Building Cost-effective Model For Predicting β-Amyloid Burden by Combination of Cortical Thickness, Volume, and Neuropsychological Assessment Measures

Abstract

In the past few years numerous studies focused on early detection of Alzheimer's disease (AD). Of various biomarkers, cerebral beta-amyloid (Aβ) is the most important evidence that implicates AD-specific neuropathology, which begins to accumulate decades before the clinical onset of AD. Thus early identification of individuals at high risk of developing AD becomes crucial. Although cerebral Aβ burden can be assessed in-vivo via positron emission tomography (PET), these techniques are expensive and are not commonly available. This study supports that simple and cost-effective methods can predict cerebral Aβ based on neuropsychological test scores with demographics derived from machine learning approaches. Specifically, we applied several well-established and simple machine learning algorithms to Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset. Our study demonstrated that neuropsychological assessment data with demographics using simple machine learning algorithms can be a cost-effective method for predicting amyloid beta compared to magnetic resonance imaging (MRI) data.