Meeting Banner
Abstract #0520

Identification of Neural Connectivity Signatures of Autism Using Machine Learning

Gopikrishna Deshpande1, 2, Karthik Ramakrishnan Sreenivasan3, Hrishikesh Deshpande4, Rajesk K. Kana5

1AU MRI Research Center, Department of Electrical and Computer Engineering, Auburn University, Auburn, AL, United States; 2Department of Psychology, Auburn University, Auburn, AL, United States; 3 AU MRI Research Center, Department of Electrical and Computer Engineering, Auburn University, Auburn, AL, United States; 4Department of Biomedical Engineering, University of Alabama, Birmingham, AL, United States; 5Department of Psychology, University of Alabama, Birmingham, AL, United States

The current study focuses on effective connectivity (EC) in autism, demonstrating the use of machine learning for identification of metrics which can be used to predict a novel subjects group membership. fMRI time-series were de-convolved using a cubature Kalman filter and the resultant neuronal variables were input into a multivariate autoregressive model (MVAR) to obtain the EC paths. These metrics were then input into a recursive cluster elimination based support vector machine (RCE-SVM) classifier which showed a prediction accuracy of 94.3% based only on causal connectivity weights indicating that EC could serve as a potential non-invasive neuroimaging biomarker for autism.

Abstract PDF Presentation Video

Keywords

abnormalities accuracy accurately activated activation addressing adopt adults altered although anatomical another approaches asked auburn autism autistic autoregressive back biomedical black blind blue brain causal causality choose classification classifier classify clear cluster clusters cognitive comic complex comprised computer connectivities connectivity considered control correlation cortical deconvolution decreased determine developing development diagnosis directional disorders disrupted effective electrical engineering even explanatory extracted feature features feed filter focuses forward functional functioning guarantee handbook identification importantly individuals influence input intentional interregional invasive involving judgment just laws lead learning life link logical machine many matrices membership mind mirror missing model multivariate neither neural novel output participants paths physical physics plot plotted potential power predict predictive previous processing psychology purple reached recursive reduction relied reports respectively response resultant rules scale scanned scanner separated separation series serve signature signatures significantly social spectrum story strip studies subject suggested support task tasks theory transactions typical typically underlying variables vector vital weaker