Folie 1

1
Multimodal Imaging MEG-NIRS integration
Neuro-vascular coupling Near-infrared
spectroscopy Magnetoencephalography and
DC-magnetoencephalography Signal processing
Independent component analysis Results Summary
T.H. Sander, M. Moeller, S. Leistner, A. Liebert,
H. Wabnitz, M. Burghoff, G.Curio, B.M. Mackert,
R. Macdonald, L. Trahms
Physikalisch- Technische Bundesanstalt
2
Neuro-vascular coupling
3
Neuro-vascular coupling
Cortical neuron
Active neuron in the brain consumes energy
supplied by blood oxygen Neuro-vascular
coupling. Why to study this ? - Neuro-vascular
coupling is basis of functional magnetic
resonance imaging (fMRI). - Diagnostical
applications, e.g. stroke. How to study this ?
Blood vessel system in the brain
4
Neuro-vascular coupling
5
Near-infrared spectroscopy (NIRS)
6
Near-infrared spectroscopy

• Brain activation
• ? vascular response DcHbO2, DcHb
• ? wavelength dependent absorption changes

7
Near-infrared spectroscopy
HbO2 Hbtot Hb
Rest Fast Rest Slow
8
Time resolved NIRS
Measure distribution of time of flight (DTOF) at
three wavelengths
Skull Cortex
Extract attenuation (DA) and mean time of flight
(lttfgt) during rest and movement sections (t
experiment time)
9
Time resolved NIRS
Analysis based on moments
D Ntot
extracerebral intracerebral
systemic focal
changes
DV
10
Time resolved NIRS Setup
3 cm
NIRS optode and prism holder attached to head,
the optical fibers are plugged into the holder.
11
Time resolved NIRS Setup
Shielded room
12
Magnetoencephalography and DC-MEG
13
Magnetoencephalography
Magnetic field B Measured using SQUIDs (brain
fields 100-500 fT, earth 40 mT).
14
Typical MEG result N20m
Source identified by ICA
Peaks at stimulus rate of 3 Hz and multiples
15
Neuro-vascular coupling How to quantify ?
0.01 0.1 1
10 100 f / Hz
Bandwidth of neuron communication and brain
currents
Bandwidth of oxygen saturation changes in blood
16
DC-Magnetoencephalography
Modulation DC-MEG (mDC-MEG) in a three layer
shielded room
Patient bed modulation frequency 0.4 Hz -gt
bandwidth DC - 0.2 Hz. Installed in a hospital.
17
Modulation DC-Magnetoencephalography
Correlation of MEG signal with bed position
Correlation amplitudes Ak.
Distributed virtual moving source fitted to Ak
Multichannel DC field sampled in 2.5 s
intervalls.
Independent component analysis on DC
field Source pattern and amplitude function.

Rest Move
18
Direct DC-MEG of Motor Activity
ICA time series (averaged over 30 epochs)
component map
19
Combined DC-MEG and NIRS
DC-magnetoencephalography sensor
3 cm
NIRS optode and prism holder.
Experiment DC-MEG and NIRS over motor cortex
contralateral to moving fingers (C3). Motor
paradigm (repeat 30 times) rest
finger movement rest t 0
30 60
90 s
20
Signal processing Independent Component Analysis
21
Independent Component Analysis
MEG Sensors
Independent component analysis tries to estimate
the simultaneously active sources using their
statistical independence.
22
Independent Component Analysis
Describe the data using a vector base
ai and amplitude functions fi(t)
ICA Search for statistically independent
amplitude functions, i.e.
, where p(.)
probability density function. Matrix A and set
of amplitude functions fi are unknown ! A second
order decorrelation algorithm (SOBI, TDSEP) is
used for the ICA.
23
Independent Component Analysis TDSEP
Caculate time-delayed correlation matrices in
parallel
Computer-Cluster
90 fold speed increase between single PC Matlab
implementation compared to 8 node cluster with
MPI C program using ATLAS library.
24
Variables for statistical analysis NIRS
DA, lttfgt
Treat each wavelength and DA and lttfgt as
independent 24 variables.
25
Classification of ICA result by demixing
Demixing Apply inverse of matrix A to average
26
Results Group study on normal subjects (NIRS
and mDC-MEG) First results for unmodulated
DC-MEG combined with NIRS Feasibility study with
subacute stroke patients (NIRS and mDC-MEG)
27
Unaveraged DC-MEG and NIRS data
finger movement
l 826 nm
NIRS DA
DC-MEG
M04
0 200 400
600 800
1000
t / s
A haemodynamic response function DA(t) HRF(t)
? DC-MEG(t)
28
Averaged mDC-MEG and NIRS data
Rest Movement Rest
DC-MEG
B HbO2 Hb
trNIRS

Group result
29
Neuro-vascular loop
X
Y
30
Neuro-vascular loop Direct DC-MEG
X
Y
Vascular Attenuation /
Neuronal Magnetic field / fT
31
Feasibility study Patient data
mDC-MEG NIRS Attenuation
As measured After ICA denoising
32
Feasibility study Comparison
Normal subject left hemisphere right
hemisphere
HbO2 Hb
Patient left hemisphere right hemisphere
HbO2 Hb
Cortical infarction of left hemisphere, mild
paresis of right hand.
33
Summary
Combined DC-MEG and time resolved NIRS technique
demonstrated. Independent Component Anaylysis
essential for signal conditioning. Neuro-vascular
coupling characterised for group of healthy
subjects. Neuro-vascular loop introduced.
Outlook NIRS imager with 9 sources and 4
detectors Estimate haemodynamic response
function with high temporal resolution
Multi-modal independent component analysis
combining NIRS and DC-MEG
Supported by Berlin Neuroimaging Center