Reference layer adaptive filtering (RLAF)in simultaneous EEG-fMRI

David Steyrl1 and Gernot R. Muller-Putz1

1Institute of Neural Engineering, BCI Lab, Graz University of Technology, Austria

Introduction

Simultaneous electroencephalography (EEG) and functional magnetic

resonance imaging (fMRI) combines advantages of both methods: high

temporal resolution of EEG and high spatial resolution of fMRI. EEG

recordings are, however, afflicted by severe artifacts caused by fMRI

scanners. Average artifact subtraction (AAS) is a common method to

reduce those artifacts [1]. Recently, we introduced an add-on method

that uses a reusable reference layer EEG cap prototype in combination

with adaptive filtering, to improve EEG data quality substantially [2, 3].

The methods applies adaptive filtering with reference layer artefact data

to optimize artefact subtraction from EEG and is named reference layer

adaptive filtering (RLAF).

Methods

The reference layer cap was developed by GUGER TECHNOLOGIES

OG, Austria, see Figure 1 panel A, B and C. It consists of 30 double-

layer electrode pairs and 2 additional ECG electrodes. 29 electrode

pairs are capturing EEG and one electrode pair serves as common

ground/reference electrode. The reference layer itself is a system of

saline-water filled tubes and galvanically isolated from the scalp except

at the common ground/reference electrode.

Figure 1: A - Rendering of the reference layer cap prototype. B - The referencelayer cap in use. C - Schematics of the double layer electrode pairs.

The experiment was designed to evaluate visual evoked potentials

(VEPs) and alpha-rhythm changes between opened and closed eyes.

The two participants underwent inverse checker-board stimuli to trig-

ger VEPs (1200 inversions). Subsequently, 10min of resting EEG during

closed eyes was recorded.

The signal processing is depicted in Figure 2 and included a 1Hz high-

pass, AAS of the gradient artefact (GA, 50 epochs for averaging), down-

sampling to 250Hz, AAS of the pulse artefact (PA, 50 epochs), a 50Hz

notch filter and finally the adaptive filtering.

HP1Hz

1

2

N

...

reference

1

2

N

...

scalp

1

2

N

...

reference

1

2

N

...

scalp

Raw

AASGA

1

2

N

...

reference

1

2

N

...

scalp

AASPA

1

2

N

...

reference

1

2

N

...

scalp

ECG

Notch50Hz

1

2

N

...

reference

1

2

N

...

scalp

AAS(GA+PA)

1

2

N

...

RLAF

RLAF

adaptive

�ltering

Figure 2: Signal processing scheme.

We optimized the step width of the adaptive filtering to guarantee sta-

bility and fast convergence of the method (7 · 10−6 to 1.5 · 10−4).

Results

We used a similar evaluation procedure as used by Jorge et al. [4]. All

results are depicted in Figure 3.

−100

0

100

−30

0

30

−30

0

30

Time (s)

1 100 200 300 400 500 600 700 800 900 1000 1100 1200 13000

0.5

1

1.5

2

2.5

Time (s)

Sca

ling

Fa

cto

rs (

AU

)

0.7

1.4

2.0

B

Channel O2

Participant 2

Channel FC1

Participant 2

Channel FC1

Participant 1

1.7

1.2

0.7

Am

plitu

de (

µV

)

EEG at O2 after AAS (GA)

EEG at O2 after AAS (GA+PA)

Ref at O2 after AAS (GA+PA)

PAPAPAPAPAPAPA PA

A3

A3

A5

A5

A2

A2

A1

A1

alphaalphaA4

A4

−30

0

30

EEG at O2 after RLAF

A5

A3A2

A1

alphaalpha A4

A

760 761 762 763 764 765 766 768 769

40

35

30

25

20

15

10

5

0µV²

alpha power after

AAS (GA+PA)alpha power after

RLAF

S2

alpha power after

AAS (GA+PA)

20181614121086420µV²

alpha power after

RLAF

S1C

Nu

mb

er

VE

P

O2 Raw

200

400

600

800

1000

1200

O2 AAS(GA+PA)

.1 .2 .3

−4

−2

0

2

4

6

Time (s)

Am

plit

ud

e (

µV

)

MSD: 1918.5

O2

O1

.1 .2 .3

Time (s)

MSD: 200.2

O2

O1

O2 AAS+RLAF

.1 .2 .3

Time (s)

MSD: 28.6

O2

O1

Num

ber

VE

P

O1 Raw

200

400

600

800

1000

1200

O1 AAS(GA+PA)

.1 .2 .3

−4

−2

0

2

4

6

Time (s)

Am

plit

ude (

µV

)

MSD: 1136.5

O1

O2

.1 .2 .3

Time (s)

MSD: 152.8

O1

O2

O1 AAS+RLAF

.1 .2 .3

Time (s)

MSD: 38.4

O1

O2

D

Figure 3: A - EEG time course comparison after artefact reduction in partici-pant 1. B - Adaptive filter coefficients’ course over time. C - Alpha power spatialdistribution. D - Selected single and average VEPs of participants 1 and 2.

Discussion

The adaptive filter scaling values are changing over time, which indicates

the necessity of adaptive filtering. Our results demonstrate that RLAF

reduces artefacts while it preserves EEG phenomenons as evoked poten-

tials and the occipital alpha rhythm. Hence, RLAF is able to improve

the quality of EEG of simultaneous EEG-fMRI experiments.

References

1. P.J. Allen, O. Josephs, and R. Turner, A method for removing imaging artifact from continuous EEG recorded during functional MRI. Neuroimage, 12(2), pp.230-239,2000. doi:10.1006/nimg.2000.0599

2. D. Steyrl, F. Patz, G. Krausz, G. Edlinger, and G. R. Muller-Putz, Reduction of EEG Artifacts in Simultaneous EEG-fMRI: Reference Layer Adaptive Filtering(RLAF), in Proceedings of the 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBC15, Milano, Italy,August 25-29, 2015, pp.3803-3806. doi.org/10.1109/EMBC.2015.7319222

3. D. Steyrl, G. Krausz, K. Koschutnig, G. Edlinger, and G. R. Muller-Putz, Reference layer adaptive filtering (RLAF) for EEG artifact reduction in simultaneousEEG-fMRI, Journal of Neural Engineering, vol.14(2), pp.1-20, 2017. doi.org/10.1088/1741-2552/14/2/026003

4. J. Jorge, F. Grouiller, R. Gruetter, W. van der Zwaag, P. Figueiredo, Towards high-quality simultaneous EEG-fMRI at 7 T: Detection and reduction of EEG artifactsdue to head motion, Neuroimage, 120, pp.143-153, 2015. doi:10.1016/j.neuroimage.2015.07.020

Acknowledgments

This work was partly supported by

the ERC Consolidator Grant 681231

‘Feel Your Reach’. This paper only

reflects the authors’ views and fund-

ing agencies are not liable for any

use that may be made of the infor-

mation contained herein.