Projection to Latent Spaces disentangles specific cerebral morphometric patterns associated to aging and preclinical AD

Published in Alzheimer\`s Association International Conference (AAIC), 2018

Authors: Adrià Casamitjana, Paula Petrone, Miquel Artigues, José Luis Molinuevo, Juan Domingo Gispert, Verónica Vilaplana for the Alzheimer\`s Disease Neuroimaging Initiative



Introduction: Partial Least Squares (PLS) is a mathematical technique that relates two sets of observable quantities by means of a few latent explanatory variables. In this work we used PLS to derive the latent factors that explain the association between brain morphometry as measured by MRI and core AD cerebrospinalfluid (CSF) biomarkers. The aim of this study was to use PLS to discover the associations between CSF biomarkers in cerebral brain volumes in preclinical AD and to disentangle their specific contribution from confounding demographic factors. Methods: Magnetic Resonance Imaging (MRI-T13D) was acquired from a subset of the ADNI cohort comprising 321 asymptomatic subjects (185 controls, 136 preclinical) for which CSF biomarkers was made publicly available. Each image was preprocessed using FreeSurfer to derive cortical thickness and grey matter volumes in the cortical regions of the Desikan-Killiany parcellation, used as predictors; age, education, number of APOE4 alleles and CSF biomarkers were taken as the response variables. PLS was used to find the latent space that best describes the variation of predictors, responses and their covariance. The latent space dimension (L) was chosen as a compromise between the total variance explained and the cross-validated total r-squared coefficient. Results: Cortical thickness shows less variation than brain volumetric information resulting in a lower dimensional latent space (L=3 and L=5). Latent variables showed significant correlation with age and CSF biomarkers, especially with tau and ptau measures while no latent process related to education and the number of APOE4 alleles was found. Figures 1 and 2 illustrate, respectively, latent processes related to cortical thickness and grey matter volume by coloring brain structures according to their importance. Patterns related to dementia-due to AD show a combination of cortical shrinkage (e.g: inferior temporal) and cortical expansion (e.g: transverse temporal) (Fig 1.c) as well as atrophy in the choroid plexus (Fig. 2c) or caudate regions (Fig. 2d). Age-related latent variables show global cortical reduction and atrophy, except for the choroid plexus regions (Fig. 1a, 2a) Conclusions: PLS was able to disentangle the cerebral morphometric patterns associated to preclinical AD stages from other demographic factors. Results with both cortical thickness and volumetric data present significant overlap, thus showing the robustness of this approach. Interestingly, volumetric data showed more significant correlations with CSF Abeta than cortical thickness.