Multimodal neuroimaging in healthy cognitive aging and alzheimer’s disease

Abstract

Early detection of brain morphological and functional alterations are important for an early diagnosis of Alzheimer’s disease (AD). However, the method to determine these changes early remains elusive. Thus, to further elucidate the mechanisms and pathophysiology underlying AD, the trajectories of the brain macro- and microstructural, functional, and neurometabolic magnetic resonance imaging (MRI) alterations co-occurring in AD were determined using multimodal neuroimaging. A total of twenty-four participants (10 AD and 14 healthy controls [HC]) who completed the Mini-Mental State Examination (MMSE), Montreal Cognitive Assessment (MoCA), and Clinical Dementia Rating (CDR) scale underwent a high-resolution threedimensional T1-weighted volumetric MRI, diffusion tensor imaging (DTI), and restingstate fMRI (rs-fMRI). The following MRI measures were derived from computational analyses – T1-weighted MRI: gray matter (GM), white matter (WM), Jacobiandeterminant GM, cortical thickness (CT), sulcus depth (SD), fractal dimension (FD), and gyrification index (GI); DTI: fractional anisotropy (FA), mean diffusivity (MD), and axial diffusivity (AxD); and rs-fMRI: brain activity. In addition, Multi-Kernel (MK) Machine Learning (ML)-based framework was used for the classification of AD from HC utilizing either a unimodal or multimodal MRI-derived metrics. Then, the ML algorithm, Kernel Ridge Regression (KRR) was used for the prediction of neuropsychological scores in patients with AD. Statistically significant results were considered at p-value < .05. There were evidence of differential patterns in the wholebrain macro- and micro-structural integrity and functional status in AD compared to HC. MKL showed comparable classification accuracy values from single and combined feature sets. KRR provides a greater insight into the interplay between multimodal whole-brain alterations and neuropsychological tests associated with AD. Thus, this study concludes as such – Firstly, multivariate pattern analysis based on machine learning employing multimodal neuroimaging techniques play a crucial role in elucidating the pathophysiological process of whole-brain alterations underlying AD. Secondly, this method is able to discriminate patients with AD from individuals with healthy cognitive status. Thirdly, it provides a robust method capable of predicting clinical outcome in AD. Therefore, this study suggests that integrating multivariate feature sets into machine learning algorithms may optimize early diagnosis and the prediction of clinical outcome in patients with AD.