Investigation of Artifact Contamination Impact on EEG Oscillations

Towards Enhanced Motor Function Characterization

Mojisola Grace Asogbon

1,2 a

, Oluwarotimi Williams Samuel

1,2 b

, Farid Meziane

1,2,

Guanglin Li

and Yongcheng Li

School of Computing, University of Derby, Derby, DE22 3AW, U.K.

Data Science Research Centre, University of Derby, Derby DE22 3AW, U.K.

CAS Key Laboratory of Human-Machine Intelligence-Synergy Systems, Shenzhen Institutes of Advanced Technology

(SIAT), Chinese Academy of Sciences (CAS), Shenzhen, Guangdong, China

Keywords: Electroencephalogram (EEG), Signal Processing, Artifact Removal Methods, Motor Recovery.

Abstract: The significant advancements in electroencephalography (EEG)-driven technology have led to its widespread

use in assessing stroke-related conditions. Over the years, various studies have explored the potential of EEG

oscillatory patterns in neurological research, with several of them giving limited attention to the signal

processing techniques employed, precluding a proper understanding of EEG oscillatory patterns under various

conditions. To resolve this issue, we systematically investigated how artifacts impact EEG oscillatory rhythms

associated with upper limb movement-related tasks. Thus, the EEG signals of motor tasks were acquired non-

invasively from healthy subjects and processed using automated artifact-attenuation methods. Subsequently,

the Mu and Beta bands in the brain's motor cortex region were extracted through time-frequency analysis and

analyzed using relevant metrics. Experimental results revealed that artifacts in EEG would substantially

influence the brain activation strength and response during motor tasks. Notably, signals preprocessed with

Reduction of Electroencephalographic Artifacts based on Multi Wiener Filter and Enhanced Wavelet

Independent Component Analysis (RELAX_MWF_wICA) showed better brain responses and high task

classification performance compared to other methods and the raw signal across motor tasks. This study's

findings revealed that the choice of signal processing technique is crucial, as it would influence its analysis

and interpretation, thus highlighting the need for careful consideration and usage.

1 INTRODUCTION

The study of neural oscillations, driven by the

coordinated activity of numerous neurons and

assessed through techniques such as functional

magnetic resonance imaging (fMRI),

electroencephalography (EEG), and magneto-

encephalography (MEG), among others, has been a

prominent and extensively explored area in

neurological research (Ward, 2015; Gui et al., 2010;

Jee, 2021).

Notable advances in EEG technology have led to

its wide usage in assessing stroke-related brain

function. EEG is a non-invasive and safe method

with an excellent temporal resolution that offers

valuable insights into brain activity through direct

https://orcid.org/ 0009-0006-1503-9356

https://orcid.org/ 0000-0003-1945-1402

https://orcid.org/ 0000-0001-9811-6914

measurement of electrical potentials from the

underlying neural tissue (Wu et al., 2016; Lin et al.,

2017; Asogbon et al., 2021; Anapama et al., 2012).

EEG signals represent recurring patterns resulting

from the coordinated activity of neurons firing in

synchrony, and they can be observed across a range

of frequencies (including delta, theta, alpha, beta, and

gamma bands). Brain activities during upper limb

movements in these bands are affected by stroke

(Maura et al., 2023; Bartur et al., 2019). Therefore,

they are considered promising predictors that can

offer valuable insights into stroke patients’ status,

helping clinicians identify distinct biological

subgroups and determine which treatment approach

might be more appropriate and effective (Cassidy et

al., 2019).

Asogbon, M., Samuel, O., Meziane, F., Li, G. and Li, Y.

Investigation of Artifact Contamination Impact on EEG Oscillations Towards Enhanced Motor Function Characterization.

DOI: 10.5220/0012373400003657

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 17th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2024) - Volume 1, pages 755-762

ISBN: 978-989-758-688-0; ISSN: 2184-4305

755

For instance, in contemporary stroke therapeutic

treatment, EEG oscillations are utilized as predictive

indicators, incorporated with clinical intervention

techniques. This integration further enhances

diagnosis, treatment, and recovery in stroke patients

with motor impairments (Keser et al., 2022). In 2019,

Cassidy et al. investigated EEG oscillations as a

potential predictor of injury and motor function

recovery in stroke survivors. By experimenting with

EEG recordings from both healthy controls and stroke

patients, the study examined the connection between

EEG oscillations and injury and motor condition,

utilizing delta and high-beta frequency bands. The

study's outcome revealed that delta-frequency

oscillations reflect both injury and motor function

recovery after a stroke.

In addition, Thibaut et al. (2017) found, in their

work, that brain activity in both lesioned and

unlesioned hemispheres of stroke patients, as

measured by EEG, provides new insights into the

relationship between high-frequency rhythms and

motor impairment. Their findings highlight the role

of an excess of beta activity in the affected central

cortical region, contributing to poor motor function

during stroke recovery.

A research study conducted by López-Larraz et al.

(2018) emphasized the significance of employing

suitable techniques to eliminate artifacts in EEG

recordings of stroke patients. The study aimed to

uncover the true neural activity by eliminating

unwanted interference. The findings revealed that

during motor tasks, EEG-cortical activation is

heightened, and the presence of artifacts can

introduce an overly optimistic bias in the performance

evaluation of brain-machine interfaces (BMIs).

Unarguably, several works have conducted

exploratory investigations using EEG oscillatory

rhythms to predict motor function recovery in stroke

patients. Considering that the EEG signal is

susceptible to contamination from various artifacts,

these disturbances can significantly influence the

resulting signal, potentially leading to

misinterpretation if a robust cleaning method is not

implemented at the signal processing stage.

Unfortunately, relatively little attention has been

directed towards the methodologies employed for

processing EEG oscillations in relation to upper limb

motor tasks, representing a fundamental drawback in

the field.

In addressing this concern, we systematically

investigated the influence of artifacts on cortical

activation and the recognition of motor tasks using

EEG-based neural oscillations, with a specific focus

on the Mu (μ) and Beta (β) bands. The study involved

the analysis of non-invasively collected EEG signals

from healthy individuals participating in four distinct

movement execution (ME) tasks.

These signals were individually processed using

five automated data-driven methods capable of

removing either single or multiple artifacts. These

methods were selected from widely used EEG artifact

attenuation techniques based on performance criteria

evident in existing works.

The ICA decomposition method is applied to the

processed signal, and an automated independent

component (IC) classification method was used to

detect or flag artefactual ICs based on specific

thresholding parameters. After that, the signal

segment that is time-locked to a specific event was

epoched and analysed using time-frequency analysis.

2 MATERIALS AND METHODS

2.1 Participants Information

In this study, 20 healthy subjects volunteered to

participate in the experiment. Specifically, right hand

dominated individuals including male and female

aged between 20 and 35 years were recruited. Prior to

the experiment, all volunteers were briefed on the

study objective and the experimental procedure.

Subsequently, they all agreed and gave written

consent for the publication of their data. The

Institutional Review Board of Shenzhen Institutes of

Advanced Technology, Chinese Academy of

Sciences, approved the recruitment and experimental

process.

2.2 Equipment Setup and Data

Collection

The experiment was conducted at the Shenzhen

Institute of Advanced Technology, Chinese Academy

of Sciences. The EEG signals were acquired from the

subjects using a 64-channel gel-based AgCl electrode

cap combined with a Neuroscan acquisition system.

The EEG cap was positioned on each of the

volunteer’s head following international 10–20

electrode placement configuration. The ground

electrode was positioned at AFz and referenced to

CPz during signal recording. All electrode channels

were sampled at 1000Hz and based on the volunteer’s

tolerance level, the impedance varies between 5-8kΩ.

Before commencing the experiment, the subjects

were trained on the experimental procedure and

instructed on how to perform the ME tasks. The ME

tasks, including key grip (KGME), power grip

BIOSIGNALS 2024 - 17th International Conference on Bio-inspired Systems and Signal Processing

756

(PGME), wrist extension (WEME), and wrist flexion

(WFME), shown in Figure 1, were performed by each

subject. The subjects were instructed to sit on a

comfortable high-back chair and watch a visual

display unit (VDU) placed 1m in front of them. To

ensure the tasks were performed correctly, pre-

recorded video containing an active (say wrist

extension) and non-active task (rest) was developed.

The VDU was used to display the video of each task

to guide them throughout the experiment. Each active

task (ME) was performed for a duration of 5s,

followed by 5s of non-active task (rest) to mitigate

fatigue. In total, participants completed two

consecutive sessions, each comprising 10 active tasks

and 10 non-active tasks.

Figure 1: A representation of a participant during the motor

execution tasks which includes key grip , power grip ,wrist

extension and wrist flexion .

2.3 Data Processing

The signal recorded for each participant underwent

offline processing and analysis utilizing the EEGLAB

(Delorme and Makeig 2004) and MATLAB (The

MathWorks Inc. 2019) toolkits. Towards

understanding the relevance of artifacts on EEG

oscillatory patterns five popularly used automatic

data-driven EEG artifact attenuation methods were

applied to the recorded signals. The procedure for the

signal processing is described as follows:

1. Utilizing EEGLAB, each of the ME tasks of the

recorded signals trials/session were merged.

Subsequently, the signals were filtered using a

passband edge frequency of 1Hz and 30Hz. The

1Hz signifies the upper limit of the lower

frequency range, and 30Hz represents the lower

limit of the higher frequency range that can pass

through the filter.

2. Afterward, the following automated EEG

artifacts elimination methods were individually

applied to the filtered signal:

(a) Independent Component Analysis (ICA)

based Extended Information-maximization

(INFOMAX) (Jutten & Herault, 1991;

Comon, 1994).

(b) Artifact Subspace Reconstruction (ASR)

(Bloniasz, 2022; Chang et al., 2019, blum et

al., 2019).

(CCACCA) (Gómez-Herrero et al., 2006;

De Clercq et al., 2006).

(d) Reduction of Electroencephalographic

Artifacts based on Multi Wiener Filter and

Enhanced Wavelet ICA

(RELAX_MWF_wICA) (Bailey et al.,

2022; Somers et al., 2018; Castellanos et al.,

2006).

(e) Reduction of Electroencephalographic

Artifacts based on Multi Wiener Filter

(RELAX_MWF) (Bailey et al., 2022;

Somers et al., 2018).

Importantly, the single and multiple automated

artifact reduction methods were chosen from

commonly used EEG artifact removal techniques

based on their superior performance. The detail

description of each method can found in the provided

references above.

3. Next, ICs of the individually processed signals

were computed using the ICA decomposition

method.

4. The IC_Label, an accurate and computationally

efficient classifier compared to other commonly

used automated IC component classification

method (Pion-Tonachini et al., 2019), was

applied to detect or flag artefactual ICs based on

thresholding parameters. Thereafter, the flagged

artefactual ICs were subtracted from the

processed signals.

Gri

Power Gri

Wrist Extension Wrist Flexion

Investigation of Artifact Contamination Impact on EEG Oscillations Towards Enhanced Motor Function Characterization

757

5. The resulting continuous EEG signals are

epoched by extracting data epochs that are time-

locked to a specified ME task.

6. In examining the task-related EEG dynamics of

the signals, each task was epoched, choosing a

window from -1s to 5s.

7. The epoched datasets are saved for time-

frequency analysis and other analyses in

MATLAB using a custom-built script.

It is worth stating that the signal processing steps

presented above were performed on each subject’s

EEG recording.

2.4 Feature Extraction and Task

Decoding

In gaining insights into how the automated methods

for reducing artifacts affect brain activation strength,

the z- The 𝜇 and 𝛽 oscillatory pattern in the brain's

motor cortex region during ME tasks were

considered. As demonstrated in existing works, these

bands were selected based on their modulation

characteristics during movement. In addition,

alterations in these bands during ME tasks have been

found to correspond with motor impairment in stroke

patients. (Bartur et al., 2019; Leonardi et al., 2022).

Apply a time-frequency analysis-based approach,

the 𝜇 and 𝛽 bands in the range of 10-14Hz and 16-

26Hz were extracted from the cleaned/processed

signal from the 18 electrodes at the motor cortex

region (excluding the midline electrodes).

The short-time Fourier transform analysis was

performed on each processed signal within the

designated frequency bands (μ and β). Following the

time-frequency decomposition, the data were z-

scored (eqn. 1). A statistical comparison of the z-

scored power in both frequency bands during rest and

movement task was conducted across all EEG

channels using the Wilcoxon rank-sum test, and the

results were subsequently topographically mapped. In

cases where no significant difference was observed

between rest and movement for a channel, the value

was set to 0.

𝑋

𝑋−𝜇

𝜎

(1)

where X ̂ is the Z-scored signal, X denotes the

processed signal, μ and σ is the mean and standard

deviation of the signal during rest time for each trial.

For the motor task recognition, the preprocessed

EEG signals were divided into smaller windows using

a sliding segmentation approach. Subsequently, a

feature extraction method based on wavelet analysis

was employed to extract pertinent features from each

segment. Each resulting feature matrix was used to

construct individual machine-learning models,

including Linear Discriminant Analysis (LDA), k-

nearest Neighbors (kNN), and Random Forest (RF).

A five-fold cross-validation technique was

applied to partition the extracted feature matrices into

training and testing datasets to ensure optimal data

utilization. The five-fold cross-validation involves

randomly dividing the entire dataset into five subsets,

and this process is repeated five times. During each

iteration, the model is trained on four of the folds,

while the remaining one is used for testing the model.

The performances of the models were assessed

using classification accuracy (CA; eqn. 2), positive

predictive value (PPV; eqn. 3), negative predictive

value (NPV; eqn. 4), and false positive rate (FPR;

eqn. 5). PPV is the percentage chance that a positive

result is a true positive. NPV is the percentage chance

that a negative result is a true negative. The FPR

measures the proportion of negative instances that are

inaccurately identified as positive instances.

𝐶𝐴



∑



𝑇𝑃



+𝑇𝑁



𝑇𝑃



+𝐹𝑁



+𝐹𝑃



+𝑇𝑁









𝑁

(

𝑃𝑃𝑉



=

𝑇𝑃



𝑇𝑃



+𝐹𝑃







(

𝑁𝑃𝑉



=

𝑇𝑁



𝑇𝑁



+𝐹𝑁







(

𝐹𝑃𝑅



=

𝐹𝑃



𝐹𝑃



+𝑇𝑁







(

where N denotes the number of ME classes, 𝑇𝑃



: true

positive, 𝐹𝑃



: false positive, 𝐹𝑁



: false positive, and

𝑇𝑁



: true negative.

The Friedman test was employed to check the

statistically significant effect between the

preprocessed signals with the artifact attenuation

methods and the original

EEG signal recordings.

3 RESULTS

3.1 Analysis of Cortical Activation via

Z-Score Power

Figure 2a-b depicts the average z-score results for the

μ and β bands during motor execution (ME) across all

participants. In the figures, each color bar represents

the raw signal and different artifact reduction

BIOSIGNALS 2024 - 17th International Conference on Bio-inspired Systems and Signal Processing

758

methods, with the z-score power varying for all

methods task by task.

Figure 2: Average z-score power for the (a) μ and (b) β

frequencies band across all participants.

Generally, the z-score value is expected to increase

(more negative) after artifacts are eliminated from the

signals. A high negative z-score value indicates a

stronger brain signal or activation and vice-versa.

Through careful examination, the ASR-based signal

obtained increased z-score value for all tasks

compared to the raw data in both bands.

Similarly, RELAX_MWF_wICA and

RELAX_MWF methods achieved better z-scores

than others for all tasks except for KGME. However,

the CCACCA method recorded the lowest z-score

value, followed by INFOMAX. The performance of

INFOMAX and CCACCA methods may be due to

removing more brain signals during pre-processing. It

could also be because of their inability to remove

other artifacts unrelated to ocular or muscular

artifacts. Across all tasks and bands, the

RELAX_MWF_wICA and ASR methods recorded

consistently higher average z-score values. At the

same time, some artifact removal methods showed an

increment in the z-score values; there is no statistical

significance between the raw data and the artifact

attenuation methods (

𝜇

: p = 0.1815 and β: p =

0.6126).

3.2 Performance Estimation Using

FPR, PPV and NPV

The effectiveness of the classifier's performances

with respect to the attenuation methods was validated

using false positive rate (FPR), positive predictive

value (PPV), and negative predictive value (NPV)

metrics. The average FPR, PPV, and NPV results for

the bands across tasks are presented in Figure 3a-b

using a scatter plot by group graph.

Figure 3: Performance evaluation of the methods for FPR,

PPV and NPV (a) 𝜇 and (b) 𝛽 bands.

Each plot consists of average (i) FPR, (ii) PPV, and

(iii) NPV (partitioned with a black dotted line) for the

classifiers (including LDA, KNN, and RF).

Observing the plots for the

𝜇

and

𝛽 bands, the

effectiveness of the raw data, and the attenuation

methods based on the RF classifier for the metrics are

also relatively the same compared to LDA and KNN

classifiers.

However, an obvious difference is noticeable

between the artifact attenuation methods and the raw

signal for all the metrics, especially PPV. Overall,

RELAX_MWF_wICA based on the LDA classifier

achieved the lowest FPR value (μ: 0.0108, β: 0.0026),

(i)

(ii) (iii)

RELAX_

MWF_wICA

RELAX_

MWF

(a) 𝝁-band

(i)

(ii)

(iii)

RELAX_

MWF_wICA

RELAX_

MWF

(b) 𝜷-band

Investigation of Artifact Contamination Impact on EEG Oscillations Towards Enhanced Motor Function Characterization

759

highest PPV (μ: 0.9679, β: 0.9923), and NPV (μ:

0.9892, β: 0.9974) values compared to other methods.

3.3 Evaluation of Individual Task

Decoding Performance

This section presents the individual ME task

recognition rate for LDA classifier because of its

performance in section 3.2. The obtained result in the

μ and β bands is presented in Figure 4a-b using a bar

plot graph.

Figure 4: Class-wise task decoding performance for (a) μ

(b) and β bands using LDA classifier.

The error bar on each preprocessing method

represents the standard deviation across participants.

From the results, the β band performs better than the

μ band, and it is clear that all artifact removal methods

were able to eliminate artifacts from the signals,

yielding varying classification performance.

Looking at the performance of each method, the

RELAX_MWF_wICA (p-value: 0.0022 for μ and β

bands) yielded the best average accuracies.

Specifically, μ: 96.98 ± 8.40%, 95.91 ± 6.05%, 96.49

± 5.00%, and 97.27 ± 4.70% for KGME, PGME,

WEME and WFME respectively. For the β band,

accuracies of KGME: 99.63 ± 1.02%, PGME: 98.77

± 2.54%, WEME: 98.97 ± 2.75% and WFME: 98.49

± 3.46%.

On the other hand, the least performance was

obtained by the raw signal with mean accuracies of

78.13 ± 10.68%, 71.62 ± 10.92%, 69.66 ± 10.60%,

71.79 ± 10.58%, for μ band. The β band recorded

87.14 ± 7.76%, 80.94 ± 10.79%, 79.30 ± 10.15%,

81.68 ± 10.29% KGME, PGME, WEME and WFME

respectively.

4 DISCUSSION AND

CONCLUSION

In this study, we demonstrated the impact of artifacts

on μ and β oscillations detected in the motor cortex

region of the brain during the execution of upper limb

motor tasks. Five automated artifact removal

methods, including single (INFOMAX) and multiple

artifact removal capabilities (ASR, CCACCA,

RELAX_MWF_wICA, and RELAX_MWF), were

individually employed to attenuate the artifacts

present in both bands. After the Independent

Component Analysis (ICA) decomposition of the

processed signal, an automated IC classifier, namely

IC_LABEL (Pion-Tonachini et al., 2019), was

utilized to detect and flag artifact-contaminated ICs

for removal from the processed signal.

The impact of artifacts were considered on

cortical activation strength and motor task

recognition. The brain activation response was

evaluated using z-score power. From the obtained

results, the average z-score values across participants

varied from task to task and between methods. Table

1 presents the average z-score power values across

participants and tasks for μ and β frequency bands.

When comparing the values between the bands, the μ

band showed better brain activation during the tasks

for all the methods compared to the β band. The ASR

Table 1: Average z-score power values all across

participants and tasks for the 𝜇 and 𝛽 bands.

Methods 𝝁 Band 𝜷 Band

RAWDATA -0.0314 -0.0182

INFOMAX -0.0302 -0.0197

ASR -0.0421 -0.0238

CCACCA -0.0244 -0.0157

RELAX_MWF_

ICA

-0.0370 -0.0231

RELAX_MWF -0.0353 -0.0202

BIOSIGNALS 2024 - 17th International Conference on Bio-inspired Systems and Signal Processing

760

method exhibited the highest z-score values in both

bands, followed by RELAX_MWF_wICA.

CCACCA recorded the least z-score value compared

to the raw data and other methods in the μ and β

bands. In other words, these methods, except

CCACCA, enhanced the brain response through the

mitigation of artifacts.

The outcome from the False Discovery Rate

(FDR) and Positive Predictive Value (PPV) and

Negative Predictive Value (NPV) validation metrics

shows that the RELAX_MWF_wICA-based method

is an accurate and effective model for processing

EEG signals compared to other methods. Similarly, in

individual task classification performance, the

RELAX_MWF_wICA method outperformed other

methods and the raw data in both bands.

Examining the performance of CCACCA, though

it has a low z-score power value compared to the raw

data, it recorded better decoding performance

compared to the raw data. One possible reason for this

could be that a strong brain response during the ME

task may not necessarily correlate with high

classification performance.

Overall, considering the impact of artifacts on

brain activation response and motor task

classification, the RELAX_MWF_wICA

demonstrated better performance, albeit with no

significant difference when compared with ASR and

RELAX_MWF methods. It performance could be

attributed to its status as a hybrid artifact attenuation

method that incorporates the advantages of MWF and

wICA.

The outcome of this work provides valuable

insights into the significance of using appropriate

methodology in the EEG signal-processing pipeline

to obtain precise estimations of motor brain activity,

thereby avoiding biased signal analyses and

interpretation.

It's important to note that this study is preliminary

and confined to a dataset consisting solely of healthy

subjects. The analysis utilized Z-score power

quantifier and statistical metrics. In our forthcoming

research, we plan to recruit stroke patients and

acquire EEG signals from them to validate our

findings. Furthermore, we will employ noteworthy

quantifiers to thoroughly investigate and analyze

EEG oscillatory rhythms.

ACKNOWLEDGEMENTS

The research work was supported in part by the

Ministry of Science and Technology of China under

grants (STI2030-Brain Science and Brain-Inspired

Intelligence Technology-2022ZD0210400), National

Natural Science Foundation of China under grant

(#62150410439), Ministry of Science and

Technology, Shenzhen (#QN2022032013L), and

Guangdong Basic and Applied Research Foundation

(#2023A1515011478).

The authors appreciates Zhengxiang Jing and

Yixin Ma, for their support in the data acquisition.

Thanks to all the recruited subjects who volunteered

to participate in the experiment.

REFERENCES

Ward, N. S. (2015). Does neuroimaging help to deliver

better recovery of movement after stroke?. Current

Opinion in Neurology, 28(4), 323-329.

Gui, X. U. E., Chuansheng, C. H. E. N., Zhong-Lin, L. U.,

& Qi, D. O. N. G. (2010). Brain imaging techniques and

their applications in decision-making research. Xin li

xue bao. Acta psychologica Sinica, 42(1), 120.

Jee, S. (2021). Brain Oscillations and Their Implications for

Neurorehabilitation. Brain & Neurorehabilitation,

14(1).

Wu, J., Srinivasan, R., Burke Quinlan, E., Solodkin, A.,

Small, S. L., & Cramer, S. C. (2016). Utility of EEG

measures of brain function in patients with acute stroke.

Journal of neurophysiology, 115(5), 2399-2405.

Lin, C. T., Chuang, C. H., Cao, Z., Singh, A. K., Hung, C.

S., Yu, Y. H., ... & Wang, S. J. (2017). Forehead EEG

in support of future feasible personal healthcare

solutions: Sleep management, headache prevention,

and depression treatment. IEEE Access, 5, 10612-

10621.

Asogbon, M. G., Samuel, O. W., Li, X., Nsugbe, E.,

Scheme, E., & Li, G. (2021). A linearly extendible

multi-artifact removal approach for improved upper

extremity EEG-based motor imagery decoding. Journal

of Neural Engineering.

Anupama, H. S., Cauvery, N. K., & Lingaraju, G. M.

(2012). Brain computer interface and its types-a study.

International Journal of Advances in Engineering &

Technology, 3(2), 739.

Maura, R. M., Rueda Parra, S., Stevens, R. E., Weeks, D.

L., Wolbrecht, E. T., & Perry, J. C. (2023). Literature

review of stroke assessment for upper-extremity

physical function via EEG, EMG, kinematic, and

kinetic measurements and their reliability. Journal of

NeuroEngineering and Rehabilitation, 20(1), 1-32.

Bartur, G., Pratt, H., & Soroker, N. (2019). Changes in mu

and beta amplitude of the EEG during upper limb

movement correlate with motor impairment and

structural damage in subacute stroke. Clinical

Neurophysiology, 130(9), 1644-1651.

Cassidy, J. M., Wodeyar, A., Srinivasan, R., & Cramer, S.

C. (2021). Coherent neural oscillations inform early

stroke motor recovery. Human Brain Mapping, 42(17),

5636-5647.

Investigation of Artifact Contamination Impact on EEG Oscillations Towards Enhanced Motor Function Characterization

761

Keser, Z., Buchl, S. C., Seven, N. A., Markota, M., Clark,

H. M., Jones, D. T., ... & Lundstrom, B. N. (2022).

Electroencephalogram (EEG) with or without

transcranial magnetic stimulation (TMS) as biomarkers

for post-stroke recovery: a narrative review. Frontiers

in Neurology, 13, 827866.

Thibaut, A., Simis, M., Battistella, L. R., Fanciullacci, C.,

Bertolucci, F., Huerta-Gutierrez, R.,... & Fregni, F.

(2017). Using brain oscillations and corticospinal

excitability to understand and predict post-stroke motor

function. Frontiers in neurology, 8, 187.

López-Larraz, E., Figueiredo, T. C., Insausti-Delgado, A.,

Ziemann, U., Birbaumer, N., & Ramos-Murguialday,

A. (2018). Event-related desynchronization during

movement attempt and execution in severely paralyzed

stroke patients: An artifact removal relevance analysis.

NeuroImage: Clinical, 20, 972-986.

Jutten, C., & Herault, J. (1991). Blind separation of sources,

part I: An adaptive algorithm based on neuromimetic

architecture. Signal processing, 24(1), 1-10.

Comon, P. (1994). Independent component analysis, a new

concept?. Signal processing, 36(3), 287-314.

Bloniasz, P. (2022). Artifact Subspace Reconstruction

(ASR) for electroencephalography artifact removal

must be optimized for each unique dataset.

Chang, C. Y., Hsu, S. H., Pion-Tonachini, L., & Jung, T. P.

(2019). Evaluation of artifact subspace reconstruction

for automatic artifact components removal in multi-

channel EEG recordings. IEEE Transactions on

Biomedical Engineering, 67(4), 1114-1121.

Blum, S., Jacobsen, N. S., Bleichner, M. G., & Debener, S.

(2019). A Riemannian modification of artifact subspace

reconstruction for EEG artifact handling. Frontiers in

human neuroscience, 13, 141.

Gómez-Herrero, G., De Clercq, W., Anwar, H., Kara, O.,

Egiazarian, K., Van Huffel, S., & Van Paesschen, W.

(2006, June). Automatic removal of ocular artifacts in

the EEG without an EOG reference channel. In

Proceedings of the 7th Nordic signal processing

symposium-NORSIG 2006 (pp. 130-133). IEEE.

De Clercq, W., Vergult, A., Vanrumste, B., Van Paesschen,

W., & Van Huffel, S. (2006). Canonical correlation

analysis applied to remove muscle artifacts from the

electroencephalogram. IEEE transactions on

Biomedical Engineering, 53(12), 2583-2587.

Bailey, N., Biabani, M., Hill, A. T., Miljevic, A., Rogasch,

N. C., McQueen, B., ... & Fitzgerald, P. (2022).

Introducing RELAX (the Reduction of

Electroencephalographic Artifacts): A fully automated

pre-processing pipeline for cleaning EEG data-Part 1:

Algorithm and Application to Oscillations. BioRxiv,

2022-03.

Somers, B., Francart, T., & Bertrand, A. (2018). A generic

EEG artifact removal algorithm based on the multi-

channel Wiener filter. Journal of neural engineering,

15(3), 036007.

Castellanos, N. P., & Makarov, V. A. (2006). Recovering

EEG brain signals: Artifact suppression with wavelet

enhanced independent component analysis. Journal of

neuroscience methods, 158(2), 300-312.

Pion-Tonachini, L., Kreutz-Delgado, K., & Makeig, S.

(2019). ICLabel: An automated

electroencephalographic independent component

classifier, dataset, and website. NeuroImage, 198, 181-

197.

Delorme, A., & Makeig, S. (2004). EEGLAB: an open

source toolbox for analysis of single-trial EEG

dynamics including independent component

analysis. Journal of neuroscience methods, 134(1), 9-

21.

The MathWorks Inc. (2019). MATLAB version: 9.7.0

(R2019b), Natick, Massachusetts: The MathWorks Inc.

https://www.mathworks.com

Leonardi, G., Ciurleo, R., Cucinotta, F., Fonti, B., Borzelli,

D., Costa, L., ... & Alito, A. (2022). The role of brain

oscillations in post-stroke motor recovery: An

overview. Frontiers in Systems Neuroscience, 16,

947421.

BIOSIGNALS 2024 - 17th International Conference on Bio-inspired Systems and Signal Processing

762