A Machine Learning Approach for Identifying Anatomical Biomarkers of Early Mild Cognitive Impairment
Authors:
Alwani Liyana Ahmad,
Jose Sanchez-Bornot,
Roberto C. Sotero,
Damien Coyle,
Zamzuri Idris,
Ibrahima Faye
Abstract:
Alzheimer's Disease (AD) is a progressive neurodegenerative disorder that primarily affects the aging population by impairing cognitive and motor functions. Early detection of AD through accessible methodologies like magnetic resonance imaging (MRI) is vital for develo** effective interventions to halt or slow the disease's progression. This study aims to perform a comprehensive analysis of mach…
▽ More
Alzheimer's Disease (AD) is a progressive neurodegenerative disorder that primarily affects the aging population by impairing cognitive and motor functions. Early detection of AD through accessible methodologies like magnetic resonance imaging (MRI) is vital for develo** effective interventions to halt or slow the disease's progression. This study aims to perform a comprehensive analysis of machine learning techniques for selecting MRI-based biomarkers and classifying individuals into healthy controls (HC) and unstable controls (uHC) who later show mild cognitive impairment within five years. The research utilizes MRI data from the Alzheimer's Disease Neuroinformatics Initiative (ADNI) and the Open Access Series of Imaging Studies 3 (OASIS-3), focusing on both HC and uHC participants. The study addresses the challenges of imbalanced data by testing classification methods on balanced and unbalanced datasets, and harmonizes data using polynomial regression to mitigate nuisance variables like age, gender, and intracranial volume. Results indicate that Gaussian Naive Bayes and RusBoost classifiers shows an optimal performance, achieving accuracies of up to 76.46% and 72.48% respectively on the ADNI dataset. For the OASIS-3 dataset, Kernel Naive Bayes and RusBoost yield accuracies ranging from 64.66% to 75.71%, improving further in age-matched datasets. Brain regions like the entorhinal cortex, hippocampus, lateral ventricle, and lateral orbitofrontal cortex are identified as significantly impacted during early cognitive decline. Despite limitations such as small sample sizes, the study's harmonization approach enhances the robustness of biomarker selection, suggesting the potential of this semi-automatic machine learning pipeline for early AD detection using MRI.
△ Less
Submitted 29 May, 2024;
originally announced July 2024.
