Resting-state fMRI + + no task or stimuli typical instructions:
keep eyes closed, or keep them open/fixation; dont fall asleep; let
your mind freely wander....
Slide 5
Resting-state fMRI resting-state signal fluctuations = ?
spontaneous neural activity (i.e., cannot be attributed to a task
or overt behavior) noise (hardware, motion, physiological...)
Slide 6
Functional connectivity analysis 0 0.8 r seed correlate seeds
time series with every other voxels time series threshold seed 0.3
0.8 r We can analyze relationships between the time series of
different brain regions E.g., seed-based correlation analysis:
Slide 7
Functional connectivity We can analyze relationships between
the time series of different brain regions Biswal et al. 1995 time
series during resting-state scan Signals from different regions
have correlated resting-state activity Regions that are correlated
tend to be functionally related
Slide 8
Resting-state networks have a close correspondence with
task-activation networks TaskRest TaskRest TaskRest Smith et al.
2010
Slide 9
Resting-state networks Rocca et a. 2012 resting-state
functional connectivity: phenomenon of correlated resting-state
fluctuations between remote brain areas resting-state networks
(RSN): set of regions with mutually high functional connectivity in
resting state
Slide 10
Implications task-free mapping of functional networks? query
multiple networks from the same dataset can be used when task
performance is not possible (fetus, coma,...) potential biomarker
of healthy & diseased brain resting-state functional
connectivity may reflect functional organization and dynamics
Meunier et al. 2011
Slide 11
Challenges Resting-state networks look real... but could also
arise due to: noise (hardware, physiology) vascular pulsation
hidden tasks: conscious thoughts, actions, sensation, etc. causing
activation within functional systems The terms FC and RSN are
purely descriptive Understanding of origins &mechanisms is
still limited Evidence that these are not trivially due to the
above
RSNs are (mostly) conserved across sessions, individuals,
states, species,... suggests not arising solely from conscious
processes Infants Monkeys Rats Horovitz et al. 2008; Vincent et al.
2007; Lu et al. 2007; Doria et al. 2010 Sleep
Slide 14
Default Mode Network higher activity during passive baseline
conditions comapred to (most) tasks Raichle at el., 2001 review:
Buckner et al., Ann. N.Y. Acad. Sci. 2008 Greicius et al. 2003
functional connectivity in resting state
Slide 15
Coherence in spontaneous electrophysiological signals Kenet et
al, 2003 spontaneous fluctuations in membrane voltage resemble
orientation columns & evoked activity
Slide 16
Simultaneous LFP-fMRI of resting-state fluctuations Shmuel
& Leopold, 2008 gamma power fluctuations in local field
potential (LFP) found to correlate with fMRI signal correlations
are spatially widespread! Scholvinck et al., 2010
Slide 17
Human ECoG of resting-state activity Keller et al. 2013 also
with slow cortical potential (He et al, 2010) macaque ECoG reveals
broadband phenomenon (Liu et al. 2014) How well do networks of
electrical signals match networks of BOLD fMRI?
Slide 18
Functional connectivity at finer spatial scales Buckner et al.
2011 Kim et al. 2013 Beckmann et al. 2005
Slide 19
Quigley et al., 2003 task activation resting-state functional
connectivity Johnston et al., 2008 Structrual connectivity affects
functional connectivity via indirect connections?
Slide 20
Clinical applications Healthy control Alzheimers Schizophrenia
Greicius et al. 2004, Whitfield-Gabrieli et al. 2009, Lewis et al.
2009 Altered functional connectivity found in a range of
neurological & psychiatric disorders Affects expected regions
and may relate to severity of disease Potential for classifying
patients vs. healthy controls No task necessary; can be used for
patients, coma,...... Underpinnings of altered functional
connectivity need further investigation
Seed-based correlation analysis 0 0.8 r seed correlate seeds
time series with every other voxels time series threshold 0.3 0.8 r
network Requires a priori seed (hypothesis) How define the seed
(atlas? functional localizer?) sensitivity of results to exact
size/placement Straightforward intepretation
Slide 23
Independent component analysis Cocktail party problem N
microphones around a room record different mixtures of N speakers
voices How to separate the voices of each speaker?? time1 Observed
data time2 time3 ICA can be applied to unmix fMRI data into
networks Multivariate
Slide 24
Original Sound sources Cocktail party mixes Estimated sources
adapted from
http://research.ics.aalto.fi/ica/cocktail/cocktail_en.cgi
http://research.ics.aalto.fi/ica/cocktail/cocktail_en.cgi by Jen
Evans
Slide 25
Independent component analysis Cocktail party problem N
microphones around a room record different mixtures of N speakers
voices How to separate the voices of each speaker?? time1 Observed
data time2 time3 ICA can be applied to unmix fMRI data into
networks Multivariate
Slide 26
Decompose fMRI data into fixed spatial components (networks)
with time-dependent weights (network time courses) McKeown et al,
1998 Thomas et al, 2002 + + + = raw_data(t) time t: a N (t) a 1 (t)
a N-1 (t) a 2 (t) Spatial ICA
Slide 27
Independent component analysis Damoiseaux et al. 2006
Slide 28
ICA + very helpful for exploring structure of data! +
multivariate; doesnt require choice of seed + useful for de-noising
(but wont completely remove it) -need to specify parameters (e.g. #
components) -interpretation difficult Review: Cole et al. 2010:
Advances and pitfalls in the analysis and interpretation of
resting-state fMRI data
Slide 29
Network analysis e.g. SEM, DCM, Granger causality, partial
correlation complex network analysis Review: Smith et al. 2013,
TICS: Functional connectomics from resting-state fMRI Review:
Rubinov & Sporns, 2011 Bullmore & Sporns, 2012 Wig et al.
2011
Resting state: signal vs. noise? No model (timing of
task/stimuli) No trial averaging Considers relationships between
the voxel time series themselves (signal + noise) stimulus
Slide 32
Thermal noise Slow drifts (magnet instability; gradient
heating) Head motion Physiological processes (respiration, cardiac)
Noise in fMRI
Slide 33
BOLD signal (whole-brain average) Respiration Breathing
variations affect BOLD signal Respiratory variations (RVT) changes
in [CO2], HR, blood pressure hemodynamic response uncoupled from
local neural activity
Slide 34
Birn et al. 2006 Changes in rate / depth of breathing over time
correlate with BOLD signal Common influence over many regions
creates false positive correlations
Slide 35
Chang et al., 2009 Reducing physiological noise whole-brain
average fMRI signal in task-free scan predicted fMRI signal derived
from respiration measuremen Model-based approaches: estimate noise
based on physiological measurements (e.g. RETROICOR, RETROKCOR,
RV/HRCOR..). Data-driven approaches: estimate noise from the data
itself e.g. CompCor, FIX, PESTICA,...
Slide 36
anti-correlated resting state networks...? Fransson 2005, Fox
et al, 2005 Global signal regression Murphy et al, 2009 are
anticorrelations state-dependent?
Slide 37
State-related variability Resting (undirected) Recalling
memories Shirer et al, 2011 Horovitz et al., 2009 eyes closed eyes
open/fixation Eyes open/closed Bianciardi et al., 2009
Slide 38
State-related variability Caffeine can influence resting-state
correlations Wong et al. 2010 Fluctuations in alertness/drowsiness
modulate FC Chang et al. 2013
Slide 39
Dynamic resting-state analysis Can we extract more information
by moving beyond static / average corrlelation? Allen et al. 2012
+
Slide 40
Xiao Liu et al. 2013
Slide 41
Variability: discussion Resting-state signals and correlations
vary over time Sources: cognitive/vigilance state, noise,
spontaneous. Consider when interpreting group differences What time
scales to study / how long to scan? Why study variability? model
within-scan variance neural basis of natural state changes
(drowsiness, emotion.) learn about dynamics of brain activity
Simultaneous recordings (EEG, physiology) during resting state can
help
Resting-state fMRI is proving valuable for clinical
applications and basic neuroscience RSNs relate to anatomic
connectivity and electrophysiology, but precise relationship still
not clear Understand analysis methods/tradeoffs no single correct
analysis of resting-state data avoid bias, fishing Noise can skew
connectivity estimates clean up the signal as best as possible! See
future lecture There can be substantial within-scan variability
need to understand these effects, determine what information is
valuable
Slide 44
Thanks! AMRI group: Jeff Duyn Xiao Liu Dante Picchioni Jacco de
Zwart Peter Van Gelderen Natalia Gudino Roger Jiang Xiaozhen Li
Hendrik Mandelkow Erika Raven Jennifer Evans Dan Handwerker Peter
Bandettini Gary Glover Mika Rubinov Zhongming Liu