Quantifiable markers of ASD via multivariate MEG-DTI combination

University Of Pennsylvania, Philadelphia PA

Investigators

Linked publications, trials & patents

Abstract

DESCRIPTION (provided by applicant): Research in ASD has aimed at identifying brain level endophenotypes or univariate biomarkers associated with the different subtypes of ASD using different imaging modalities. DTI-based studies have provided insights into pathology-induced and neuro-developmental changes in white matter (WM) architecture and structural connectivity for ASD, and the neurobiological basis for neuronal deficits associated with ASD. Specific symptoms such as LI are also being investigated via MEG derived temporal signatures based on evoked neuromagnetic activity during speech perception, processing and cognition. Encouraged by the fact that DTI and MEG studies have individually been able to characterize different aspects of ASD and some of its related symptoms like LI, in this proposal we aim at combining the MEG-based temporal signatures with the DTI-based structural connectivity signatures to create a multi-parametric spatio temporal signature of ASD using pattern classification. Creating joint multivariate signatures for an ASD population is rendered challenging by high heterogeneity imposed by disease and development~ manifest as various subtle subtypes possibly associated with differential neuronal deficits and the presence of datasets with missing modalities due to the inability of the subject in completing both diffusion and electrophysiology protocols. In this proposal, we aim to address these challenges, by creating quantifiable multi-parametric spatio-temporal markers of autism learnt from the underlying pathology patterns of the population, by combining the MEG-based temporal signatures with the DTI-based structural connectivity signatures using pattern classifiers created on an ASD population with LI. We identify spatio-temporally compatible DTI- MEG features (aim 1) and create various multi- parametric pattern classifiers that will quantify ASD pathology by learning the population heterogeneity despite partially missing data (aim 2). Via the application in aim 3, to an ASD population with LI as one of the sources of heterogeneity, we will have created classifiers that will elucidate he importance of combination diffusion and MEG information, identify the regional and connectivity combinations that best characterize ASD, and provide abnormality scores associated with each subject that can aid in quantifying the likelihood of impairment, thereby enriching diagnosis decisions. The classifiers that embrace population heterogeneity and missing data will be generic and flexible in applicability to any population, as these can be easily retrained with new MEG and DTI features, and will aid future studies that wish to incorporate spatio-temporal information.