Laboratory of Artificial Intelligence, Data Science and Emerging Systems, National School of Applied Sciences Fez, Sidi Mohamed Ben Abdellah University, Fez, Morocco
³ Department of Epidemiology, Clinical Research and Community Health, Faculty of Medicine and Pharmacy of Fez, Sidi Mohamed Ben Abdellah University, Fez, Morocco

^* Corresponding author. Email: hbourkhime@gmail.com
^* Corresponding author. Email: noura.qarmiche@usmba.ac.ma
^* Corresponding author. Email: mohammed.omari@usmba.ac.ma
^* Corresponding author. Email: nassibabahra@gmail.com
^* Corresponding author. Email: nabil.tachfouti@usmba.ac.ma
^* Corresponding author. Email: samira.elfakir@usmba.ac.ma
^* Corresponding author. Email: nada.otmani@usmba.ac.ma

Abstract

Background: Alzheimer’s disease (AD) is a progressive neurodegenerative disease that leads to dementia and eventual death, the reason why screening is so beneficial in its early stages. Recent evidence suggests that memory and vision impairments are closely linked to Alzheimer’s disease. Moreover, assessing vision disorders may improve early detection and treatment of dementia. Therefore, some research has been conducted on screening for AD disease using new machine learning (ML) techniques on novel ophthalmologic biomarkers data.

Objective: To summarize existing findings on machine learning models exploring eye changes data to predict cognitive decline in the context of AD.

Methods: Systematic review of original research between January 2016 and August 2021. A search covered two databases on (Scopus) and (PubMed).

Results: From 104 search results, 13 articles were selected after using the eligibility criteria: 5 machine learning models used retinal texture data, 5 models included eye movement data, 2 proposed models used iris change data, and 1 proposed model used corneal nerve loss data.

Conclusion: Promising results are reported in almost all 13 studies, but very few have been implemented in research or clinical practice. The principal constraints in this area are limited standardization and comparability of results..