Individualized Gaussian Process-based Prediction of Memory Performance and Biomarker Status in Ageing and Alzheimer's disease

Abstract

Neuroimaging markers based on Magnetic Resonance Imaging (MRI) combined with various other measures (such as informative covariates, vascular risks, brain activity, neuropsychological test etc.,) might provide useful predictions of clinical outcomes during progression towards Alzheimers disease (AD). The Bayesian approach aims to provide a trade-off by employing relevant features combinations to build decision support systems in clinical settings where uncertainties are relevant. We tested the approach in the MRI data across 959 subjects, aged 59-89 years and 453 subjects with available neuropsychological test scores and CSF biomarker status (amyloid-beta (A{beta})42/40 & and phosphorylated tau (pTau)) from a large sample multi-centric observational cohort (DELCODE). In order to explore the beneficial combinations of information from different sources, we presented a MRI-based predictive modelling of memory performance and CSF biomarker status (positive or negative) in the healthy ageing group as well as subjects at risk of Alzheimers disease using a Gaussian process multikernel framework. Furthermore, we systematically evaluated predictive combinations of input feature sets and their model variations, i.e. (A) combinations of brain tissue classes and feature type (modulated vs. unmodulated), choices of filter size of smoothing (ranging from 0 to 15 mm full width at half maximum), and image resolution (1mm, 2mm, 4mm and 8mm); (B) incorporating demography and covariates (C) the impact of the size of the training data set (i.e., number of subjects); (D) the influence of reducing the dimensions of data and (E) choice of kernel types. Finally, the approach was tested to reveal individual cognitive scores at follow-up (up to 4 years) using the baseline features. The highest accuracy for memory performance prediction was obtained for a combination of neuroimaging markers, demographics, genetic information (ApoE4) and CSF-biomarkers explaining 57% of outcome variance in out of sample predictions. The best accuracy for A{beta}42/40 status classification was achieved for combination demographics, ApoE4 and memory score while usage of structural MRI improved the classification of individual patients pTau status.