Transcript 09_MEEG_DCM_etc - Wellcome Trust Centre for Neuroimaging

The other stuff
Vladimir Litvak
Wellcome Trust Centre for Neuroimaging
UCL Institute of Neurology, London, UK
• SPM resources
• Fieldtrip in SPM8
• MEEGTools and Beamforming
• DCM
Image time-series
Realignment
Spatial filter
Design matrix
Smoothing
General Linear Model
Statistical Parametric Map
Statistical
Inference
Normalisation
Anatomical
reference Parameter estimates
RFT
p <0.05
Software: SPM8
• Open Source academic freeware (under GPL)
• Documented and informally supported
• Requirements:
– MATLAB: 7.1 (R14SP3) to 7.11 (R2010b)
no Mathworks toolboxes required
– Supported platforms (MEX files):
Linux (32 and 64 bit)
Windows (32 and 64 bit)
Mac Intel (32 and 64 bit)
– File Formats:
• Images: NIfTI-1 (& Analyze, DICOM)
• Surface meshes: GIfTI
• M/EEG: most manufacturers (with FieldTrip’s fileio)
SPMweb
• Introduction to SPM
• SPM distribution:
SPM2, SPM5, SPM8
• Documentation &
Bibliography
• SPM email discussion
list
• SPM short course
• Example data sets
• SPM extensions
http://www.fil.ion.ucl.ac.uk/spm/
SPM Toolboxes
• User-contributed SPM extensions:
http://www.fil.ion.ucl.ac.uk/spm/ext/
SPM Documentation
Peer reviewed literature
Online help
& function
descriptions
SPM Books:
Human Brain Function I & II
Statistical Parametric Mapping
SPM Manual
SPM Online Bibliography
External Resources
• SPM @ Wikipedia
http://en.wikipedia.org/wiki/Statistical_parametric_mapping
• SPM @ Scholarpedia
http://www.scholarpedia.org/article/SPM
• SPM @ WikiBooks
http://en.wikibooks.org/wiki/SPM
• MRC-CBU Imaging/MEG wiki
http://imaging.mrc-cbu.cam.ac.uk/imaging/CbuImaging
http://imaging.mrc-cbu.cam.ac.uk/meg
• SPM @ NITRC
http://www.nitrc.org/projects/spm/
SPM Mailing List
• [email protected]
• Web home page
– http://www.fil.ion.ucl.ac.uk/spm/support/
– Archives, archive searches, instructions
• Subscribe
– http://www.jiscmail.ac.uk/
– email [email protected]
– join spm Firstname Lastname
• Participate & learn
– email [email protected]
– Monitored by SPMauthors
– Usage queries, theoretical discussions,
bug reports, patches, techniques, &c…
http://www.fil.ion.ucl.ac.uk/spm/support/
[email protected]
FieldTrip
Powered by:
http://fieldtrip.fcdonders.nl/
What is FieldTrip?
A MATLAB toolbox for electrophysiological data
analysis
Features: high-level functions for
electrophysiological data analysis
Data reading
all commercial MEG systems, many different EEG systems
Preprocessing
filtering, segmenting
Time-locked ERF analysis
Frequency and time-frequency analysis
multitapers, wavelets, welch, hilbert, parametric spectral
estimates
Features: high-level functions for
electrophysiological data analysis
Functional connectivity analysis
coherence, phase locking value, granger causality,
and many more
Source reconstruction
beamformers, dipole fitting, linear estimation
Statistical analysis
parametric, non-parametric, channel and source level
All other operations that are required around it
But…
Features
Analysis steps are incorporated in functions
ft_preprocessing
cfg = [ ]
cfg.dataset = ‘Subject01.ds’
cfg.bpfilter = [0.01 150]
ft_rejectartifact
...
rawdata = ft_preprocessing(cfg)
ft_freqanalysis
ft_multiplotTFR
ft_freqstatistics
ft_multiplotTFR
FieldTrip toolbox - code reused in SPM8
SPM8 end-user
perspective
SPM8 main functions
main functions
with graphical user interface
fileio
distrib.
forwinv preproc
comput.
public
fieldtrip
private
Fieldtrip-SPM8 integration
• Full version of Fieldtrip is contained in SPM8 under
/external/fieldtrip.
• Fieldtrip raw, timelock and freq structures can be
converted into SPM8 datasets with spm_eeg_ft2spm.
• D.ftraw and D.fttimelock can be used to export SPM
dataset to Fieldtrip raw and timelock/freq structs
respectively.
• Fieldtrip and SPM share common forward modelling
framework. Head models created in SPM can be used in
Fieldtrip.
Fieldtrip-SPM8 integration – the future
• Time-frequency analysis will be done using
shared code.
• Matlabbatch interface as in SPM will be created
for all top-level Fieldtrip function, so Fieldtrip will
have GUI for the first time.
• Matlabbatch and distributed computing toolbox
from FieldTrip will be combined for easy-to-use
job parallelization framework that will work with
both FieldTrip and SPM.
MEEGTools
•MEEGTools toolbox includes some useful functions
contributed by SPM developers and power users.
•Many of these functions combine SPM and FieldTrip
functionality.
•Other functions solve system-specific problems that cannot
be handled in by the main SPM code.
Beamforming
•
Functions in the beamforming toolbox make it
possible to perform source reconstruction using
beamforming methods in the time and
frequency domains and extract source activity
using beamformer spatial filters.
•
They make use of SPM-generated forward
models (see ‘Source reconstruction’) and
(where relevant) generate images that can be
entered into the SPM statistics pipeline.
•
Some of these functions are based on FieldTrip
code and others are being developed by Gareth
Barnes at the FIL.
•
We are now working on optimizing these
functions for Neuromag but this is still in
progress.
DCM for fMRI
DCM for fMRI
Single region
u1
u1
c
u2
a11
z1
z1
z2
stimulus functions
u
t
activity
x(t )
neural state equation
vasodilatory signal
s  x  s  γ( f  1)
f
s
s
hemodynamic state
equations
flow induction (rCBF)
f  s
f
changes in volume
τv  f  v
1 /α
v
changes in dHb
τq  f E ( f,E0 ) qE0  v1 /α q/v
q
v
BOLD signal
y (t )   v, q 
Estimated BOLD
response
Modelled neural
activity
Predicted
BOLD
Predicted
BOLD
+ noise
=
observed data
Multiple regions
u1
c
a11
z1
u2
a21
z2
a22
u1
z1
z2
Modulatory inputs
u1
u2
c
u1
a11
z1
b21
a21
z2
a22
u2
z1
z2
Reciprocal connections
u1
u2
c
u1
a11
z1
b21
a12
a21
z2
a22
u2
z1
z2
Bayes‘ Theorem
new data
p( y |  )
prior information
p ( )
p( | y )  p( y |  ) p( )
posterior
 likelihood
∙ prior
Reverend Thomas Bayes
1702 - 1761
“Bayes‘ Theorem describes how an ideally rational
person processes information."
Wikipedia
Bayesian model inversion
•
Knowing the probability of data given the
model (which is something we can define)
Bayes rule makes it possible to compute the
probability of model parameters given the
data.
•
This requires specifying prior beliefs about
the parameters values.
•
Bayes rule is a mathematically optimal way to combine prior knowledge and
information derived from the data.
•
Model parameters will be moved from their prior values only if there is a need for
it to fit the data. Thus, in a model with many parameters we can make inferences
just about those that are important.
•
Bayesian model evidence, approximated by a quantity called ‘free energy’ is a
single number combining a measure of ‘goodness of fit’ of a model with
‘complexity penalty’. It allows comparing different models for the same data.
Accuracy
F=-
+
Complexity
Summary: the outputs of DCM
• Predicted data as similar as possible to the real
data.
• Posterior values of models parameters and
posterior precisions (measures of confidence
about those values).
• Free energy value (F) which can be used to
compare models fitted to the same data.
macro-scale
meso-scale
micro-scale
external granular
layer
external pyramidal
layer
internal granular
layer
internal pyramidal
layer
AP generation zone
Daunizeau et al. 2009, NeuroImage
David et al. 2006, NeuroImage
Kiebel et al. 2006, NeuroImage
Moran et al. 2009, NeuroImage
synapses
Spatial model
Depolarisation of
pyramidal cells
x0
L
L

Sensor data
y
Spatial model
Kiebel et al., NeuroImage, 2006
Daunizeau et al., NeuroImage, 2009
DCMs for M/EEG
input
 DCM for ERP
(+second-order mean-field DCM)
1st and 2d order moments
depolarization
250
0
0
200
-20
-20
150
-40
-40
100
-60
-60
50
-80
-80
0
0
100
200
300
time (ms)
auto-spectral density
LA
-100
0
100
200
300
time (ms)
-100
0
100
200
300
time (ms)
auto-spectral density
CA1
cross-spectral density
CA1-LA
frequency (Hz)
frequency (Hz)
 DCM for steady-state responses
frequency (Hz)
 DCM for induced responses
 DCM for phase coupling
Summary
• Dynamic Causal Modelling (DCM) is an approach
combining computational neuroscience and
neuroimaging data analysis.
• DCM makes it possible to estimate hidden
parameters from observable measurements given
a model that links between the two.
• Although there is complex theoretical background
behind DCM, its application is straightforward and
does not necessarily require mathematical training
or programming skills.
Thanks to
The people who contributed material to
this presentation:
• Guillaume Flandin
• Stefan Kiebel
• Robert Oostenveld
• Gareth Barnes
• Karl Friston
Thank you for your attention