The impact of global signal regression on resting state networks

Slides:



Advertisements
Similar presentations
Bayesian inference Lee Harrison York Neuroimaging Centre 01 / 05 / 2009.
Advertisements

Wellcome Dept. of Imaging Neuroscience University College London
General Linear Model L ύ cia Garrido and Marieke Schölvinck ICN.
1st level analysis - Design matrix, contrasts & inference
Resting state functional connectivity MRI in isoflurane-anesthetized rat brain Resting state functional connectivity MRI (rs-fcMRI) provides a unique opportunity.
1 st Level Analysis: design matrix, contrasts, GLM Clare Palmer & Misun Kim Methods for Dummies
SPM 2002 C1C2C3 X =  C1 C2 Xb L C1 L C2  C1 C2 Xb L C1  L C2 Y Xb e Space of X C1 C2 Xb Space X C1 C2 C1  C3 P C1C2  Xb Xb Space of X C1 C2 C1 
Outline What is ‘1st level analysis’? The Design matrix
Quality Assurance NITRC Enhancement Grantee Meeting June 18, 2009 NITRC Enhancement Grantee Meeting June 18, 2009 Susan Whitfield-Gabrieli & Satrajit Ghosh.
FMRI Data Analysis: I. Basic Analyses and the General Linear Model
Classical inference and design efficiency Zurich SPM Course 2014
Statistical Analysis fMRI Graduate Course October 29, 2003.
fMRI data analysis at CCBI
Multiple testing Justin Chumbley Laboratory for Social and Neural Systems Research Institute for Empirical Research in Economics University of Zurich With.
The General Linear Model (GLM) Methods & models for fMRI data analysis in neuroeconomics November 2010 Klaas Enno Stephan Laboratory for Social & Neural.
Multiple testing Justin Chumbley Laboratory for Social and Neural Systems Research Institute for Empirical Research in Economics University of Zurich With.
Dissociating the neural processes associated with attentional demands and working memory capacity Gál Viktor Kóbor István Vidnyánszky Zoltán SE-MRKK PPKE-ITK.
The General Linear Model (GLM)
The General Linear Model (GLM) SPM Course 2010 University of Zurich, February 2010 Klaas Enno Stephan Laboratory for Social & Neural Systems Research.
Multiple comparison correction Methods & models for fMRI data analysis 29 October 2008 Klaas Enno Stephan Branco Weiss Laboratory (BWL) Institute for Empirical.
Signal and Noise in fMRI fMRI Graduate Course October 15, 2003.
General Linear Model & Classical Inference Guillaume Flandin Wellcome Trust Centre for Neuroimaging University College London SPM M/EEGCourse London, May.
The General Linear Model
From Localization to Connectivity and... Lei Sheu 1/11/2011.
With a focus on task-based analysis and SPM12
FMRI Methods Lecture7 – Review: analyses & statistics.
SPM short course – Oct Linear Models and Contrasts Jean-Baptiste Poline Neurospin, I2BM, CEA Saclay, France.
Statistical Analysis fMRI Graduate Course November 2, 2005.
Contrasts & Statistical Inference
SPM short course Functional integration and connectivity Christian Büchel Karl Friston The Wellcome Department of Cognitive Neurology, UCL London UK http//:
Functional Brain Signal Processing: EEG & fMRI Lesson 14
Methods for Dummies Overview and Introduction
The General Linear Model (for dummies…) Carmen Tur and Ashwani Jha 2009.
Ch. 5 Bayesian Treatment of Neuroimaging Data Will Penny and Karl Friston Ch. 5 Bayesian Treatment of Neuroimaging Data Will Penny and Karl Friston 18.
Statistical Analysis An Introduction to MRI Physics and Analysis Michael Jay Schillaci, PhD Monday, April 7 th, 2007.
FMRI Modelling & Statistical Inference Guillaume Flandin Wellcome Trust Centre for Neuroimaging University College London SPM Course Chicago, Oct.
Idiot's guide to... General Linear Model & fMRI Elliot Freeman, ICN. fMRI model, Linear Time Series, Design Matrices, Parameter estimation,
The General Linear Model
The general linear model and Statistical Parametric Mapping I: Introduction to the GLM Alexa Morcom and Stefan Kiebel, Rik Henson, Andrew Holmes & J-B.
Analysis of FMRI Data: Principles and Practice Robert W Cox, PhD Scientific and Statistical Computing Core National Institute of Mental Health Bethesda,
Bayesian inference Lee Harrison York Neuroimaging Centre 23 / 10 / 2009.
The general linear model and Statistical Parametric Mapping II: GLM for fMRI Alexa Morcom and Stefan Kiebel, Rik Henson, Andrew Holmes & J-B Poline.
Group Analysis Individual Subject Analysis Pre-Processing Post-Processing FMRI Analysis Experiment Design Scanning.
SPM short course – Mai 2008 Linear Models and Contrasts Jean-Baptiste Poline Neurospin, I2BM, CEA Saclay, France.
The General Linear Model (GLM)
The General Linear Model (GLM)
Variational Bayesian Inference for fMRI time series
The general linear model and Statistical Parametric Mapping
The General Linear Model
Design Matrix, General Linear Modelling, Contrasts and Inference
Seed-based Resting State fMRI Analysis in FreeSurfer
Statistical Inference
The General Linear Model (GLM): the marriage between linear systems and stats FFA.
and Stefan Kiebel, Rik Henson, Andrew Holmes & J-B Poline
The General Linear Model (GLM)
Contrasts & Statistical Inference
The General Linear Model
Signal and Noise in fMRI
Statistical Parametric Mapping
Bayesian Methods in Brain Imaging
The General Linear Model
The General Linear Model (GLM)
Contrasts & Statistical Inference
MfD 04/12/18 Alice Accorroni – Elena Amoruso
The General Linear Model
The General Linear Model (GLM)
Probabilistic Modelling of Brain Imaging Data
The General Linear Model
The General Linear Model
Contrasts & Statistical Inference
Presentation transcript:

The impact of global signal regression on resting state networks Are anti-correlated networks introduced? Kevin Murphy, Rasmus M. Birn, Danil A. Handwerker, Tyler B. Jones, Peter A. Bandettini

Introduction Low frequency fluctuations (~0.1 Hz) Brain is intrinsically organized into dynamic, anti- correlated functional networks (Fox et al., 2005) common assumption: correlated fluctuations in resting state networks are neuronal As you have understand by now, low frequency fluctuations have been analyzed to map restings state networks. Patrickj just told you that researchers claimed that brain is intrinsically organized into dynamic anti-correlaed functional networks. Disruptions in connection of parts of Resting state networks are related to disorders (Alzheimer, schizophrenia, autism),  large interest in this topic. The common assumption in those connectivity analysis is that correlated fluctuations are neuronal in origin.

Introduction non neuronal sources of fluctuation (noise): cardiac pulsation, respiration  physiological measured changes in CO2 (Wise et al., 2004) magnetic noise, subjects head sinks… Noise reduction: Preprocessing: body, head correction... Global signal regression (GLM) filter out global signal Fluctuation in signal can also be caused by changes in cardiac pulsation, respiration. These noise can be controlled by measuring these parameters. It has also been shown that changes in arterial carbon dioxide cause very low frequency fluctuations. And then you can blame the hardware and subject for more noise. There are several methods to reduce noise. To account for several potential sources of noise Global signal regression is very often performed. That means you average the time series over all voxels and use it as a regressor in a general linear model.

Introduction Is global signal just uninteresting source of noise? only global signal and experimental conditions are orthogonal / uncorrelated PET: resulting time course not orthogonal to task-induced activations (Andersson, 1997) task-related voxels included in global regressor  underestimating true activation  introducing deactivations covariation for global signal  reduce intensity and introduce new negatively activated areas  default mode network But if you filter that signal out – the question is: Is this global signal really just uninteresting source of noise? One could say that if the global signal and the experimental conditions are uncorrelated. But a PET study by Andersson has shown that the time course was not uncorrelated to the task induced activation. So task related voxels were included in the global regressor which could lead to an underestimating of the true activation and can even introduce deactivations In a simple button-press study they did a covariation for the global signal and found out that the global signal reduced intensity and also introduced new negatively activated areas. The interesting thing is that those areas lie within the default mode network.

Introduction Global signal regression can cause reductions in sensitivity and introduce false deactivations in resting state data experimental condition is undefined exact timing, spatial extent and relative phase between areas are unknown correlation between global signal and resting state fluctuations cannot be determined this could lead to wrong results in seed voxel correlation analyses Global signal regression can cause reductions in sensitivity and introduce false deactivations when experimental induced activations contaminate global signal. By definition the task is undefined in resting state data therefore the exact timing, the spatial extent and the relative phase between the areas are unknown and correlation between global signal and resting state fluctuations can‘t be calculated. That could lead to wrong results in seed voxel correlation analyses.

Introduction seed voxel analyses 1 time series (hypothesized fluctuations of interest) correlate with every other voxel Studies have used global signal regression default mode network = task negative network anti-correlated network = task positive network If global signal is uncorrelated with resting state fluctuations then finding is correct If not  brain may not be organized into anti-correlated networks

Introduction How does global signal regression affect seed voxel functional connectivity analyses? different aspects of resting state fluctuations theory  global signal regression in seed voxel analyses always results in negative mean correlation value (math) simulation  empirical demonstration… breath-holding and visual task visual task – localisable connectivity maps breath-holding as comparatively global fluctuation resting state scans

General linear model voxel-wise GLM is expressed by Y =βX+ε Y … column vector of N rows X … design matrix with N rows × p columns – regressors β… column vector with p rows - unknown parameters associated to each regressor ε … column vector, with N rows, estimation error (residuals) The voxel-wise GLM is expressed as: Y =βX+ε (1) Y is a column vector of N rows (the number of collected time-points) representing the time-series BOLD signal associated to a single voxel. X represents the design matrix with N rows × p columns, each representing a regressor (i.e. an explanatory variable). Of interest are the columns representing manipulations or experimental conditions, although the matrix often may include regressors of non-interest, modelling the mean signal (i.e. the intercept), trends (typically linear and quadratic) and other design specific confounds. β is a column vector with p rows representing the unknown parameter associated to each regressor. Finally, ε is also a column vector, with N rows, representing the estimation error (or residuals) defined as Y − βˆX.

Theory Si(t) ... voxel‘s time series g(t) ... global signal βi ... regression coefficient xi(t) … time series after global signal regression

Theory After Global Signal Regression, the sum of correlation value of a seed voxel across the entire brain is less than or equal to 0 For all voxels that correlate positively with the seed, negatively correlated voxels must exist to balance the equation.

Simulations Matlab 1000 time series 2 time courses Resting state fluctuations generated by sine wave, randomly choosen frequency Gaussian noise added (global) Each time serie‘s global signal regressed with GLM

Simulation Results high SNR low SNR phase shifting – sine wave and gaussian noise is shifted

Breath holding & visual data 8 adults scanned on 3T scanner (27 sagittal slices) Pulse oximeter Pneumatic belt

Breath holding & visual data

Breath holding & visual data 5 conditions VisOnly = 30s OFF (fixation) / 20s ON (flashing checkerboard) Synch 30s countdown – „breath in (2s)“, „breath out“ (2s) then breath holding & checkerboard Synch+10 = like above but 10s delayed checkerboard Asynch = visual ON period ended when breath holding ON commenced??? RandVis = event-related design var. ISI, each second 50% probability of checkerboard

Breath holding & visual data Preprocessing AFNI (Cox, 1996) RETROICOR (remove cardiac and repiration effects) Correction of timing for slices bandpass filtering (0.01 Hz – 0.1 Hz) 1 Dataset with GLM | 1 Dataset without GLM

Breath holding & visual data While in the condition which only presented the checkerboard (VisOnly) global signal resgression doesn‘t change too much about the activation. In the other conditions global signal regression result in anticorrelated areas. The highest impact can be seen in the Synch+10 condition, where the „resting state“ fluctuations are similiar to the global signal.

Resting state data 12 subjects – 2 resting state scans (5 min) correlation maps from seed region in posterior cingulate/precuneus (PCC) with global signal removed without global signal removal with RVT (respiration volume per time) correction voxels correlating with PCC ROI  task-negative network

Resting state data

Resting state data

Conclusions Mathematically global signal regression forces half of the voxels to become anti-correlated On data with known respiration confound (global signal) global signal regression not effective in removing noise & location of anti-correlated effect is dependent on relative phase of global and seed voxel time series In resting state data, anti correlated networks are not evident until global signal regression