Meeting Banner
Abstract #2230

Minimizing Spurious Functional Connectivity Findings from Resting State fMRI

Prantik Kundu1, Noah Brenowitz1, Souheil J. Inati2, Ziad S. Saad3, Petra Vertes4, Yulia Worbe4, Valerie Voon4, Ed Bullmore4, Peter A. Bandettini5

1Section on Functional Imaging Mehthods, NIMH, Bethesda, MD, United States; 2Functional MRI Core Facility, NIMH, Bethesda, MD, United States; 3Scientific and Statistical Computing Core, NIMH, Bethesda, MD, United States; 4Dept. of Psychiatry, University of Cambridge, Cambridge, Cambridgeshire, United Kingdom; 5Functional MRI Core Facility, National Institute of Mental Health, Bethesda, MD, United States

We present a solution to the critical problem for resting state fMRI (rs-fMRI) that functional connectivity estimates are severely biased by any level of in-scanner subject head movement. We show that separating BOLD from non-BOLD signals using T2* decay analysis of multi-echo fMRI and independent components analysis (ME-ICA) entirely removes both linear and non-linear manifestations of motion artifact. This denoising is achieved without arbitrary processing such as data censoring or band pass filtering, making ME-ICA the first physically and statistically principled approach for comprehensive denoising of rs-fMRI data. Our study gives a deeper understanding of rs-fMRI while solving a critical problem for this exciting methodology.

Abstract PDF

Keywords

accounting acquisition activation alternating anatomical approved arise artifact artifacts assessments background band bias bold brain called categorically channel closely coefficients coil committee component components comprehensively compute computing conditioning connectivity contrast contrasts core correlation correlations critical critically datasets decay decomposed decomposition degrees dependence dept differentiating distributions effective efficacy enable engine equalized equivalent error estimation ethical facility false filtering finally findings fisher flat fluctuations freedom functional head highly hypothesis identical identify inconsistent increasingly independent inference integrated involves isolating kingdom knowledge linear local manifestations maps matrix medical methodology metrics minimizing mitigates motion movements nearly networks nonlinear nuisance oblique patterns permuted peter positive power preprocessed principal problem processing psychiatry question realignment reconstruction regression related remove removes removing resolution respective resting robust robustly sample scanner scientific scrubbing section seed separate separating series severely simplifies slice slices smoothing solutions sources space spatial spurious statistical studies subject subsequently subspace systematic temporal term terms trace transformation validity volume volunteers