A Statistical Method for EEG Channel Selection

BIANCA-ALEXANDRA ZÎRNĂ, MĂDĂLIN CORNELIU FRUNZETE

Department of Bioengineering and Biotechnology,

National University of Science and Technology Politehnica Bucharest,

Splaiul Independenței 313, Bucharest,

ROMANIA

Abstract: - A new statistical analysis of medical signals is proposed based on information extracted for

diagnosis. Several specialized tests can be run on electroencephalography (EEG) signal due to its nonlinear

behavior. Epileptic EEG signals are the main topic of this research. A statistics-based approach is proposed for

the automatic selection of the relevant channels that will not only provide more accurate characterization but

also require fewer computing resources. Moreover, the outcomes given by specific electrode pairs might

indicate epileptic focus. By selecting the most efficient electrode placement, alarms for detecting inappropriate

medical behavior could then be triggered.

Key-Words: - EEG, epilepsy, channel selection, probability density function, cumulative distribution function,

statistical methods.

Received: April 9, 2024. Revised: August 13, 2024. Accepted: September 18, 2024. Published: October 30, 2024.

1 Introduction

This paper proposes a new method for medical

signal interpretation by considering the EEG

(electroencephalography) signal as a vector that is

computed to extract statistical features, [1]. Such

interpretation is oriented toward characterization in

a specific case study: epileptic activity.

Epilepsy is one of the most common

neurological diseases and affects people of all ages,

ethnic groups, socioeconomic levels, and

geographical regions, [2]. Despite not being a

mental disease, epilepsy affects the brain and

frequently causes seizures, [3]. Seizures are

temporary spikes in brain activity that can cause a

wide range of symptoms, such as trembling

uncontrollably, unconsciousness, stiffness, tingling,

fainting, etc. [4]. EEG is a low-cost, accurate time-

resolution method of non-invasively measuring the

electrical fields of the brain with the aid of

electrodes placed on the scalp, [5], [6]. Depending

on how the electrodes are placed and distributed

across the surface of the scalp, potential differences

between certain pairs of electrodes or the difference

between a single electrode and the reference

electrode may be measured. The resulting signals

are amplified and displayed in quasi-real time as

particular sequences of waves.

A strategy for choosing the best acquisition

channels is suggested using statistical analysis in

order to improve the effectiveness of the signal

processing on those channels, shorten acquisition

times, and optimize the entire process, [7]. This

approach can also be used to distinguish between

signals with epileptic seizures and normal ones or to

determine the epileptic focus of the seizures based

on the channel locations, [8].

In Section 2, some medical information is

recalled to support the diagnosis claimed in this

paper. The results of the statistical analysis are

presented in Section 3. The conclusion is given in

the final section of the paper.

2 Problem Formulation

2.1 Various Methods for Channel Selection

In order to obtain the best possible results from a

potential seizure classification, choosing the optimal

recording channel is a very important step, focusing

on the channels that contain the most useful

information. This process is essential for reducing

computational complexity and time, enabling the

use of portable headsets at a low cost, or activating

only certain electrodes in that region.

Using the CHB-MIT database, the paper [9]

proposes a multi-objective optimization approach

for EEG channel selection that is based on the non-

dominated sorting genetic algorithm. With only one

EEG channel, the results showed an accuracy of up

to 100%, indicating that portable EEG seizure

detection systems and the classification of epileptic

seizures with a few electrodes are feasible. Papers

WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE

DOI: 10.37394/23208.2024.21.34

Bianca-Alexandra Zîrnă, Mădălin Corneliu Frunzete

E-ISSN: 2224-2902

339

Volume 21, 2024

[10] and [11] present a method for EEG channel

selection and seizure prediction based on statistical

probability distributions of the EEG signals. To

reduce the use of memory and computational

complexity while making the system suitable for

real-time applications, the paper [12] proposes a

channel selection algorithm, which includes testing

every feature for every channel and selecting the

best outcomes. This algorithm is also tested on the

CHB-MIT EEG database. Using Model Agnostic

Meta-Learning (MAML) applied to a Deep Neural

Network (DNN), the paper [13] provides an

improved channel reduction for seizure prediction,

selecting and optimizing the number of channels

from all the subjects of the CHB-MIT Dataset.

The channels can be selected based on the results

obtained from feature extraction, as mentioned

before [9], [12]. The disadvantage of this method is

that it is time-consuming to test all features on all

channels. The channels can also be chosen

according to the location of the epileptic seizure and

its nature, but the disadvantage of this method is that

the type of seizure must be known, and, in the case

of a general seizure, all the channels seem to be

relevant.

Therefore, the proposed method involves a

statistical method that uses the probability density

function (PDF). A probability density function

shows the values that, for a certain draw or time, are

most likely to occur in a data process. This method

can also represent a solution to determine the type

of epileptic seizure because finding the optimal

channel means determining the best pair of

electrodes. Depending on the location of this pair,

the exact type of seizure can be identified [14], such

as a partial or generalized seizure, as presented in

Figure 1.

Fig. 1: EEG signals with partial and generalized

seizures, [15]

2.1.1 Database

For testing the proposed algorithm, a database from

PhysioNet, named CHB-MIT Database, was used

[16]. This database, obtained at Boston Children's

Hospital, contains EEG recordings from 23 children

with intractable seizures. The subjects were

monitored for several days after stopping the

antiepileptic medication to characterize their

seizures and evaluate their suitability for surgical

interventions. As the purpose of this paper is to

determine the best channels for epilepsy detection,

only seizure signals were selected from 24 subjects

(subject 1 was recorded twice), resulting in 140

signals on 23 channels (other additional channels

were removed). The signals were sampled at a 256

Hz rate.

3 Problem Solution

The EEG signals from the database recalled in the

previous section were computed in separate

channels, resulting in a total of 3220 signals. The

probability density function was applied to compute

the mean (µ) and standard deviation (σ), values that

are then used to select the best channels for each

subject (Figure 2).

Fig. 2: The block diagram of the proposed method

As mentioned in [17], the same EEG signals with

the best binary classification results were the filtered

ones. Therefore, a bandpass filter was used in order

to keep the frequencies in the [0.5, 12] range, where

most of the seizures occur, [18].

All 23 channels that are first used are shown in

Figure 3, displayed in the same range ([-2000,

2000]) µV in order to highlight variations in

amplitude.

Fig. 3: Patient 1, signal 18 – the 23 corresponding

channels

WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE

DOI: 10.37394/23208.2024.21.34

Bianca-Alexandra Zîrnă, Mădălin Corneliu Frunzete

E-ISSN: 2224-2902

340

Volume 21, 2024

Signals with seizures have larger amplitudes and

repetitive wave sequences than those without;

therefore, both µ and σ are larger. Due to these

factors, this method can also be used to discriminate

between signals with and without seizure or ictal

(seizure) and inter-ictal (seizure-free) epochs, [19].

Inter-ictal epochs have smaller, constant amplitudes,

while ictal epochs are identified by a sudden

increase in the amplitude, [20]. Good channels

contain a lot of information, and the weaker

channels contain less information. However,

channels that contain too much information might

be affected by noise or artifacts, which is why the

filtering stage is necessary. Due to these factors, the

data was normalized in the [0, 1] range, and some

thresholds were set compared to the mean values of

µ and σ. Figure 4 presents three PDF representations

from three channels of one signal from subject 1,

with different means and probabilities of the

outcomes.

Fig. 4: Three PDF representations - same signal,

different channels

For a better visual representation, the cumulative

distribution function (CDF), another method used to

describe the distribution of random variables, was

computed for the same three signals (Figure 5).

The mean and standard deviation were computed

for each signal and for each subject. If both values

were in the proper range (between 20% and 70%

higher than the mean value), that channel was stored

in a vector, and the rest were removed. For each

subject, it resulted in a vector with all the stored

channels, and the number of times a channel

appeared was computed. The channels that appeared

the most were finally kept. The channel selection

algorithm is presented in Figure 6.

Fig. 5: CDF representations corresponding to the

previous channels

Two different results for each channel are

presented in Figure 9: the number of appearances

and the number of times each channel appeared the

most. In the first case, channels FP1-F7, F7-T7,

FP1-F3, FP2-F4, F8-T8, P8-O2, and P7-T7 have the

highest number of appearances. If the subjects with

less than 5 seizure signals are removed, these

channels still have the highest number of

appearances, except for the P7-T7 channel. As for

the number of times each channel appeared the

most, the F7-T7 channel is the first one, followed by

the P7-T7 channel.

Some channels never appeared or had never been

the most efficient ones, such as T8-P8, F4-C4, or

C4-P4. Therefore, these channels can be removed

from the computations or simply deactivated during

the recording sessions.

Fig. 6: Block diagram of the channel selection

algorithm

To illustrate the facts mentioned above, Figure 7

and Figure 8 present two PDF and CDF

WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE

DOI: 10.37394/23208.2024.21.34

Bianca-Alexandra Zîrnă, Mădălin Corneliu Frunzete

E-ISSN: 2224-2902

341

Volume 21, 2024

representations of one of the best channels (F7-T7)

and one of the weakest (F4-C4).

Fig. 7: PDF representations – 2 different channels

Fig. 8: CDF representations – 2 different channels

Fig. 9: Channels statistics

The fact that for most of the subjects, there were

a few efficient channels means that they

experienced partial seizures since a generalized

seizure is visible on every channel. Besides the

removal or deactivation of the inefficient pairs of

electrodes, the type of seizure can be identified

depending on the location of the electrodes. It is

known that the International 10-20 system of EEG

electrode positions and nomenclature was used for

these recordings. In Figure 10, these electrode

placements are presented alongside the highlighted

electrodes and the marked pairs.

Fig. 10: The International 10-20 system of EEG

electrode positions and nomenclature and the best

pairs of electrodes.

From Figure 10, it can be seen that the pairs are

located on the exterior of the hemispheres and not in

the center. Also, the positions are almost

symmetrical on the left and right sides. These

locations suggest that even though the type of

epilepsy might be different (although all subjects

seem to experience partial seizures), the same

channels offer the best information. This means that

there is no need to use all the channels; therefore,

the weaker channels can be removed from the

computations or deactivated during the recording

sessions. There are improvements in time,

computing complexity, and the risk of inaccurate

data in both scenarios.

4 Conclusion

This study attempted to determine the optimal

acquisition channel of an EEG signal using

statistical methods, more specifically PDF, which is

also crucial for obtaining accurate and effective

characteristics. If the best channel is not selected,

then the corresponding signals will contain too little

useful information or too much information affected

by noise and artifacts, and will also complicate the

computations due to the higher number of signals

(channels), some of which do not even contain

relevant information.

On one hand, this method can be used to identify

the type of epileptic seizure. If the seizure appears

on all channels, it means that the subject

experiences generalized epilepsy. If the seizure

occurs on specific channels, it means it is partial

WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE

DOI: 10.37394/23208.2024.21.34

Bianca-Alexandra Zîrnă, Mădălin Corneliu Frunzete

E-ISSN: 2224-2902

342

Volume 21, 2024

epilepsy (e.g., frontal lobe epilepsy). As was

previously presented, the best channels were mostly

located in the frontal and temporal lobes, indicating

two common types of epilepsy.

On the other hand, by discovering a method to

detect the optimal channel according to the location

or type of seizure, an algorithm that "filters" the

channels can be obtained, i.e., automatically

activates only certain electrodes and sends the best-

acquired signals without the need to manually

choose the best channel. However, finding the best

channel means finding the best pair of electrodes,

and based on their placement, the specific type of

seizure can be recognized.

The proposed approach is very simple but

effective, with relatively few computations. Thus,

the algorithm can be used in a portable device for

real-time EEG signal acquisition since the channel

selection is precise as well as fast.

Acknowledgement:

This work was supported by a grant

from the National Program for Research of the

National Association of Technical Universities -

GNAC ARUT 2023.

References:

[1] A. Avasiloaie, B.A. Zîrnă, M. Oniga, A.E.

Sultana, M.C. Frunzete, Classification of

Mental Effort vs. Relaxation States Using

EEG Signals, E-Health and Bioengineering

Conference (EHB), Iasi, Romania, 2022,

pp.1-4, doi:

10.1109/EHB55594.2022.9991299.

[2] E. Beghi, The Epidemiology of Epilepsy,

Neuroepidemiology, Vol.54, No.2, 2020,

pp.185-191, doi:10.1159/000503831.

[3] K.A. Mora Rodríguez, S.R. Benbadis,

Managing Antiepileptic Medication in

Dialysis Patients, Current Treatment Options

in Neurology, Vol.20, No.45, 2018, pp.1-11,

doi: 10.1007/s11940-018-0530-5.

[4] C.E. Stafstrom, L. Carmant, Seizures and

Epilepsy: An Overview for Neuroscientists,

Cold Spring Harbor Perspectives in

Medicine, Vol.5, No.6, 2015, p.a022426, doi:

10.1101/cshperspect.a022426.

[5] A. Biasiucci, B. Franceschiello, M.M.

Murray, Electroencephalography, Current

Biology, Vol.29, No.3, 2019, pp.R80-R85,

doi: 10.1016/j.cub.2018.11.052.

[6] L. Zhiwei, S. Minfen, Classification of

Mental Task EEG Signals Using Wavelet

Packet Entropy and SVM, 8th International

Conference on Electronic Measurement and

Instruments, Xi'an, China, 2007, pp.3-906-3-

909, doi: 10.1109/ICEMI.2007.4351064.

[7] S. Nakano, F.N. Kondo, Customer

Segmentation with Purchase Channels and

Media Touchpoints Using Single Source

Panel Data, Journal of Retailing and

Consumer Services, Vol.41, 2018, pp.142-

152, doi: 10.1016/j.jretconser.2017.11.012.

[8] Y. Zhang, Y. Guo, P. Yang, W. Chen, B. Lo,

Epilepsy Seizure Prediction on EEG Using

Common Spatial Pattern and Convolutional

Neural Network, IEEE Journal of

Biomedical and Health Informatics, Vol.24,

No.2, 2019, pp.465-474, doi:

10.1109/JBHI.2019.2933046.

[9] L.A Moctezuma, M. Molinas, EEG Channel-

Selection Method for Epileptic-Seizure

Classification Based on Multi-Objective

Optimization, Frontiers in neuroscience,

Vol.14, 2020, p.537633, doi:

10.3389/fnins.2020.00593.

[10] F. Ibrahim, S. Abd-Elateif El-Gindy, S.M.

El-Dolil, A.S. El-Fishawy, E.S.M. El-Rabaie,

M.I. Dessouky, I.M. Eldokany, T.N.

Alotaiby, S.A. Alshebeili, F.E. Abd El-

Samie, A Statistical Framework for EEG

Channel Selection and Seizure Prediction on

Mobile, International Journal of Speech

Technology, Vol.22, 2019, pp.191-203, doi:

10.1007/s10772-018-09565-7.

[11] H.M. Emara, M. Elwekeil, T.E. Taha, A.S.

El-Fishawy, E.S.M. El-Rabaie, T. Alotaiby,

S.A. Alshebeili, F.E. Abd El-Samie, Hilbert

Transform and Statistical Analysis for

Channel Selection and Epileptic Seizure

Prediction, Wireless Personal

Communications, Vol.116, 2021, pp.3371-

3395, doi: 10.1007/s11277-020-07857-3.

[12] H. Daoud, M.A. Bayoumi, Efficient

Epileptic Seizure Prediction Based on Deep

Learning, IEEE Transactions on Biomedical

Circuits and Systems, Vol.13, No.5, 2019,

pp.804-813, doi:

10.1109/TBCAS.2019.2929053.

[13] A. Romney, V. Manian, Optimizing Seizure

Prediction from Reduced Scalp EEG

Channels Based on Spectral Features and

MAML, IEEE Access, Vol.9, 2021,

pp.164348-164357, doi:

10.1109/ACCESS.2021.3134166.

[14] X. Gao, Y. Yang, F. Zhang, F. Zhou, J. Zhu,

J. Sun, K. Xu, Y. Chen, A Feature Extraction

Method for Seizure Detection Based on

WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE

DOI: 10.37394/23208.2024.21.34

Bianca-Alexandra Zîrnă, Mădălin Corneliu Frunzete

E-ISSN: 2224-2902

343

Volume 21, 2024

Multi-Site Synchronous Changes and Edge

Detection Algorithm, Brain Sciences,

Vol.13, No.1, 2022, p.52, doi:

10.3390/brainsci13010052.

[15] M.T. Sadiq, H. Akbari, A.U. Rehman, Z.

Nishtar, B. Masood, M. Ghazvini, J. Too, N.

Hamedi, M.K. Kaabar, Exploiting Feature

Selection and Neural Network Techniques

for Identification of Focal and Nonfocal EEG

Signals in TQWT Domain, Journal of

Healthcare Engineering, No.1, 2021, p.

6283900, doi: 10.1155/2021/6283900.

[16] A.L. Goldberger, L.A. Amaral, L. Glass,

J.M. Hausdorff, P.C. Ivanov, R.G. Mark, J.E.

Mietus, G.B. Moody, CK Peng, H.E.

Stanley, PhysioBank, PhysioToolkit, and

PhysioNet: Components of a New Research

Resource for Complex Physiologic Signals,

circulation, Vol.101, No.23, 2000, pp.e215–

e220, doi: 10.1161/01.CIR.101.23.e215.

[17] B.A. Zîrnă, D. Mihailovschi, M.C. Frunzete,

EEG Signal Processing for Epileptic Seizure

Detection, E-Health and Bioengineering

Conference (EHB), Bucharest, Romania,

2023, unpublished.

[18] R. Hopfengärtner, B.S. Kasper, W. Graf, S.

Gollwitzer, G. Kreiselmeyer, H. Stefan, H.

Hamer, Automatic Seizure Detection in

Long-Term Scalp EEG Using an Adaptive

Thresholding Technique: A Validation Study

for Clinical Routine, Clinical

Neurophysiology, Vol.125, No.7, 2014, pp.

1346-1352, doi:

10.1016/j.clinph.2013.12.104.

[19] S.S. Alam, M.I.H. Bhuiyan, Detection of

Seizure and Epilepsy Using Higher Order

Statistics in the EMD Domain, IEEE journal

of biomedical and health informatics,

Vol.17, No.2, 2013, pp.312-318, doi:

10.1109/JBHI.2012.2237409.

[20] F. Kalin, T.C. Akinci, D. Türkpence, S.

Seker, U. Korkmaz, Detection of Epileptic

Seizure Using STFT and Statistical Analysis,

Advances in Neural Signal Processing.

IntechOpen, 2020, doi:

10.5772/intechopen.89026.

Contribution of Individual Authors to the

Creation of a Scientific Article (Ghostwriting

Policy)

The authors equally contributed in the present

research, at all stages from the formulation of the

problem to the final findings and solution.

Sources of Funding for Research Presented in a

Scientific Article or Scientific Article Itself

This work was supported by a grant

from the National Program for Research of the

National Association of Technical Universities -

GNAC ARUT 2023.

Conflict of Interest

The authors have no conflicts of interest to declare.

Creative Commons Attribution License 4.0

(Attribution 4.0 International, CC BY 4.0)

This article is published under the terms of the

Creative Commons Attribution License 4.0

https://creativecommons.org/licenses/by/4.0/deed.en

_US

WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE

DOI: 10.37394/23208.2024.21.34

Bianca-Alexandra Zîrnă, Mădălin Corneliu Frunzete

E-ISSN: 2224-2902

344

Volume 21, 2024