=Paper=
{{Paper
|id=Vol-2249/paper1
|storemode=property
|title=Machine Learning of Multi-channel Electroencephalographic Data
|pdfUrl=https://ceur-ws.org/Vol-2249/AIIA-DC2018_paper_1.pdf
|volume=Vol-2249
|dblpUrl=https://dblp.org/rec/conf/aiia/Saibene18
}}
==Machine Learning of Multi-channel Electroencephalographic Data==
https://ceur-ws.org/Vol-2249/AIIA-DC2018_paper_1.pdf
Machine Learning of Multi-channel
Electroencephalographic Data?
Aurora Saibene1
Department of Informatics, Systems and Communication, University of Milano -
Bicocca, Milan, Italy
a.saibene2@campus.unimib.it
Abstract. Machine Learning techniques have been recently applied in
the healthcare field and particularly for electroencephalographic signal
classification, opening new possibilities for brain activities and diseases
analysis through peculiar applications like the Brain Computer Inter-
faces.
The project proposal for the Ph.D. thesis work briefly described in the
following wants to address the problems arising from these biomedical
heterogeneous data, starting from the preliminary signal processing for
noise removal, moving to possible data normalisation for subject and pop-
ulation based analysis and exploiting the outputted manipulated data to
create classifiers for peculiar brain activities labelling, diseases identifi-
cation, Brain Computer Interface development.
These steps will require an evaluation of the state-of-the-art, which present
mostly semi-automatic or manual signal processing techniques, that will
be used to create fully automated denoising modules for every type of
data and integrated for scenario-dependent signal reconstruction proce-
dures. Also, there is a narrow number of studies addressing the normali-
sation problem, which is to be considered for population-based analysis.
Finally, the recent works on electrophysiological signal classification will
be used to evaluate commonly used Machine Learning algorithms and to
create best-practices for feature extraction, a benchmark for deep learn-
ing techniques application and the study of Brain Computer Interface
mainly for rehabilitation purposes.
Keywords: Brain Computer Interface · Deep learning · Electroencephalo-
gram · Machine Learning · Signal processing.
1 Introduction
In the last decades the constant technological improvement and the availability
of a greater amount of data have led to an increasing interest over the applica-
tion of Machine Learning (ML) techniques in the biomedical field, posing new
challenges for the development of faster and more accurate classification algo-
rithms.
?
Main advisor: Francesca Gasparini, Department of Informatics, Systems and Com-
munication, University of Milano - Bicocca, Milan, Italy.
2 A. Saibene
The project proposal for the Ph.D. thesis work described in the following, wants
to address the problems arising from the heterogeneity of a peculiar kind of
biomedical data, i.e. the Electroencephalographic (EEG) signals.
Recently, EEG has begun to be extensively used in the medical and research fields
due to its characteristics: it is non-invasive, records the cerebral bio-electric po-
tentials through multiple sensors (electrodes) placed on the scalp [12], has a high
temporal resolution [8]. The recording produces a multi-channel signal, i.e. the
EEG data structure is usually in the form of a matrix, whose rows correspond
to the sensors and the columns to the electric potential recorded in a specific
time.
The EEG has been used, for example, in face recognition experiments [9], in the
analysis of vegetative or minimally conscious states [5], for the development of
Brain Computer Interfaces (BCIs) for rehabilitation purposes and to allow the
control of medical devices (e.g. wheelchairs) [6].
However, the development of these kinds of applications encounters some difficul-
ties due to the fact that the EEG signal is weak, time varying and easily affected
by biological (ocular, muscular, cardiac movements) and non-physiological (di-
rect current, electrical leakage) noises [11], which must be removed to allow a
better analysis without losing useful experimental data.
Also, to make assumptions over a population, the specificity of each recording
must be considered and so the difficulty of normalising the data, i.e. trying to
fit a specific recording into a canonical space, arises. This necessity, to which an
unique solution is yet to be found, may allow researchers to move from a classical
subject-based analysis to a population-based one.
2 Problem statement
Starting from an experiment of face recognition conducted in collaboration with
professor Daini1 [9], the Ph.D. project mainly wants to refine the denoising
techniques used in the aforementioned work and move to the identification of
metrics for discrimination of noisy patterns in the EEG signal.
Afterwards, the obtained features will be used to train a classifier to develop a
tool for semi- to completely-automated noise removal, which will be expected to
run with slight variations in any kind of case scenario.
Finally, the Ph.D. work will consider the problem of normalisation to allow
analysis inter-subject or between different populations (e.g. access the differences
in the brain activations between a normal recogniser population and an impaired
one). This last step will be introduced as an addition to the defined pipeline,
lacking of state-of-the-art references and having the heterogeneity and subject-
specificity of the EEG data as difficult issues to address.
1
Department of Psychology, University of Milano - Bicocca, Milan, Italy
Machine Learning of Multi-channel Electroencephalographic Data 3
3 State-of-the-art and Methodology
As well depicted by Urigüen et al. [11], concerning the EEG signal pre-processing,
the state-of-the-art presents a great number of noise removal algorithms, but
there are no automatic procedures for the identification and correction of noisy
patterns that could be considered completely effective and efficient, preferring
more consolidated methods for semi-automatic inspection and noise suppression
or even manual rejection of the noisy recording portions.
As stated in the introduction, the difficulties of developing an automated pro-
cedure arise from the specific nature of the EEG signal: it varies from patient
to patient, depends on the recording hardware, may present a mixture of the
actual electrophysiological data and contaminated components.
Denoising techniques are mostly scenarios-dependent, but despite that, some re-
cent studies confirm the success of methods used in the identification of noisy
components and suggests some useful tips for EEG automatic signal manipu-
lation. For example, Al-Qazzaz et al. [1] present an automatic noise removal
pipeline specifically suitable for working memory tasks in normal and affected
by dementia populations.
Therefore, starting from the study of semi-automatic procedures, the noise re-
moval methodology suggested for the Ph.D. work wants to obtain a modular set
of procedures, that could be algorithms applicable to any kind of experiment
and to scenario-dependent ones and move to the identification of noisy patterns
through some metrics, which could be used for the training of peculiar classifiers.
In fact, the state-of-the-art presents a good amount of ML techniques for EEG
signal classification. The most used are k-Nearest Neighbour (k-NN), which as-
signs to a tested sample the label of the k-nearest training sample, and Support
Vector Machine (SVM), that segregates the data through an hyper plan with
maximal margins, as supervised classifiers and Naı̈ve Bayes (NB), which is based
on Bayes’ theorem and determines the class of earlier probabilities through a
maximum probability algorithm and uses a feature probability distribution from
a training set [2], as a probabilistic one. The common characteristic of these
methods is the necessity of having a validated training set.
To allow a classification that could be executed with both un- and supervised
approaches, recently the research community involved in healthcare began to
explore and develop applications based on deep learning techniques.
In this regard, the review edited by Miotto et al. [7] describes these applications
challenges and opportunities, which arise even in the more specific domain of
EEG classification.
Therefore, the Ph.D. project wants to (1) start from the supervised classifiers
and create best practices for feature extraction in scenario-based experiment
and for noise patterns classification, where the most used features would be ex-
tracted computing the Power Spectral Density (PSD) over the frequency bands
that characterise the EEG signal (e.g. average spectral power, spectral power for
each frequency and approximate entropy) and (2) create a benchmark through
which evaluate the possibility of developing a deep learning classifier.
The last cited item opens new issues, like the fact that a deep learning model
4 A. Saibene
requires a great amount of data, which should be clean and well-structured. This
could be achieved by manipulating the raw state of the EEG signal to be more
clean and interpretable through the signal processing procedure previously cited.
Also, there are no robust and well-maintained deep learning procedures on EEG
applications, even though there have been recent studies for the use of Con-
volutional Neural Networks (ConvNets). Schirrmeister et al. [10] developed a
new method for visualising learned features and showed how to design and train
ConvNets to decode task-related information from raw EEG data. However, the
outputs given by ConvNets are frequently difficult to interpret and this method
involves a good number of hyperparameters, but ConvNets have some interesting
characteristics that could represent a good compromise to choose them among
the other ML algorithms: there is not the necessity of a priori features selection,
they are scalable on large datasets and exploit the hierarchical structure typical
of the natural signals [10].
Finally, from the necessity of evaluating brain activities and functions in non-
and pathological conditions or between different states, comes the issue regard-
ing a proper way to normalise the EEG signal, which is not only characterised by
a high dimensionality, but also - as for other biomedical data - presents hetero-
geneity, temporal dependency, sparsity and irregularity. This problem has been
discussed between numerous researchers (mainly on ResearchGate 2 ), but an uni-
vocal solution has yet to be found.
The advised approaches emerging from the discussions and, only minimally, in
the state-of-the-art are mainly (1) the normalisation of the power spectra for
each frequency band and sensor [4], which could be difficult to apply due to the
EEG nature and (2) the standardisation of the sensors voltage by using the z-
score to detect differences between groups, usually applied on resting state EEG
recording and thus inappropriate for task-based experiments.
The Ph.D. project wants to evaluate, as an additional step, the suggested solu-
tions and find better methods for data normalisation, moving from a subject-
based approach to a population-based one.
4 Conclusion
The project for the Ph.D. work is divided in three main steps: signal processing,
classification of heterogeneous EEG data and normalisation.
Each of them could be divided in and expanded with different sub-modules: novel
algorithms for denoising based on peculiar signal characteristics, normalisation
procedures less commonly cited as the min-max normalisation [13], new features
and classifiers that could be useful for BCI development and whose accuracy
could be evaluated verifying for example classes balance, kappa metric, confusion
matrix on offline data [6].
The cleared signal and noisy patterns classification will be validated by experts
of the Department of Psychology at University of Milano - Bicocca, evaluating
2
https://www.researchgate.net/post/Normalization_of_resting_EEG_data_
for_comparisons_between_different_subjects
Machine Learning of Multi-channel Electroencephalographic Data 5
the accuracy, sensitivity and specificity of the obtained results [3].
Therefore, the presented paper wants to give guidelines for EEG signal processing
and classification, given that the research field that the Ph.D. project wants to
address is in constant evolution and improvement.
References
1. Al-Qazzaz, N.K., Hamid Bin Mohd Ali, S., Ahmad, S.A., Islam, M.S., Escudero,
J.: Automatic artifact removal in EEG of normal and demented individuals using
ICA–WT during working memory tasks. Sensors 17(6), 1326 (2017)
2. Amin, H.U., Mumtaz, W., Subhani, A.R., Saad, M.N.M., Malik, A.S.: Classification
of EEG signals based on pattern recognition approach. Frontiers in computational
neuroscience 11, 103 (2017)
3. Barua, S., Ahmed, M.U., Ahlström, C., Begum, S.: Automatic driver sleepiness
detection using EEG, EOG and contextual information. Expert systems with ap-
plications 115, 121–135 (2019)
4. Haegens, S., Cousijn, H., Wallis, G., Harrison, P.J., Nobre, A.C.: Inter-and intra-
individual variability in alpha peak frequency. Neuroimage 92, 46–55 (2014)
5. Lehembre, R., Bruno, M.A., Vanhaudenhuyse, A., Chatelle, C., Cologan, V.,
Leclercq, Y., Soddu, A., Macq, B., Laureys, S., Noirhomme, Q.: Resting-state
EEG study of comatose patients: a connectivity and frequency analysis to find dif-
ferences between vegetative and minimally conscious states. Functional neurology
27(1), 41 (2012)
6. Lotte, F., Bougrain, L., Cichocki, A., Clerc, M., Congedo, M., Rakotomamonjy,
A., Yger, F.: A review of classification algorithms for EEG-based brain–computer
interfaces: a 10 year update. Journal of neural engineering 15(3), 031005 (2018)
7. Miotto, R., Wang, F., Wang, S., Jiang, X., Dudley, J.T.: Deep learning for health-
care: review, opportunities and challenges. Briefings in bioinformatics (2017)
8. Radüntz, T., Scouten, J., Hochmuth, O., Meffert, B.: EEG artifact elimination by
extraction of ICA-component features using image processing algorithms. Journal
of neuroscience methods 243, 84–93 (2015)
9. Saibene, A., Corchs, S., Daini, R., Facchin, A., Gasparini, F.: EEG Data of Face
Recognition in Case of Biological Compatible Changes: A Pilot Study on Healthy
People. In: Proceedings of the 15th International Joint Conference on e-Business
and Telecommunications - Volume 1: ICETE,. pp. 414–420. INSTICC, SciTePress
(2018). https://doi.org/10.5220/0006909104140420
10. Schirrmeister, R.T., Springenberg, J.T., Fiederer, L.D.J., Glasstetter, M.,
Eggensperger, K., Tangermann, M., Hutter, F., Burgard, W., Ball, T.: Deep learn-
ing with convolutional neural networks for EEG decoding and visualization. Human
brain mapping 38(11), 5391–5420 (2017)
11. Urigüen, J.A., Garcia-Zapirain, B.: EEG artifact removalstate-of-the-art and guide-
lines. Journal of neural engineering 12(3), 031001 (2015)
12. Zani, A., Mado Proverbio, A., Mangun, G., M. Fletcher, E., Brattico, E., Olcese, C.,
Tervaniemi, M., Ntnen, R., L. Wilding, E., Federmeier, K., Kutas, M., T. Knight,
R., Scabini, D., Luu, P., Tucker, D.: Metodi Strumentali nelle Neuroscienze Cog-
nitive. EEG ed ERP- Instrumental Methods in Cognitive Neuroscience. EEG and
ERP (11 2013)
13. Zhang, X., Yao, L., Zhang, D., Wang, X., Sheng, Q.Z., Gu, T.: Multi-person
brain activity recognition via comprehensive EEG signal analysis. arXiv preprint
arXiv:1709.09077 (2017)