analysis and characterization of the physiological noise in ......i. introduction • fmri signals,...
TRANSCRIPT
University of Rome, La Sapienza Department of Physics
G1 Group
Analysis and characterization of the physiological noise in Resting state fMRI measurements
1 July 2010 1
I. INTRODUCTION• fMRI signals,• fMRI Noise,• Physiological Noise,• Techniques of Reduction Physiological Noise,• Resting State.
II. Physiological Noise Characterization
III. Graphic Theory Analysis
Outline
IMAGE ACQUISTIONIMAGE PREPROCESSING RESULTS
GRAPH THEORYMETHODRESULTS
1 July 2010 2
IINTRODUCTION
1 July 2010 3
Magnetic susceptibility difference across the blood vessels
1 July 2010 4
5
Change in CBF appear to more than is necessary to support the small increase in oxygen metabolism (∆CBF>>∆CMRO2)
1 July 2010
CBF
CMRO2
Task Task OnOn
Time
BOLD Signal
Deoxyhaemoglobin change
1 July 2010
Time courses of CBF and CMRO2 that would predict an early neagative BOLD response 6
I. INTRODUCTION
The signals of interest include:
task-related signal, function-related signal, and transiently task-related signal.
The signals not of interest include:
physiology-related signal, motion related signal, and scanner-related signal.
Functional magnetic resonance imaging (fMRI) is a noninvasive technique that uses Blood oxygen level dependent (BOLD) effect to explore neural activity.
1 July 2010 7
I. INTRODUCTION
Thermal Noise
System Noise
Motion Noise
External Noise
Physiological Noise
................. etc1 July 2010
Changes in signal intensity over time due to thermal motion of electrons within the subject and within the scanner electronics.
Arises as a result to static field inhomogeneities, nonlinearities and instabilities in the gradient field and variations in imaging hardware.
Severe head motion during an fMRI scan can severely corrupt the data e.g. eyes, head….etc
Interference from outside sources can also lead to distortions and artefacts in the data.
Cardiac and respiration noise.
Acoustic, Non- task related neural variability, .......
8
Cardiac
Resp.
Frequency (Hz )
cardiac here (~0.8Hz)
Resp. here (~0.15Hz)
Glover et al, 2000
fMRI Techniques and Protocols,2009
I. INTRODUCTION
TR= 250 msec.
1 July 2010 9
There are several methods have been developed for reducing such physiological noise in fMRI time series, including:
navigator echo (Hu X, Kim SG, 1994),
retrospective gating (Hu X, et al, 1995),
digital filtering (Biswal B, et al, 1996),
k-space based estimation and correction (Wowk B, et al 1997),
pulse sequence gating (Guimaraes AR, et al, 1998),
motion-ordered data acquisition (Stenger VA, et al, 1999),
RETROICOR (Glover GH, et al, 2000),
IMPACT (Chuang KH, et al, 2001),
CORSICA (perlbarg V, et al, 2007),
RVHRCOR (Chang G,et al, 2009)
I. INTRODUCTION
1 July 2010 10
Fox MD et al, PNAS 2005; 102, 9673-9678Greicius M et al, PNAS 2003; 100, 253-258Raichle M et al, PNAS 2001; 98, 676-682
I. INTRODUCTION
0.01 Hz < f < 0.1 Hz
1 July 2010 11
Intrinsic Correlations between PCC and all other Voxels in the Brain during Resting State
Resting state networksM. De Luca,2006;29, 1359-1367
IIPhysiological Noise Characterization
1 July 2010 12
II. Physiological Noise Characterization
ICA-based artifact removal in functional connectivity analysis
• Subjects: 13 healthy right-handed subjects (27±7 years), healthy and right handed
• MRI Scanner: 3T Allegra (Siemens, Erlangen, D)
EPI-SIEMENS:
• TR/TE=2100/30 ms
• 3x3x2.5 mm3
• 32 slices
• 240 vol
20 40 60 80 100 120 140 160 180 200 220 240
1 July 2010 13
Voxel’s time series
II. Physiological Noise Characterization
ICA-based artifact removal in functional connectivity analysis
Movement Regression, Temporally band-pass filtered
(0.009 < f < 0.15), Detrend,Centering,Whiting
Statistical Parametric Mapping (SPM8)http://www.fil.ion.ucl.ac.uk/spm/
1 July 2010 14
II. Physiological Noise Characterization
ICA-based artifact removal in functional connectivity analysis
Create CSF & WM Images (New approach)
CSF common mask WM common mask
1 July 2010 15
II. Physiological Noise Characterization
ICA-based artifact removal in functional connectivity analysis
Extract CSF & WM Time Series (using ICA)
• 20 CSF time series
• 20 WM time series
fMRI Images CSF ImagesWM Images
CSF Time seriesWM Time series
CSF mask
WM mask
ICA
ICA
1 July 2010 16
ICA-based artifact removal in functional connectivity analysis
II. Physiological Noise Characterization
“Independent component analysis (ICA) is a method for finding underlying factors or components from multivariate (multidimensional) statistical data. What distinguishes ICA from other methods is that it looks for components that are both statistically independent, and nonGaussian.”
A.Hyvarinen, A.Karhunen, E.Oja‘Independent Component Analysis’1 July 2010 17
II. Physiological Noise Characterization
ICA-based artifact removal in functional connectivity analysis
=
fMRI Data
Tim
e
Space # IC
Tim
e
x = As
# IC
Space
Mixing Matrix Spatial MapsN x K
N= Num. of Scans
K= Num. of IC
N x M
N= Num. of Scans
M= Num. of Voxels
K x M
K= Num. of IC
M=Num. of Voxels1 July 2010 18
II. Physiological Noise Characterization
ICA-based artifact removal in functional connectivity analysis
Avoid the Global Signal Effect.
Regress out CSF and WM from fMRI data .
Create Region of Interest (ROI)
PCC (0, -40, 30)
0 50 100 150 200-0.5
0
0.5
Images Number
PCC ROI1 July 2010
Average time series of ROI19
II. Physiological Noise Characterization
ICA-based artifact removal in functional connectivity analysis
Create Z- maps.⎟⎟⎠
⎞⎜⎜⎝
⎛−+
∗=),,(1),,(1log5.0),,(
zyxrzyxrzyxZ
0.3
0.09
Z-map for first subject
Group analysis (for 13 Subjects)
1 July 2010 20
II. Physiological Noise Characterization
ICA-based artifact removal in functional connectivity analysis
Group analysis (N=13) of regions having significant positive correlations (p<0.05 FWE) with the PCC.
none
Global
WM/CSF
CSF
WM
1 July 2010 21
II. Physiological Noise Characterization
ICA-based artifact removal in functional connectivity analysis
none
Global
WM/CSF
CSF
WM
Regions having negative correlation (p<0.05 uncorrected) with the PCC in a group-level (N=13).
1 July 2010 22
II. Physiological Noise Characterization
ICA-based artifact removal in functional connectivity analysis
WM CSF WM/CSF Global-60
-50
-40
-30
-20
-10
0
perc
ent o
f pos
itive
ly c
orre
late
d vo
xels
%
WM CSF WM/CSF Global100
101
102
103
104
105
perc
ent o
f neg
ativ
ely
cor
rela
ted
voxe
ls %
Spatial extent of (A) positive (r>0.15) and (B) negative (r<−0.15) correlations.
A B
1 July 2010 23
II. Physiological Noise Characterization
ICA-based artifact removal in functional connectivity analysis
CSF
CSF/WM
2
0
2
0
2
0WM
F -maps for single Subject
1 July 2010 24
II. Physiological Noise Characterization
ICA-based artifact removal in functional connectivity analysis
•Regions having negative correlation after WM/CSF correction overlapped substantially
with those having the greatest negative correlation magnitudes after global signal removal.
• The spatial extent of significant positive correlations is diminished after WM/CSF
correction is applied, and even further diminished after global signal removal. CSF signal
removal and WM signal correction also introduced slight decreases in the extent of
positive correlations.
• Global signal removal greatly increased the negative correlations over more widespread
regions of the brain, as well as inter-subject variability.
1 July 2010 25
II. Physiological Noise Characterization
II. Physiological Noise Characterization
As a further step we would like to compare this approach with other approaches:
RETROICOR
CORSICA
RVHRCOR
Discussion the difference between physiological signal regression results and our results.
1 July 2010 26
IIIGraphic Theory Analysis
1 July 2010 27
III. Graphic Theory Analysis
node
edge
Van den Heuvel et al., 2010
• Using graph theory, functional brain networks can be defined as a graph G=(V,E).
Functional brain network
• To characterize the topological properties of a network, a number of parameters have
been described:
• Node Degree (K),
• Degree Density,
• Degree Distribution,
• Clustering coefficient (Ci)
Ci= 2Ei / Ki(Ki-1)
C=1/N . ƩCi
• Path Length
• Other topological properties.
1 July 2010 28
Fox, et al. 2005
III. Graphic Theory Analysis
Intrinsically defined anticorrelated processing networks in the brain
1 July 2010 29
III. Graphic Theory Analysis
- 9 subjects.-13 ROI. “Fair, et al 2008”- Pajek software. “http://pajek.imfm.si/doku.php”
DMNCorrelation Matrix Topological propertiesGraph
Schematic illustration of the graph analysis.
1 July 2010 30
III. Graphic Theory Analysis
DMN_Rest graph DMN_Task graphNodes
Edges
1 July 2010 31
III. Graphic Theory Analysis
aMPFC L.Sup.F R.Sup.F vMPFC L.IT R.IT L.PHC R.PHC PCC Rsp L.LatP L.LatP R.LatP0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
ROI
Clu
ster
ing
Coe
ff.
DMN-TaskDMN-Rest
Clustering coefficient for first subject (N=1)
1 July 2010 32
III. Graphic Theory Analysis
1 2 3 4 5 6 7 8 90
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Subjects
Average clustering coefficient
R-DMNT-DMN
mean±SDR-DMN= 0.2526 ± 0.0533T-DMN= 0.2221± 0.0384
Average clustering coefficient for all subject (N=9)
1 July 2010 33
III. Graphic Theory Analysis
aMPFC L.Sup.F R.Sup.F vMPFC L.IT R.IT L.PHC R.PHC PCC Rsp L.LatP R.LatP Cereb0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
DMN nodes
DMN-Task DMN-Rest
Average clustering coefficient for each ROI (N=9)
1 July 2010 34
III. Graphic Theory Analysis
aMPFC L.Sup.F R.Sup.F vMPFC L.IT R.IT L.PHC R.PHC PCC Rsp L.LatP R.LatP Cereb0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
DMN nodes
DMN-Task DMN-Rest
Degree value for each ROI (N=9)
1 July 2010 35
The analysis of DMN topological properties demonstrates that the overall effect of the cognitive load is to weaken the network connectivity, without inducing a reduction of the cortical hubs participating to the network itself. Because of a decrease of the node degree for almost all the nodes, during working memory with respect to the resting condition, it can be speculated that the communication within the network is reduced in behalf of the best performance during the task. As already reported, among all the nodes, the PCC seems to be the center of the information processing within the DMN.
III. Graphic Theory Analysis
1 July 2010 36
III. Graphic Theory Analysis
III. Graphic Theory Analysis
We are currently working on the description of the scale-free distribution of functional connections,
Study Community structure in networks of functional connectivity,
Extract the restating state networks directly from fMRI signals at voxel resolution.
1 July 2010 37
1 July 2010 39
Matrix Formulation
Equation for scan j
Simultaneous equations forscans 1..N(J)
…that can be solvedfor parameters β1..P(L)
1 July 2010 40
Regressors
Scan
s
+
error
vecto
r
ε+
data ve
ctor
y
=
design
matr
ix
= X
β1
β2
β3
β4
β5
β6
β7
β8
β9
param
eters
β×
1 July 2010 41
1 July 2010 42
this line is a 'model' of the data
µ
slope β
ε
Y = βx + µ + ε
Y
X
Brain region Abbreviations MNI Coordinates1. Medial prefrontal cortex (anterior) aMPFC -3,54,18 2. Left superior frontal cortex L.Sup.F -15,54,423. Right superior frontal cortex R.Sup.F 18,42,484. Medial prefrontal cortex (ventral) vMPFC -6,36,-95. Left inferior temporal cortex L.IT -60,-9,-246. Right inferior temporal cortex R.IT 57,0,-277. Left parahippocampal gyrus L.PHC -24,-18,-278. Right parahippocampal gyrus R.PHC 27,-18,-249. Posterior cingulate cortex PCC -3,-48,3010. Retrosplenial Rsp 9,-54,1211. Left lateral parietal cortex L.LatP -48,-69,3912. Right lateral parietal cortex R.LatP 48,-66,3613. Cerebellar tonsils Cereb -6,-54,-48
SEED REGIONS FOR DEFAULT NETWORK
1 July 2010 43