Differences in atypical resting-state effective connectivity distinguish autism from schizophrenia

Abstract

Autism and schizophrenia share overlapping genetic etiology, common changes in brain structure and common cognitive deficits. A number of studies using resting state fMRI have shown that machine learning algorithms can distinguish between healthy controls and individuals diagnosed with either autism spectrum disorder or schizophrenia. However, it has not yet been determined whether machine learning algorithms can be used to distinguish between the two disorders. Using a linear support vector machine, we identify features that are most diagnostic for each disorder and successfully use them to classify an independent cohort of subjects. We find both common and divergent connectivity differences largely in the default mode network as well as in salience, and motor networks. Using divergent connectivity differences, we are able to distinguish autistic subjects from those with schizophrenia. Understanding the common and divergent connectivity changes associated with these disorders may provide a framework for understanding their shared cognitive deficits.

Overlaps in Brain Dynamic Functional Connectivity between Schizophrenia and Autism Spectrum Disorder

“…As noticed empirically that the NBS was producing about 300 connection using a p-value threshold related to the t-test of α = 0.05 (proportionally in agreement with Mastrovito et al [16]), we lowered the NBS threshold to α = 0.01 to allow visual inspection of those results.…”

“…Recently, Yahata et al identified a small number of connections which can discriminate ASD subjects from healthy control [15]. Mastrovito et al compared atypical functional connectivity between ASD to typically developing children and schizophrenia to normal control, highlighting also common connectivity features between ASD and schizophrenia [16].…”

Multi-Link Analysis: Brain Network Comparison via Sparse Connectivity Analysis

“…Furthermore, sparse models follow a feature selection agenda to subselect among existing variables, whereas PCA dimensionality reduction follows a feature engineering agenda to generate a set of new variables. A feature selection by sparse model is indeed similar to the RFE used by Mastrovito et al 19 . However, the stability of RFE approach depends heavily on the type of model used for feature ranking at each iteration, and as shown empirically, using regularized ridge regression jointly to stability selection criteria can provide more stable results in terms of stability selection of features, and yields finite sample familywise error control 28,29 .…”

“…While, Chen et al 18 enhanced NBS regulating the topological structures comprised. Other research groups [19][20][21] leveraged support vector machines (SVM) weights to identify discriminating regions. SVM is a supervised learning method which constructs a hyperplane or set of hyperplanes in a high-or infinite-dimensional space used for classification.…”

“…Specifically, Ng et al 20 used a projection of covariance estimates onto a common tangent space to reduce the statistical dependencies between elements. Then, while Mastrovito et al 19 employed recursive feature elimination (RFE) to identify connections relevant for the classification, both Goankar et al 21 and Ng et al 20 found meaningful connections using t-tests on the SVM weights. A more advanced machine learning approach is based on SVM coupled with Riemannian/Grassmannian geometry 22 .…”