The Predictive Power of Structural MRI in Autism Diagnosis

The Predictive Power of Structural MRI in Autism Diagnosis

August 19, 2019

Abstract
Diagnosis of Autism Spectrum Disorder (ASD) using structural magnetic resonance imaging (sMRI) of the brain has been a topic of significant research interest. Previous studies using small datasets with well-matched Typically Developing Controls (TDC) report high classification accuracies (80-96%) but studies using the large heterogeneous ABIDE dataset report accuracies less than 60%. In this study we investigate the predictive power of sMRI in ASD using 373 ASD and 361 TDC male subjects from the ABIDE. Brain morphometric features were derived using FreeSurfer and
classification was performed using three different techniques:Random Forest (RF), Support Vector Machine (SVM) and Gradient Boosting Machine (GBM). Although high classification accuracies were possible in individual sites (with a maximum of 97% in Caltech), the highest classification accuracy across all sites was only 60% (sensitivity = 57%,specificity = 64%). However, the accuracy across all sites improved to 67% when IQ and age information were added to morphometric features. Across all three classifiers, volume and surface area had more discriminative power. In general, important features for classification were present in the frontal
and temporal regions and these regions have been implicated in ASD. This study also explores the effect of demographics and behavioral measures on the predictive power of sMRI. Results show that classification accuracy increases with autism severity and that ASD detection with sMRI is easier before the age of 10 years.

Introduction

Autism Spectrum Disorder (ASD) is a polygenetic neurodevelopmental disorder characterized by repetitive behavior, intellectual disability, and impaired language and social skills. The behavioral phenotype of ASD is well characterized, but its etiology and pathogenesis remains elusive. However, structural magnetic resonance imaging (sMRI) studies have reported subtle anatomical differences in brain structures of ASD subjects versus Typically Developing Control (TDC) subjects, including differences in the structure of the frontal lobe, parietal lobe, temporal lobe, limbic system and cerebellum.

At present, ASD diagnosis is primarily based on behavioral criteria. This approach is subjective, time consuming and does not help understanding the underlying etiology. This makes ASD diagnosis based on imaging data highly desirable. Brain biomarkers derived from sMRI can help identify the neuroanatomical basis of heterogeneity in ASD and can be a powerful tool for early diagnosis and intervention. Mass-univariate techniques such as Voxel Based Morphometry (VBM) are commonly employed in brain imaging studies to detect brain anatomical differences. Although VBM has high exploratory power, it lacks the statistical power required to detect subtle multivariate structural differences. Multivariate pattern recognition techniques (MVPT) are capable of detecting subtle and spatially distributed differences in data and thus hold promise for extracting higher predictive power.

In recent years, there have been a number of studies applying MVPT for ASD vs. TDC classification. These studies can be broadly categorized into two groups based on the data) data matched for demographics and behavioral (DB) measures such as age, sex and IQs ) large heterogeneous data. The following classification methods and accuracies have been reported by the first group of studies: Support Vector Machine (SVM) on gray matter scans (81%) , Logistic Model Trees (LMT) on regional cortical thickness (87%) , SVM on gray matter in default mode network regions (90%) , and SVM on regional and inter-regional cortical and subcortical features (96%) . The second group of studies uses the large heterogeneous ABIDE dataset and reports classification accuracies less than 60%.

In this study, we investigate the predictive power of sMRI in ASD utilizing three heterogeneous classifiers– Random Forest (RF) , SVM  and Gradient Boosting Machine (GBM) . We perform classification within each and across all sites of the ABIDE dataset. We investigate the effects of DB measures on the predictive power of sMRI in ASD and the incremental power that can be gained from them. This has not been addressed by previous studies. In addition, we investigate the relationship between Autism Diagnostic Observation Score (ADOS) and autism class probability, an autism score produced by a classifier. In addition we compare the discriminative power of morphometric properties such as volume, area, thickness, curvature and folding index with three classifiers. We discuss issues related to over-fitting due to small sample size and feature selection, using results from individual sites. Finally, we conclude by discussing the challenges and future directions on predicting ASD using sMRI.

Authors:Gajendra J. Katuwal,  Nathan D. Cahill, Stefi A. Baum,Andrew M. Michael

For more details click here

Source:https://www.researchgate.net