AUTHOR=Mohanty Rosaleena , Sinha Anita M. , Remsik Alexander B. , Dodd Keith C. , Young Brittany M. , Jacobson Tyler , McMillan Matthew , Thoma Jaclyn , Advani Hemali , Nair Veena A. , Kang Theresa J. , Caldera Kristin , Edwards Dorothy F. , Williams Justin C. , Prabhakaran Vivek TITLE=Early Findings on Functional Connectivity Correlates of Behavioral Outcomes of Brain-Computer Interface Stroke Rehabilitation Using Machine Learning JOURNAL=Frontiers in Neuroscience VOLUME=Volume 12 - 2018 YEAR=2018 URL=https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2018.00624 DOI=10.3389/fnins.2018.00624 ISSN=1662-453X ABSTRACT=The primary goal of this work was to apply data-driven machine learning regression to assess if changes in resting state functional connectivity could predict changes in behavioral domains in stroke subjects who completed brain-computer interface (BCI) interventional therapy for motor rehabilitation. The study cohort consisted of 20 chronic-stage stroke subjects exhibiting persistent upper-extremity motor deficits who received therapy using a closed-loop neurofeedback BCI device. Over the course of this therapy, resting-state functional MRI scans were collected at four distinct time points: namely, pre-therapy, mid-therapy, post-therapy and one month after completion of all sessions of therapy. Behavioral assessments were administered outside the scanner at each time point to collect objective measures such as the Action Research Arm Test, Nine-Hole Peg Test, and Barthel Index as well as subjective measures including the Stroke Impact Scale. The present analysis focused on neuroplasticity and behavioral changes occurring between pre-therapy and post-therapy, as it is expected that peak changes would be observed across these time points. The dynamics of changes in resting-state functional connectivity (rs-FC) across the motor network were used as input features and changes in behavioral measures were used as outcomes for machine-learning-based linear and non-linear support vector regression (SVR) models. Potential clinical confounding factors such as age, gender, lesion hemisphere, and stroke severity were included as additional features in each of the regression models. Sequential forward feature selection procedure narrowed the search for important predictors. Rs-FC within the thalamus and supplementary motor area predicted several behavioral changes. Additionally, individual predictors specific to behavioral gains on each scale were identified. Comparatively, linear SVR models aided in evaluation of contribution of individual predictors and seed regions while non-linear SVR models achieved higher prediction performance of behavioral outcomes.