Detecting gamma frequency neural activity using simultaneous ...

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 *equal leads in this work

Intro Heating Image Quality Motor study Conclusion

Introduction EEG gamma band (30-100 Hz) synchronization is linked to cognitive and sensory function e.g. • • • • • •

Visual Attention task

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)


Introduction

• 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


Introduction

FFT of EEG data during fMRI of an agar phantom Uncorrected MR Gradient artefact

Corrected MR Gradient artefact

Cz 1200

Uncorrected MR gradient

Cz 50

Without MR Scanning

Corrected MR gradient 40

1000 800

Amplitude

Amplitude

Without MR Scanning

600

30 20

400 10

200 0

50

100 150 Frequency (Hz)

200

0

250

50

T7


200

250

T7 50

4000

Uncorrected MR gradient

Without MR Scanning

Corrected MR gradient

Without MR Scanning

3000

Amplitude

Amplitude

40

2000 1000 0

30 20 10

50


200

250

0

50


200

250


Introduction

• 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


Introduction

Purpose of work • To examine the feasibility of recording EEG simultaneously with multiband fMRI in humans • To demonstrate its potential for investigating relationships between gamma frequency EEG activity and BOLD responses


Methods

Recording 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: Safety testing • 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


Results: Heating effects Cable bundle

ECG

Cz

TP7

FCz

TP8

Scanner Bore

0.8

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

0.6

ΔTemp (°C)

0.4

0.2

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

0

-0.2 -0.4 0

1

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17 18 19 20 Time (min)

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


Methods: Image quality

• 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: 𝑡𝑆𝑁𝑅𝑣𝑜𝑥𝑒𝑙 = 𝑚𝑒𝑎𝑛 𝑠𝑖𝑔𝑛𝑎𝑙 𝑜𝑣𝑒𝑟 𝑡𝑖𝑚𝑒𝑣𝑜𝑥𝑒𝑙 𝑠𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝑑𝑒𝑣𝑎𝑡𝑖𝑜𝑛 𝑜𝑣𝑒𝑟 𝑡𝑖𝑚𝑒𝑣𝑜𝑥𝑒𝑙 tSNR map

• tSNR was averaged over grey matter (GM), defined from segmenting subject’s T1 anatomical

GM mask


Results: Image quality Multiband Factor

Slice acquisition spacing

tSNR

1

Equidistant

74 ± 40

2

Equidistant

72 ± 39

2

Sparse

67 ± 37

3

Equidistant

68 ± 37

3

Sparse

74 ± 38

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.

Little variability in tSNR between sequences So MB=3, sparse acquisition chosen for further experiments


Methods: EEG-fMRI motor study • 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:

Sparse MB=3 fMRI sequence

4 abduction movements

Resting baseline Time

MR acquisition (0.75s)

Quiet period (2.25s)

• 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)

EMG recorded from right FDI muscle


Methods: EEG-fMRI motor study 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)


Methods: EEG-fMRI motor study 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 secondlevel fixed and third-level mixed effects analysis


Results: EEG-fMRI motor study 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 timefrequency 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.


Results: EEG-fMRI motor study Stimulus period

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

Residual MR gradient artefacts













Left

Right 2

4.5

2

3

Main effect BOLD response to abductions

Z-stats

Positive Gamma-BOLD correlation

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