Predicting symptom severity in autism spectrum disorder based on cortical thickness measures in agglomerative data

Abstract

Machine learning approaches have been widely used for the identification of neuropathology from neuroimaging data. However, these approaches require large samples and suffer from the challenges associated with multi-site, multiprotocol data. We propose a novel approach to address these challenges, and demonstrate its usefulness with the Autism Brain Imaging Data Exchange (ABIDE) database. We predict symptom severity based on cortical thickness measurements from 156 individuals with autism spectrum disorder (ASD) from four different sites. The proposed approach consists of two main stages: a domain adaptation stage using partial least squares regression to maximize the consistency of imaging data across sites; and a learning stage combining support vector regression for regional prediction of severity with elastic-net penalized linear regression for integrating regional predictions into a wholePreprint submitted to NeuroImage September 4, 2017 brain severity prediction. The proposed method performed markedly better than simpler alternatives, better with multi-site than single-site data, and resulted in a considerably higher cross-validated correlation score than has previously been reported in the literature for multi-site data. This demonstration of the utility of the proposed approach for detecting structural brain abnormalities in ASD from the multi-site, multi-protocol ABIDE dataset indicates the potential of designing machine learning methods to meet the challenges of agglomerative data.

Does size matter? The relationship between predictive power of single-subject morphometric networks to spatial scale and edge weight

“…7. They appear in the lingual, supra-marginal, post-and precentral areas, which are consistent with previous reports on Autism studying the group differences in developmental patterns of cortical thickness (Smith et al 2016;Scheel et al 2011), as well as found to be important in other prediction tasks (Moradi et al 2017).…”

A systematic review of structural MRI biomarkers in autism spectrum disorder: A machine learning perspective

“…It was also unclear from the reviewed studies in Supplementary Table 2 as to the developmental trajectory of ASD, with four studies reporting accelerated thinning (Auzias et al, 2016;Koolschijn and Geurts, 2016;Wallace et al, 2015;Zielinski et al, 2014) and another four studies reporting slowed thinning (Doyle-Thomas et al, 2013a;Khundrakpam et al, 2017;Sussman et al, 2015;Yang et al, 2016), reflecting the inconsistencies in ASD development observed for the whole brain volume (ICV). Despite this, increased CT was associated with poorer outcomes across multiple ASD severity measures, including social communication language ability Sharda et al, 2016aSharda et al, , 2016b and the ADOS (Moradi et al, 2017;Sato et al, 2013), supporting this measure as a useful biomarker of autism. Furthermore, several studies investigate the covariance of CT across all regions of the cortex, and have established altered correlations and reduced covariance in CT with Euclidean distance when comparing ASD with TDC (Bethlehem et al, 2017;Sato et al, 2013), which may reflect the long-distance under-connectivity indicative of autism (Kikuchi et al, 2015).…”

“…Many studies found correlations between brain structure and clinical severity of autism, such as the ADOS (Dougherty et al, 2016b;Grecucci et al, 2016;Khundrakpam et al, 2017;Moradi et al, 2017;Sato et al, 2013) and ADI-R (Floris et al, 2016;Xiao et al, 2017), which are considered the two 'gold standard' diagnostic evaluations for children with autism (Reaven et al, 2008), as well as the Clinical Evaluation of Language Fundamentals (CELF) (Sharda et al, 2016a), Intelligence Quotient (IQ) (Fredo et al, 2014;Sharda et al, 2016b), the Movement Assessment Battery for Children (MABC) (Hanaie et al, 2016), and the Repetitive Behaviour Scale-Revised (RBS-R) (Eisenberg et al, 2015). Of the studies outlined in Supplementary Table 8, several reported strong correlations (Pearson's r between 0.4 -0.5) with outcome (Eisenberg et al, 2015;Gebauer et al, 2015;Grecucci et al, 2016;Hanaie et al, 2016;Joseph et al, 2014;Knaus et al, 2017;Laidi et al, 2017;Mahajan et al, 2016).…”

“…Therefore N-fold cross validation, which splits the data equally into N groups, trains the model on (N-1) of the groups and tests on the remaining one group 'N' times, seems the optimal approach in this setting. Indeed, several studies utilise 5-fold (Demirhan, 2018;Sen et al, 2018) and 10-fold cross validation (Akhavan Aghdam et al, 2018;Moradi et al, 2017;Wang et al, 2016). This approach can achieve an accurate estimation of model performance, and future imaging studies would be able to use such an approach to validate structural biomarkers of ASD, regardless of dataset limitations.…”

Applications of Supervised Machine Learning in Autism Spectrum Disorder Research: a Review

“…Neuroimaging data has also been examined using machine learning approaches. Moradi et al (2017) predicted symptom severity of individuals with ASD based on cortical thickness using support vector regression (SVR) and ENet penalized linear regression. Participants included 156 individuals with ASD ages 8 to 40 years old.…”