Abstract #3515

Localization and Detrending of Physiological Noise in Resting State FMRI Using Machine Learning

Thomas WJ Ash¹, John Suckling², Martin Walter³, Cinly Ooi², T Adrian Carpenter¹, Guy B. Williams¹

¹Wolfson Brain Imaging Centre, University of Cambridge, Cambridge, Cambridgeshire, United Kingdom; ²Brain Mapping Unit, University of Cambridge, Cambridge, Cambridgeshire, United Kingdom; ³Department of Psychiatry, University of Magdeburg

Using machine learning tools on fMRI imaging data, we can predict the output of a physiological monitoring device with accuracy far better than chance. The model thus derived shows physiological noise to be localized mainly to the cerebrovascular system, CSF and the brain edge. Upon detrending this noise to the extent that it is no longer predictable, voxel autocorrelation as measured by the Hurst exponent is significantly decreased in the brain parenchyma, in contrast to results when using common physiological noise correction tool RETROICOR, which does not affect autocorrelation in our dataset.