Simultaneous EEG and f MRI Acquisition

Cohen 7/23/04

UCLA BrainMapping CenterIPAM - UCLA

Simultaneous EEG and fMRI Acquisition

Simultaneous EEG and fMRI Acquisition

Mark CohenUCLA Brain Mapping

Center

Mark CohenUCLA Brain Mapping

Center

Cohen 7/23/04


OutlineOutline

• Motivation: It’s hard, why bother?• Problems and Solutions• Early Results

Cohen 7/23/04


QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

Levels of UnderstandingLevels of Understanding

fMRI-EEG

Fiber Tracingregional connectivityFiber Tracingregional connectivity

fMRIfunctional nuclei or processing centersfMRIfunctional nuclei or processing centers

EEG & Autoradiographycell assembliesEEG & Autoradiographycell assemblies

Multi-unit Recordinglocal circuits: columns, retina…

Multi-unit Recordinglocal circuits: columns, retina…

Single Unit Electrophysiologyaction potentials, chemomodulation

Single Unit Electrophysiologyaction potentials, chemomodulationCrystallography, Chromatography (etc…)

transmitters, ion channels, membrane proteinsCrystallography, Chromatography (etc…)

transmitters, ion channels, membrane proteins

Cohen 7/23/04


Presumed Origin of the EEGPresumed Origin of the EEG-- -- -- --

++++

++++ ++

++

Skin

Bone

CSF

Cohen 7/23/04


Many Neurons are Not “Seen” by EEGMany Neurons are Not “Seen” by EEG

Bone

Skin

Bone

CSF

Cohen 7/23/04


General Limitations in EEG LocalizationGeneral Limitations in EEG Localization

• Deeper Sources Show Weaker Signals• Magnitude Depends on Dipole Orientation• Magnitude Depends on Temporal

Synchrony• Magnitude Depends on Spatial Coherence• Conductivity of Body Tissues (CSF, scalp)

Blur the Scalp Potentials

• Deeper Sources Show Weaker Signals• Magnitude Depends on Dipole Orientation• Magnitude Depends on Temporal

Synchrony• Magnitude Depends on Spatial Coherence• Conductivity of Body Tissues (CSF, scalp)

Blur the Scalp Potentials

Cohen 7/23/04


EEG Source LocalizationEEG Source Localization

1) High resolution raw dataa) High temporal resolution EEGb) High spatial resolution imaging

2) Computational Capacity3) Accurate Mathematical Models

a) Maxwell’s Equation forward model

b) Non-singular Inversion4) Accurate Physiological Parameters

a) Conductivity of all relevant tissues

1) High resolution raw dataa) High temporal resolution EEGb) High spatial resolution imaging

2) Computational Capacity3) Accurate Mathematical Models

a) Maxwell’s Equation forward model

b) Non-singular Inversion4) Accurate Physiological Parameters

a) Conductivity of all relevant tissues

≥10 kHz≤ 1 mm



after Massoud Akhtari

Cohen 7/23/04


EEG Source LocalizationEEG Source Localizationafter Massoud Akhtari

?

?

?

??

Cohen 7/23/04


The Scalp Waveform is DistortedThe Scalp Waveform is Distorted

MEG ECoG

Cohen 7/23/04


the Brain and Skull Exhibit Complex Impedancethe Brain and Skull Exhibit Complex Impedance

100 1000 Hz10

100 1000 Hz10

Con

duct

ivity

,

phas

e

The observed spectrally-dependent conductivity implies not only Resistive elements, but both Capacitive and Inductive energy storage.

R

R

L

C

Akhtari, unpublished

Such elements result in substantial phase lags that can alter the dipole localization by several millimeters.

Cohen 7/23/04


How does BOLD relate to neural firing?How does BOLD relate to neural firing?

Energy Demands in TransmissionPre-synaptic:

Transmitter SynthesisExocytosisTransmitter re-uptake

Post-SynapticMaintenance of membrane potential after ion leakageExcitatory: Removal of Sodium (Na/K pump)Inhibitory: ???

Energy Demands in TransmissionPre-synaptic:

Transmitter SynthesisExocytosisTransmitter re-uptake

Post-SynapticMaintenance of membrane potential after ion leakageExcitatory: Removal of Sodium (Na/K pump)Inhibitory: ???

Cohen 7/23/04


Red and Green SpikesRed and Green Spikes

EEG-correlated fMRI, New York

Seizure Activity Spreads from an Irritative Zone

Seizure Activity Spreads from an Irritative ZoneHypotheses:

•Initial Event is Energetically Costly

•Spreading Depolarization is Not

Hypotheses:

•Initial Event is Energetically Costly

•Spreading Depolarization is Not

Cohen 7/23/04


Spike-triggered ImagingSpike-triggered Imaging

Cohen 7/23/04


Spike-Triggered fMRISpike-Triggered fMRI

• Complex partial seizures, rare generalization

• EEG: generalized interictal discharges, some with left temporal onset

• MRI: normal

• Complex partial seizures, rare generalization

• EEG: generalized interictal discharges, some with left temporal onset

• MRI: normal

• Complex partial seizures, occasional generalization

• EEG: multifocal and generalized interictal discharges

• MRI: symmetric subependymal heterotopias

• Complex partial seizures, occasional generalization

• EEG: multifocal and generalized interictal discharges

• MRI: symmetric subependymal heterotopias

Warach, et al. (1996)

RR RR

Cohen 7/23/04


Project GoalsProject Goals

• Unaltered MR Image Quality• Diagnostic Quality EEG During functional

MRI:Artifact FreeDense Array of Channels

• Tomographic Correlation of Scalp Electrical Activity

• Amplifiers Suitable for Single Units• Subject Safety

• Unaltered MR Image Quality• Diagnostic Quality EEG During functional

MRI:Artifact FreeDense Array of Channels

• Tomographic Correlation of Scalp Electrical Activity

• Amplifiers Suitable for Single Units• Subject Safety

Cohen 7/23/04


Artifacts - MRIArtifacts - MRI

RF Noise

Magnetic Field DistortionNon-magnetic material such as Silver

• Properly-shielded Amplifiers

• Softened Logic Pulses

• Careful Lead Dress

• Eliminate RF Loops

Signal Losses

Cohen 7/23/04


Inductive Pickup by EEG leadsInductive Pickup by EEG leads

BallistocardiogramBallistocardiogram

Imaging Field GradientsImaging Field Gradients

++

––

Cohen 7/23/04


Ballistocardiogram SubtractionBallistocardiogram Subtraction

0 1 2 3time (sec)

QRS

EKG

A

B

A-B

Goldman, et al., Clinical Goldman, et al., Clinical Neurophysiology, 2000Neurophysiology, 2000

Cohen 7/23/04


ECG correctionECG correction

•Fourier detection of nominal heart rate

•Bootstrap detection of QRS based on template

•Adaptively create template of complete ECG

•Outlier and error monitoring

•Further detection based on statistical correlation peak

•Continuous adaptation of nominal HR and waveforms

Cohen 7/23/04


Approach to MR Artifact RemovalApproach to MR Artifact Removal

EEGEEGkk

SSkk=EEG=EEGkk++ArtifactArtifact

ArtifactArtifact

• EEGEEGkk and and ArtifactArtifact are are uncorrelateduncorrelated

• EEGEEGkk and and ArtifactArtifact add add linearlylinearly

• ArtifactArtifact is identical at each is identical at each time (time (kk))

This approach requires that:

kk=1=1

NN

SSkk

NN

–

+

Cohen 7/23/04


EEG Lead ConfigurationEEG Lead Configuration

Minimized Current Induction:• High Impedance Carbon Fiber Leads• Reduced loop area• Self canceling currents

Minimized Current Induction:• High Impedance Carbon Fiber Leads• Reduced loop area• Self canceling currents

Hard-Wired Bipolar Montage:• Dual lead electrodes• Twisted lead pairs• Local differential amplifiers

Hard-Wired Bipolar Montage:• Dual lead electrodes• Twisted lead pairs• Local differential amplifiers

A B C

+ + ––

Goldman, et al., Clinical Neurophysiology, 2000

Patent Applied For

Cohen 7/23/04


EEG Physical ApparatusEEG Physical Apparatus

Goldman, et al., Clinical Goldman, et al., Clinical Neurophysiology, 2000Neurophysiology, 2000Patent Applied For

Cohen 7/23/04


Twisted Lead Noise ReductionTwisted Lead Noise Reduction

twisted

untwisted

0 1 2time (sec)

fp2f8

c4p4

fp1f7

c3p3

no scan scan


Cohen 7/23/04


Amplifier RecoveryAmplifier Recovery

1 sec1 sec

Cohen 7/23/04


Receiver SaturationReceiver Saturation

++

––

ININ

2 Hz2 Hz

OUT OUT

150 mV150 mV

Time (sec)Time (sec)11 22

OUT OUT

150 mV150 mV

Time (sec)Time (sec)11 22

ININ

DC offsetDC offset

Cohen 7/23/04


EEG Functional Block DiagramEEG Functional Block Diagram

TwistedPair

Leads

DifferentialAmp

IsolationBarrier

ShieldDriver

+

–1 358

64

72

Low Pass

73

High Pass

Range detect

+

–

Patent Applied For

Cohen 7/23/04


After DC Offset control & Low Pass FilterAfter DC Offset control & Low Pass Filter

0 0.1 0.2 0.3sec

msec

100 µV

0 5 10 15 20 25

10 µV

Gradient Artifacts are not eliminated completely by analog means.

Cohen 7/23/04


Fast Sampling is NOT enoughFast Sampling is NOT enough

Raw SignalRaw Signal

After After Subtraction of Subtraction of Averaged Averaged ArtifactArtifact

After After Subtraction of Subtraction of Averaged Averaged ArtifactArtifact

0 0.1 0.2 0.3 0.4 0.5 0.6

0 5 10 15 20 25 0 5 10 15 20 25

100 µV

10 µV

Sampling rate: 10 kHz

Cohen 7/23/04


Residual ErrorsResidual Errors

cos(2ft) cos(2ft )

cos(2ft)cos( 1) sin(2ft)sin

…where:ƒ is the frequency of the artifact is the phase error, equal to 2πf0/fs,

- f0 is the EPI readout frequency and- fs is the sampling frequency.

At high sampling frequency (small At high sampling frequency (small ) the error, ) the error, , is , is linearly proportional to the sampling frequencylinearly proportional to the sampling frequency

tt

Cohen 7/23/04


––++

––++

Triggered Adaptive CorrectionTriggered Adaptive Correction

Trigger once per trTrigger once per tr

EEGEEG

++ ––

Fc=100 HzFc=100 Hz≥≥200 s/s200 s/s16 bits16 bits

AveragingAveraging

++

––

DigitizationDigitizationAnalog FilterAnalog Filter

DifferentialDifferentialAmplificationAmplification DC OffsetDC Offset

CorrectionCorrection

Corrected EEGCorrected EEG

Patent Applied For

Cohen 7/23/04


Sampled Gradient Drive CurrentSampled Gradient Drive Current

0 0.5 1 1.5 2 2.5 3Time (seconds)

Uncorrected

Average N=30

Corrected

Corrected X 5

Cohen 7/23/04


EEG Correction During Continuous ScanEEG Correction During Continuous Scan

tr = 2.5 ste = 30 ms20X20 cm15 slices

2000 3000 4000 5000 6000 7000Time (msec)

50 µV

Cohen

UCLA BrainMapping CenterColumbia University 5/11/04

MRI-Compatible EEG System DesignMRI-Compatible EEG System Design

Within the MR scanner

64 ChannelLocal Amplifier

Analog to Digital Convertor & Digital Signal Processor

EEG Display

USB over Optical Fiber

EEG Leads

(anonymized subject)(anonymized subject)

Cohen 7/23/04


State MeasurementsState Measurements

EEG may be the best available measure of state:

Sleep Attentiveness Arousal Responsiveness

Cohen 7/23/04


0 1 2 3 4 5 6 7 8 9 10Time (seconds)

EEG during Sleep (corrected)EEG during Sleep (corrected)

fp2-f8

fp8-t4

t4-t6

t6-o2

fp2-f4

c4-p4

p4-o2

f4-c4

fp2-f8

fp8-t4

t4-t6

t6-o2

fp2-f4

c4-p4

p4-o2

f4-c4

Cohen 7/23/04


UCLA: Simultaneous fMRI & Epileptiform EEGUCLA: Simultaneous fMRI & Epileptiform EEG

fp2-f8

f8-t4

t4-t6

t6-o2

fp1-t7

t7-t3

t3-t5

t5-o1

fp2-f4

f4-c4

c4-p4

p4-o2

fp1-f3

f3-c3

c3-p3

p3-O1

1 sec

Cohen 7/23/04


EEG Spectral ContentEEG Spectral Content


Cohen 7/23/04


Linear Systems ApproachLinear Systems Approach• Linear• Time-

Invariant• (LTI) System

x(t)x(t) y(t) = ƒy(t) = ƒ[[x(t)x(t)]]

ƒ(A + B) = ƒ(A) + ƒ(B)ƒ(A + B) = ƒ(A) + ƒ(B)

If h(t) is the response to an impulse:

y(t) h()x(t )d

x(t) * h(t)

Cohen 7/23/04


Response Latency vs. Stimulus DurationAverage of 10 recordings

Data courtesy R. Savoy / Mass. General Hospital

1 sec flash

17 ms flash

100 ms flashStimulusOnset

Signal

1840

1860

1880

1900

1920

15 20 25 30 35Time (seconds)

Cohen 7/23/04


•••••••••••••••

••

•

•••••

•

•

•

•••••

-10

0

10

20

30

40

50

0 2 4 6 8 10 12 14 16

Brain Impulse Response

Time (seconds)

MR Signal (a.u.)

Raw Data from R. Savoy

Light Flash

Cohen 7/23/04


0 25 50 75 100 125 150 175 200 225 2500

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

time (sec)time (minutes)

1 2 3 40

Ave

rage

Pow

er (

µV

2 )

Alpha MappingAlpha Mapping

spectral power in the alpha bandpredicted BOLD

response

Cohen 7/23/04


Spectral Time CourseSpectral Time Course

8

10

12

14

16

18

20

22

24

26

28

0 50 100 150 200 250

µV

Time (seconds)

Theta Alpha Gamma

5

10

15

20

25

15 20 25Gamma

Theta

Cohen 7/23/04


Simultaneous Imaging for Tomographic Electrophysiology

Cohen 7/23/04


EEG spectral fMRIEEG spectral fMRI

+ 0.6

± 0.3

– 0.6

Goldman, et al., NeuroReport, 2003

Patent Applied For

Cohen 7/23/04


Q&AQ&A

QQ: Do the spectral bands adequately : Do the spectral bands adequately describe the EEG?describe the EEG?

QQ: Is the forward convolution model : Is the forward convolution model valid?valid?

QQ: What is the explanatory power of : What is the explanatory power of the data?the data?

AA: : Ask a mathematician.Ask a mathematician.

Cohen 7/23/04


NNkk

PARAFACPARAFAC

EEGEEGSSftdftd

kk

aa

bb

cckk

kk

kk

F. Miwakeichi, et al., NeuroImage 22, 2004

frequency

time

channel

S(Nd X Nw X Nt) is a three

dimensional matrix, d are the electrode pairs, w is frequency and t is time.

ˆ S dwt adkbwkc tk dwt

k1

Nk

Each atom, Each atom, kk, is the trilinear combination of , is the trilinear combination of f, tf, t and and dd

Cohen 7/23/04


NNkk

PARAFACPARAFAC

EEGEEGSSftdftd

kk

aa

bb

cckk

kk

kk

AA

BB

CC

F. Miwakeichi, et al., NeuroImage 22, 2004

frequency

time

channel

ˆ S dwt adkbwkc tk dwt

k1

Nk

The “corcondia”The “corcondia” (=core consistency diagnostic)(=core consistency diagnostic) constraint determines N constraint determines Nkk..

A a k , B bk , & C ck Find:Find:

that explain that explain SS with minimal error. with minimal error.

Cohen 7/23/04


the Atomthe Atom



Cohen 7/23/04


PARAFAC applied to SITE dataPARAFAC applied to SITE data0.40.4

0.30.3

0.20.2

0.10.1

1010 2020 3030 4040 50500.10.1

00

Rel.Rel.EnergyEnergy

Frequency (Hz)Frequency (Hz)

thetatheta

alpha gamma

Cohen 7/23/04


Trilinear Partial Least SquaresTrilinear Partial Least Squares

kk

Maximize covarianceMaximize covariance

EEGEEGSS ftdftd

kk

aa

bb

cc kk

kk

kk

AA

BB

CCfrequency

time

channel

voxel

time

uu

vvkk

kkUU

VV

fMRIfMRIff stst

E. Martinez-Montes, et al., in preparation, 2003

Cohen 7/23/04


Trilinear Partial Least SquaresTrilinear Partial Least Squares

ˆ S dt adkbkctk edkk1

Nk

ˆ F st uskvtk stk1

Nk

Establish a structural model:

EEG

fMRI

then,Maximize the covariance of ck and uk


Cohen 7/23/04


Correlation of EEG and fMRI dataCorrelation of EEG and fMRI data

Alphar2 = 0.83p < 0.05

Thetar2 = 0.56p ≈ 0.07

Gammar2 = -0.03p = n.s.


Cohen 7/23/04


JackKnife pseudo t-imageJackKnife pseudo t-image

E. Martinez-Montes, et al., NeuroImage 22:1023-34, 2004

LL RR

Alpha

Cohen 7/23/04


Direct Dipole MappingDirect Dipole Mapping

LORETA solution to alpha atom - Source Spectrum Imaging


LL RR

Cohen 7/23/04


Apple PowerPCApple PowerPC

Subjectin Magnet

the UCLA Autocerebroscope

the UCLA Autocerebroscope

Statistical Detection of Signal Changes

Time Series of Images

MR ScannerMR Scanner

GE/ANMRGE/ANMR

Radio Signal

SITE E-physSITE E-phys

Stimulus Stimulus PresentationPresentation

Resonance TechnologyResonance Technology

Functional Functional ImageImage

Research supported under DA13054

Cohen 7/23/04


Changing the LandscapeChanging the Landscape

Non-Invasive Highly Invasive

3

2

1

0

-1

-2

-3

-4

brain

map

column

layer

neuron

dendrite

synapse

-3 -2 -1 0 1 2 3 4 5 6 7millisecond second minute hour day

Log10 seconds

Log10

millimeters

LesionsPET

2-Deoxyglucose

Microlesions

Light Microscopy

Light MicroscopyPatch ClampPatch Clamp

Optical ImagingOptical Imaging

Single UnitSingle Unit

MEG / ERPMEG / ERP

fMRI

Cohen 7/23/04




Levels of UnderstandingLevels of Understanding

fMRI-EEG

fMRI- EPhys

fMRI-Microdialysis

Ephys-Microdialysis

Cohen

UCLA BrainMapping CenterColumbia University 5/11/04

Richard DuBoisRichard DuBois

Robin GoldmanRobin Goldman

Susan BookheimerSusan Bookheimer

Ahmad HaririAhmad HaririAlison

Burggren-Clements

Alison Burggren-Clements

Meredith BraskieMeredith Braskie

Fred SabbFred Sabb

Zrinka BilusicZrinka Bilusic Alex KorbAlex Korb

Thanks to the GangThanks to the Gang

Jennifer BramenJennifer Bramen

Smoke Imagery Invert EndICAEnergy

QuickTime™ and aTIFF (LZW) decompressor


Massoud AkhtariMassoud Akhtari

Cohen 7/23/04


in my dreams…in my dreams…

RF Probe XRF Probe X

RF Probe YRF Probe Y

RF Probe ZRF Probe Z

Electrical ProbesElectrical Probes

Inner CannulaInner CannulaSemi-Semi-permeable permeable MembraneMembrane

Outer CannulaOuter Cannula

Cohen 7/23/04


thanksthanksNIDA DA13054 DA13627

DA15549 DA14093NEI EY12722Epilepsy Foundation of America UCLA Eugene Cota-Robles Fellowship

Amir AbrishamiDeane AikinsPete EngelEduardo Martinez-MontesFumikazu MiwakeichiMark SimonJohn SternNelson TrujilloPedro Valdes-Sosá

John Mazziotta

Robin GoldmanRobin Goldman

Cohen 7/23/04


EEG spectral ƒMRI (2)EEG spectral ƒMRI (2)

alpha (8-12 Hz)

beta (12-30 Hz)

Cohen 7/23/04


EEG Functional Block DiagramEEG Functional Block Diagram

Multi-plexer

OptoIsolator

RFShield

74 78 82Single Board Computer

TwistedPair

Leads

DifferentialAmp

IsolationBarrier

Latch

DAC

ADC

ShieldDriver

+

–

+

–

1 358

6368

70

64

66

72

Low Pass

76

OptoIsolator

TCP/IP TCP/IP

GradientTriggerGradient

Trigger

ECG

Cohen 7/23/04


Applications (triggered sampling)Applications (triggered sampling)

• EEG-fMRI.EEG-fMRI.

• Other physiological signals and fMRI Other physiological signals and fMRI (EKG, EMG, etc…)(EKG, EMG, etc…)

• Hum (60 Hz noise) in digital audio and Hum (60 Hz noise) in digital audio and telephony.telephony.

• Carrier Suppression in Radio, Radar Carrier Suppression in Radio, Radar and Television.and Television.

• ……any conditions in which signals are any conditions in which signals are digitized in conditions where digitized in conditions where repeatable noise contaminates the repeatable noise contaminates the signal.signal.

Simultaneous EEG and f MRI Acquisition

Documents

Transcript of Simultaneous EEG and f MRI Acquisition