Detecting gamma frequency neural activity using simultaneous multiband EEG fMRI

(1)

Detecting gamma frequency neural activity

using simultaneous multiband EEG-fMRI

Makoto Ujia, Ross Wilsona, Susan T. Francisb, Stephen D. Mayhewa*, and Karen J. Mullingera,b*

a Centre for Human Brain Health (CHBH), University of Birmingham, Birmingham, UK b SPMIC, School of Physics and Astronomy, University of Nottingham, Nottingham, UK

(2)

EEG gamma band (30-100 Hz)

synchronization is linked to cognitive

and sensory function e.g.

• Visual

• Somatosensory • Auditory

• Memory processing • Motor control

• Attention

However, incomplete understanding of

relationship between gamma and

BOLD responses with the majority of

work carried out in the visual domain

(Logothetis, Nature, 2008; Koch et al., J Neurosci, 2009

Magri, J Neurosci, 2012) Scheeringa et al., Neuron (2011)

(3)

• Simultaneous EEG-fMRI: novel method to investigate EEG gamma – fMRI BOLD correlates

• However, residual gradient artefacts typically obscure gamma frequency EEG activity when acquired with fMRI

(4)

50 100 150 200 250 0 200 400 600 800 1000 1200 Cz Frequency (Hz) A m p lit u d e

Uncorrected MR gradient Without MR Scanning

50 100 150 200 250

0 1000 2000 3000 4000 T7 Frequency (Hz) A m p lit u d e

Uncorrected MR gradient Without MR Scanning

Corrected MR Gradient artefact Uncorrected MR Gradient artefact

Introduction

FFT of EEG data during fMRI of an agar phantom

50 100 150 200 250

0 10 20 30 40 50 Cz Frequency (Hz) A m p lit u d e

Corrected MR gradient Without MR Scanning

50 100 150 200 250

0 10 20 30 40 50 T7 Frequency (Hz) A m p lit u d e

(5)

• Simultaneous EEG-fMRI: novel method to investigate EEG gamma – fMRI BOLD correlates

• However, residual gradient artefacts typically obscure gamma frequency EEG activity when acquired with fMRI

• Multiband (MB) fMRI can provide

 Shorter TR

 Increased brain coverage for a given TR

 Shorter acquisition time in sparse fMRI which extends “quiet periods” for

stimulus presentation

MB-fMRI potentially useful to overcome gradient artefacts in EEG

• Before MB EEG-fMRI can be acquired needs to be assessment of • Safety (due to increased peak RF power required in MB excitation) • EPI image quality due to the combined effect of EEG and MB

(6)

Introduction

Purpose of work

• To examine the feasibility of recording EEG simultaneously with

multiband fMRI in humans

• To demonstrate its potential for investigating relationships

(7)

EEG

• MRplus EEG amplifiers, 5kHz

• 64-channel EEG cap (Brain Products)

fMRI

• 3T Philips Achieva MRI scanner

• Multiband acquisition (Gyrotools, Zurich)

• Physiological recording of cardiac and respiratory traces

• MR-EEG clocks were synchronised

Methods

(8)

• The EEG cap was placed on and connected to a conductive agar

head-shaped phantom

• Fibre-optic thermometers (Luxtron) monitored the temperature

change during 20-minute scans at:

 Electrodes (ECG, Cz, TP7, FCz & TP8),

 cable bundle,

 the scanner bore (as a control location)

• Tested the upper bound of SAR for fMRI sequences:

1) GE-EPI (TR/TE=1000/40ms, slices=48, B1 RMS=1.09μT, SAR/head=22%)

2) PCASL-GE-EPI (TR/TE=3500/9.8ms, slices=32, B1 RMS=1.58μT, SAR/head=46%)

With MB factor = 4 and SPIR fat suppression for both sequences

(9)

Temperature changes at EEG electrodes, cable bundle and a control location on the scanner bore during a 20-minute GE-EPI scan. Temperature changes were calculated relative to a 5-20-minute baseline recording made

before the scan at each location.

• The greatest temperature change seen for the GE-EPI sequence was 0.6°C in the ECG electrode

• The PCASL-GE-EPI resulted in the greatest heating effect (ECG ~0.9°C)

-0.4 -0.2 0 0.2 0.4 0.6 0.8

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Δ T e m p ( °C) Time (min)

Cable bundle ECG Cz TP7 FCz TP8 Scanner Bore

(10)

• 3 healthy young subjects

• On each subject acquired 5 different GE-EPI sequences, varying:

 MB factor: 1, 2 or 3

 Slice acquisition spacing: equidistant or sparse

 Kept constant: TR/TE=3060/40ms; SENSE=2; Slices=36; FA=79°; 41 volumes

• Image quality was assessed by computing temporal signal-to-noise

ratio (tSNR) maps where:

Methods:

Image quality

𝑡𝑆𝑁𝑅_{𝑣𝑜𝑥𝑒𝑙} = 𝑚𝑒𝑎𝑛 𝑠𝑖𝑔𝑛𝑎𝑙 𝑜𝑣𝑒𝑟 𝑡𝑖𝑚𝑒𝑣𝑜𝑥𝑒𝑙

𝑠𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝑑𝑒𝑣𝑎𝑡𝑖𝑜𝑛 𝑜𝑣𝑒𝑟 𝑡𝑖𝑚𝑒_{𝑣𝑜𝑥𝑒𝑙}

tSNR map GM mask

• tSNR was averaged over grey matter

(GM), defined from segmenting

(11)

Mean temporal SNR (tSNR) (±standard deviation) calculated over grey matter across 3 participants during five MR sequences: MB: 1-3; acquisition type = equidistant or sparse.

All other parameters were constant: TR/TE=3060/40ms, SENSE=2, slices=36, FA=79°, Volumes=41.

Results:

Image quality

Little variability in tSNR between sequences

So MB=3, sparse acquisition chosen for further experiments

Multiband Factor

Slice acquisition spacing

tSNR

1 Equidistant

74 ± 40

2 Equidistant

72 ± 39

2 Sparse

67 ± 37

3 Equidistant

68 ± 37

(12)

• 10 healthy, young adult (8 males, 2 females) – all right-handed

• 4 right-hand index finger abduction movements per trial with 16s interstimulus interval

• Abductions performed in response to auditory cues, 1kHz beep • One trial:

• 4 runs of 30 trials per subject, 120 trials in total per subject • Sparse GE-EPI scheme:

MB=3, TR/TE=3000/40ms, 33slices, Voxels=3mm3_{, 192 volumes, SAR/head < 7%}

• EEG electrode positions digitised (Polhemus Fastrak)

Methods:

EEG-fMRI motor study

Time MR acquisition (0.75s) Quiet period (2.25s)

4 abduction movements

Resting baseline

Sparse MB=3 fMRI sequence

EMG recorded from right FDI

(13)

EEG data analysis

• Corrected gradient and pulse artefacts: average artefact subtraction (BrainVision Analyzer2) • Downsampled (600Hz) & epoched from -16 to 2s relative to auditory cue onset

• Removed (visual inspection) channels/trials contaminated with large artefacts/baseline EMG movements

• ICA to remove eye-blinks/movements (EEGLAB) • Average referenced to non-noisy channels

• A LCMV beamformer with individual BEM headmodels (Fieldtrip) to localise changes in gamma frequency (55–80Hz) power to abduction movements (Darvas et al, NeuroImage,2004)

[active: 0 to 1.5s; passive: -9.0 to -7.5s time windows]

• A broadband (1-120Hz) timecourse of neural activity was extracted from the peak T-statistic location (virtual electrode, VE) in the contralateral primary motor cortex (M1)

• Time-frequency spectrograms were calculated using a multitaper wavelet approach

(Oostenveld et al., Comput. Intell. Neurosci.2011; Scheeringa et al., Neuron,2011)

• Mean gamma power per trial (0-1.5s after auditory cue onset) formed a regressor for fMRI analysis (Mayhew et al., NeuroImage,2010; Mullinger et al., NeuroImage, 2014)

(14)

fMRI data analysis (FSL)

• fMRI data were motion corrected

• Spatially smoothed (5mm)

• Normalised to the MNI template

• First-level GLM analysis employed 2 regressors convolved

with the HRF:

1) Boxcar abduction movement

2) Parametric modulation of single-trial gamma neuronal activity

• Data were grouped over runs and subjects using

second-level fixed and third-second-level mixed effects analysis

(15)

Virtual electrode location of peak gamma frequency (55-80Hz) response

• Increase in gamma power observed during finger abductions over

contralateral M1

• Peak gamma response in contralateral M1 used as the virtual electrode

location from which to extract

broadband timecourse for further time-frequency analysis

Group average (N=10) T-stat beamformer maps of increase in gamma-power during abduction movement compared to baseline. Cross hair denotes the mean location of the peak location found for

each subject. Plotted on the MNI brain.

(16)

Group mean time-frequency spectrograms demonstrating changes in the EEG signal power in contralateral M1 a) 18s whole-trial duration: shows residual gradient artefacts during EPI acquisition

b) Gamma power increases and beta power decreases are seen during the active compared with passive window

Results:

EEG-fMRI motor study

(17)

Results:

EEG-fMRI motor study

Residual MR

(18)

(19)

(20)

(21)

Group average (N=10) fMRI mixed effects results.

Main effect BOLD response to finger abductions (red-yellow) and areas of positive gamma-BOLD correlation (green). Main effects and single-trial correlations are cluster corrected with p<0.05, masked to motor cortex.

Results:

EEG-fMRI motor study

Z-stats

2 4.5

2 3

Main effect BOLD response to abductions

Positive Gamma-BOLD correlation

(22)

Group average (N=10) fMRI mixed effects results.

Main effect BOLD response to finger abductions (red-yellow) and areas of positive gamma-BOLD correlation (green). Main effects and single-trial correlations are cluster corrected with p<0.05, masked to motor cortex.

Results:

EEG-fMRI motor study

Z-stats

2 4.5

2 3

Main effect BOLD response to abductions

Positive Gamma-BOLD correlation

(23)

Safety

• MB EEG-fMRI acquisition was safe for GE-EPI with MB factor = 4

• However, for future work testing of the specific sequence/implementation of MB/set-up must still be done.

• Our sequences were within safe limits (<1°C) for heating with our set up, but were relatively close (maximum 0.9°C) (Carmichael et al., J of Magnetic Resonance Imaging,2008)

Image quality

• The addition of the EEG to MB-fMRI did not degrade MR image quality

Neurovascular coupling

• Our results support a tight, positive coupling between BOLD fMRI and gamma EEG responses in motor cortex, similar to previous reports in visual cortex (Logothetis et al., 2001)

• Our work shows the potential of combining EEG-MB fMRI for advanced study of human brain function

(24)

We thank the Birmingham-Nottingham Strategic

Collaboration Fund for funding this research

(25)

GMmasked tSNR map MB3 Equidistant

GMmasked tSNR map MB3 Sparse

GM mask

(26)

Virtual electrode location of

beta frequency (15-30Hz) response

Group average (N=10) T-stat beamformer maps of decreases in beta-power during abduction movement compared to baseline. Plotted on the MNI brain.

• Decrease in beta power observed during finger abductions over M1 bilaterally

• Beta response in contralateral M1 used as the virtual electrode location from which to extract broadband

timecourse for further time-frequency analysis