Meeting Banner
Abstract #3306

Support Vector Machine Classification of Stroke Using Resting State Functional Connectivity

Svyatoslav Vergun1, Veena A. Nair2, Matthew Jensen3, Marcus Chacon3, Justin Sattin3, Vivek Prabhakaran2

1Medical Physics, UW-Madison, Madison, WI, United States; 2Radiology, UW-Madison, Madison, WI, United States; 3Neurology, UW-Madison, Madison, WI, United States

Multivariate pattern analysis methods have been shown successful in extracting significant information and classifying individual scans. In this work, a support vector machine classifier accurately discriminated between stroke and normal aging subjects based on their resting state functional connectivity. 50 resting state fMRI scans from 24 normal and 26 stroke subjects were preprocessed and time series from 160 functional ROIs were correlated to produce a functional connectivity matrix for each subject. Each subjects correlations were input as features into the classifier, which predicted subjects with 80% accuracy using leave-one-out cross validation. Sensorimotor network connectivity was most influential for classification.

Abstract PDF Presentation Video

Keywords

ability able accessed accuracy accurate accurately activity acute adapted added affected aging allowing allows analyses aspect audience band brain classification classifications classifier classify connection connections connectivity containing correction correlations cross custom default defined discriminated disease drive driving dynamics error every examine extract extracted feature features female filtering focused functional functions generated global gradient healthy implemented importance important included indicating individual influenced insight investigate kernel labels language learning leave linear lists machine maturity medical memory meta mode motion multivariate network networks neural neurology pair particular pass pathological pattern people physics populations potentially powerful predict predicted predicting prediction preprocessed previously processing project projects quadratic radiology regression represents resting scale scanners science scientists scripts seek sensitivity series slice slices slightly smoothing space spatial specificity spider stroke subject subjects successful support table target timing toolbox tools transformation underlying validation vector volumes white whole width years