Electroencephalography-based Motor Hotspot Detection

Ga-Young Choi

, Chang-Hee Han

, Hyunmi Lim

, Jeonghun Ku

, Won-Seok Kim

and Han-Jeong Hwang

Department of Medical IT Convergence Engineering, Kumoh National Institute of Technology,

Gumi 39177, Republic of Korea

Machine Learning Group, Berlin Institute of Technology (TU Berlin), 10623 Berlin, Germany

Department of Biomedical Engineering, School of Medicine, Keimyung University, Republic of Korea

Department of Rehabilitation Medicine, Seoul National University College of Medicine, Seoul National University

Bundang Hospital, Seongnam-si, Republic of Korea

Keywords: Neuronavigation, Hotspot, Electroencephalography, Transcranial Magnetic Stimulation.

Abstract: The motor-evoked potential (MEP) induced by transcranial magnetic stimulation (TMS) has been generally

used to identify a motor hotspot, and it has been used as a target location for transcranial electrical stimulation

(tES). However, the traditional MEP-based method needs a bulky TMS device, and it involves the empirical

judgement of an expert. In this study, we propose a machine-learning-based motor hotspot identification

method using electroencephalography (EEG) that is portably acquired in a tES device. EEG data were

measured from ten subjects while they performed a simple finger tapping task. Power spectral densities

(PSDs) were extracted from the EEG data as features, and they were used to train and test artificial neural

network (ANN). The 3D coordinate information of individual motor hotspots identified by TMS were also

used as the ground-truth motor hotspot locations in ANN, and they were compared with those estimated by

ANN. A minimum distance between the motor hotspots identified by TMS and EEG features was only 0.24

cm, demonstrating the feasibility of our proposed novel motor hotspot identification method based on EEG

features.

1 INTRODUCTION

Non-invasive brain stimulation (NIBS) is an

emerging technique that applies electrical current or

magnetic field to the scalp for the modulation of

cortical excitability (Paulus, 2000). NIBS is divided

into two types according to whether electrical current

or magnetic field is used. NIBS based on electrical

current is called transcranial electrical stimulation

(tES) that is divided into three types: i) transcranial

direct current stimulation (tDCS) (Nitsche et al.,

2000), transcranial alternating current stimulation

(tACS) (Herrmann et al., 2013), transcranial random

noise stimulation (tRNS) (Antal et al, 2016). NIBS

https://orcid.org/0000-0003-2209-5517

https://orcid.org/0000-0001-8668-3989

https://orcid.org/0000-0001-7074-7757

https://orcid.org/0000-0002-9610-0078

https://orcid.org/0000-0002-1199-5707

https://orcid.org/0000-0002-1183-1219

based on magnetic field is called transcranial

magnetic stimulation (TMS) (Wassermann et al,

2001).

TMS has been widely used to identify muscle

representations in the motor cortex as well as to

investigate corticomotor excitability. An optimal

TMS site is called as the motor hotspot, and it is

generally identified based on the TMS-induced motor

evoked potential (MEP).

The motor hotspot identified by TMS has been

used to validate the feasibility of tES on corticomotor

excitability (Cabral et al, 2015). Some studies have

shown that tES is effective for motor function

rehabilitation in patients with stroke, Parkinson’s

Choi, G., Han, C., Lim, H., Ku, J., Kim, W. and Hwang, H.

Electroencephalography-based Motor Hotspot Detection.

DOI: 10.5220/0008937201950198

In Proceedings of the 13th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2020) - Volume 4: BIOSIGNALS, pages 195-198

ISBN: 978-989-758-398-8; ISSN: 2184-4305

195

disease, amyotrophic lateral sclerosis (ALS), and so

on (Hummel et al., 2006). Most tES studies have used

the anodal electrode attached to the motor hotspot

identified by TMS, and the cathodal electrode

attached to the contralateral motor area or

contralateral supraorbital area (Ferreira et al., 2019).

Although TMS is an ideal tool to find the motor

hotspot, a cumbersome procedure involving the

empirical judgement of an expert is required to find

the motor hotspot. Also, it is impractical to use a TMS

device for finding the motor hotspot as a target area

for tES because a TMS device is relatively bulky and

heavy. A potential alternative to TMS identifying the

motor hotspot is to use electroencephalography

(EEG) measured while performing a motor task

related to a targeted motor hotspot because EEG

provides the representation information related to

motor functions even though its spatial resolution is

relatively low as compared to TMS. Therefore, in this

study, we propose an EEG-based machine-learning

approach to identify an individual motor hotspot that

is used as a target location for tES.

2 METHODS

2.1 Subjects

Ten right-handed subjects (five females and five

males; 25.3 ± 1.36 years) were recruited for this

study. They have no history of psychiatric diseases

that might affect research results. They received the

information about the details of experiment

procedure, and signed an informed consent for study

participation. Appropriate compensation for their

participation was provided after the experiment. This

study was approved by the Institutional Review

Board (IRB) of Kumoh National Institute of

Technology (No. 6250), and was conducted in

accordance with the principles of the declaration of

Helsinki.

2.2 Experiment Protocols

Subjects sat on a comfortable armchair. An individual

motor hotspot was first identified using TMS. The

motor hotspot was defined as the TMS coil location

that shows a MEP with an amplitude of at least 50 μV

more than 5 out of 10 consecutive stimuli when a

minimum stimulation intensity was applied. Because

a target region of interest was the right first dorsal

interosseous (FDI) muscle in this study, MEP was

measured from the FDI muscle using Ag-AgCl

disposable electrodes while single-pulse TMS was

applied to a corresponding brain area (REMED.,

Daejeon, Korea). We searched a motor hotspot on the

contralateral motor area (around C3 based on the

international 10-20 system); the coil was held at

approximately 45 degrees with the handle facing the

rear in order for TMS to be directed perpendicular to

the brain. Individual motor hotspot locations were

represented in the 3D coordinate (x, y, and z) based

on the vertex (Cz in the 10-20 international system)

using a polhemus patriot digitizer (Polhemus Inc.,

Colchester, Vermont, USA). The 3D locations of

individual motor hotspots were used as the ground

truth, and they were compared with those identified

by EEG to verify the feasibility of our proposed EEG-

based motor hotspot identification approach.

To measure motor-task-specific brain activity, 64

EEG electrodes were mounted on the scalp using the

international 10-20 system (Figure 1), and the

location of the EEG electrodes were also represented

in the 3D coordinate as that of the motor hotspot

identified by TMS-induced MEPs. The ground and

reference electrodes were attached on Fpz and FCz,

respectively. The EEG data were sampled at 1,000 Hz

using a multi-channel active electrode EEG

acquisition system (actiChamp, Brain Products

GmbH, Gilching, Germany) while each subject

performed a motor task that presses a button 30 times

using a right index finger whenever a red circle

appeared in the center of a monitor (Figure 2). The

subjects were given enough rest in the middle of the

experiment to avoid fatigue whenever they wanted. In

addition, they were instructed to remain relaxed

during the experiment without any movements to

prevent unwanted physiological artifacts.

Figure 1: Electrode positions used for recording EEG data.

BIOSIGNALS 2020 - 13th International Conference on Bio-inspired Systems and Signal Processing

196

Figure 2: Experimental paradigm.

2.3 Data Analysis

EEG data were analysed using the MATLAB

(MathWorks, Natick, MA, USA). The raw EEG data

were down-sampled into 200 Hz. We applied

common average reference (CAR) and bandpass

filtering between 0.5 and 50.5 Hz (zero-phase third-

order Butterworth filter) sequentially to the down-

sampled data. We also applied multiple artifact

rejection algorithm (MARA) based on independent

component analysis (ICA) to the filtered data in order

to remove physiological artifacts.

After the EEG preprocessing, we epoched the

EEG data between -0.5 and 0.5 sec based on an onset

time when a subject pressed a button for each trial.

Power spectral density (PSD) was estimated for each

trial and each channel using the fast Fourier transform

(FFT), and the PSDs of six frequency bands were

calculated (delta:1 – 4 Hz, theta: 4 – 8 Hz, alpha: 8 –

13 Hz, beta: 13 – 30 Hz, gamma: 30 – 50 Hz, full: 1

– 50 Hz). To identify the motor hotspot based EEG, a

multi-layer feedforward artificial neural network

(ANN) was trained and tested using EEG PSD

features (Figure 3). The input labels of the ANN were

the 3D coordinate information of the motor hotspots

identified by TMS, and the outputs were their

corresponding 3D coordinate information produced

by the ANN based on the EEG PSD features. A 10-

fold cross-validation was performed with early

stopping to prevent overfitting. The distance between

the 3D coordinates of the motor hotspots identified by

TMS and EEG was calculated using Euclidean

distance, which was defined as the error distance. The

mentioned procedure was performed for each

frequency band (delta, theta, alpha, beta, and gamma)

and the whole frequency band (full) to find an optimal

EEG frequency band to extract PSD features.

3 RESULT

Figure 4 presents a representative example from one

subject, showing the 3D coordinate locations of

motor hotspots identified by TMS (red) and EEG

PSD features (blue). The detected motor hotspots are

Figure 3: Schematic diagram of an ANN model used to find

motor hotspots based on EEG features.

located in the contralateral motor area of the right

index finger, and the motor hotspot locations

identified by TMS and EEG PSD features are close to

each other (0.65 cm).

Figure 4: 3D coordinate information of motor hotspots

identified TMS (red) and EEG PSD features (blue).

Figure 5 shows the mean error distances for each

frequency band. A minimum error distance of 0.24

cm was obtained when a full band was used to extract

PSD features (1.09 ± 0.38 cm for delta, 0.89 ± 0.43

cm for theta, 0.85 ± 0.46 cm for alpha, 0.43 ± 0.31 cm

for beta, and 0.35 ± 0.34 cm for gamma).

Figure 5: Mean error distances for each frequency band.

Electroencephalography-based Motor Hotspot Detection

197

Individual error distances for each frequency band

are presented in Table 1.

Table 1: Motor hotspot error distances of each subject for

each frequency band.

Subject Delta Theta Alpha Beta Gamma Full

0.82 0.92 0.31 0.13 0.03 0.11

1.31 0.91 0.81 0.28 0.18 0.16

0.73 0.87 0.35 0.28 0.17 0.25

0.38 0.52 0.47 0.15 0.08 0.22

1.22 0.71 0.48 0.33 0.16 0.09

1.69 1.02 1.39 0.64 0.69 0.17

1.36 0.51 1.30 1.10 0.74 0.30

0.97 0.57 1.10 0.23 0.26 0.21

1.03 0.89 0.72 0.39 0.15 0.21

S10

1.39 2.00 1.55 0.76 1.03 0.65

Mean

1.09 0.89 0.85 0.43 0.35 0.24

± Std.

0.38 0.43 0.46 0.31 0.34 0.16

4 CONCLUSIONS

In this study, we proposed an EEG-based novel motor

hotspot identification algorithm using machine

learning technique to provide a target location for tES

without using TMS. A minimum distance between

motor hotspots identified by TMS-induced MEP and

EEG features was 0.24 cm when using a full

frequency band information. As a tES electrode size

is generally bigger than 1 cm, it is expected that the

motor hotspot identified by EEG features could be

covered by a tES electrode with a small error

distance. However, additional tES experiments

should follow to verify the feasibility of our proposed

motor hotspot identification method based on EEG on

corticomotor excitability.

Instead of using a TMS device, an EEG device is

required to apply our proposed machine-learning-

based motor hotspot identification method. Note that

it is possible to integrate an EEG device to a tES

device with retaining its portability, and a

commercially available tES/EEG device already

exists (e.g., NeuroElectronics Starstim). Thus, we

expect that the EEG-based hotspot detection

algorithm will facilitate use of tES, in particular, for

home-based tES treatment. One limitation of our

algorithm is that TMS was used to find the 3D

coordinates of motor hotspots. Thus, we will develop

an advanced method that use the 3D coordinates of

motor hotspots identified by TMS to construct a

motor hotspot identification algorithm, after which it

uses only EEG features to find individual motor

hotpots for new subjects.

ACKNOWLEDGEMENTS

This work was supported by Ministry of Trade

Industry & Energy (MOTIE, Korea), Ministry of

Science & ICT (MSIT, Korea), and Ministry of

Health & Welfare (MOHW, Korea) under

Technology Development Program for AI-Bio-

Robot-Medicine Convergence (20001650).

REFERENCES

Paulus, W. 2011. Transcranial electrical stimulation (tES–

tDCS; tRNS, tACS) methods. Neuropsychological

rehabilitation, 21(5), 602-617.

Nitsche, M. A., & Paulus, W. 2000. Excitability changes

induced in the human motor cortex by weak transcranial

direct current stimulation. The Journal of physiology,

527(3), 633-639.

Herrmann, C. S., Rach, S., Neuling, T., & Strüber, D. 2013.

Transcranial alternating current stimulation: a review of

the underlying mechanisms and modulation of

cognitive processes. Frontiers in human neuroscience,

7, 279.

Antal, A., & Herrmann, C. S. 2016. Transcranial alternating

current and random noise stimulation: possible

mechanisms. Neural plasticity.

Wassermann, E. M., & Lisanby, S. H. (2001). Therapeutic

application of repetitive transcranial magnetic

stimulation: a review. Clinical Neurophysiology,

112(8), 1367-1377.

Cabral, M. E., Baltar, A., Borba, R., Galvão, S., Santos, L.,

Fregni, F., & Monte-Silva, K. (2015). Transcranial

direct current stimulation: before, during, or after motor

training?. Neuroreport, 26(11), 618-622.

Hummel, F. C., & Cohen, L. G. 2006. Non-invasive brain

stimulation: a new strategy to improve

neurorehabilitation after stroke?. The Lancet

Neurology, 5(8), 708-712.

Ferreira, I. S., Costa, B. T., Ramos, C. L., Lucena, P.,

Thibaut, A., & Fregni, F. 2019. Searching for the

optimal tDCS target for motor rehabilitation. Journal of

neuroengineering and rehabilitation, 16(1), 90.

BIOSIGNALS 2020 - 13th International Conference on Bio-inspired Systems and Signal Processing

198