Page 2150

www.rsisinternational.org

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026

Using ICA Neuroimaging Techniques to Detect Neurological

Abnormalities.

Smt. Preethi.Warrier

M.E .Electronics(Dig Systems) Shah and Anchor Kutchhi Engineering College

DOI: https://doi.org/10.51583/IJLTEMAS.2026.150600154

Received: 28 June 2026; Accepted: 07 July 2026; Published: 18 July 2026

ABSTRACT

EEG (Electroencephalogram) measures electrical potentials on the scalp surface that occur as a result of dynamic

brain function. This paper presents the method to extract EEG Data and various methods used to perform EEG

Neuroanalysis, and why the author has chosen to work on ICA method. EEG data is collected by placing sensors

on a person’s scalp at different positions, amplifying data by 100db, digitizing it with a sampling rate of 256HZ

and using BLUETOOTH Technology to wirelessly transmit it to the computer. EEGLAB, a C library and tool

integrated in MATLAB, has been described. The methodology used to obtain the EEG datasets and implement

both ICA methods on the data has been presented. Baseline wander and power line interference were also

removed from baseline wander and power line interference from the EEG data. The time taken for ICA and

FAST ICA has been compared. Finally, neurological abnormalities were detected abnormalities in brain, using

ICA, by detecting spikes and sharp waves in EEG.

Keywords:

EEGLAB, ICA, Artifacts, Whitening, Sphering

INTRODUCTION

EEG (Electroencephalography) 1 measures electrical potentials on the scalp surface that occur as a result

of dynamic brain function. The procedure involves placing multiple sensors on the scalp. These sensors

have the ability to measure potential changes with microvolt sensitivity.

One challenge of using scalp-based data is that each sensor is actually measuring surface potential changes

caused by a superposition of underlying signals from various sources within the brain, as well as extra

brain sources. These signals are transmitted to the scalp via volume conduction through various tissue and

bone structures. Each sensor is actually receiving a mixture of different signals, and a given EEG signal is

obtained from sensors, and at different intensities.

The hardware used, does introduce some power line interference in the EEG waveforms. Also, one neuron

in brain does not influence 1 sensor, but many sensors at a time, so channel data are dependent on each

other. Infomax and Fast ICA, mathematically remove these problems.

The signal mixtures at each sensor can be separated into several independent components. Some of these

will correspond to artifacts such as eye blinks, heartbeats and in some cases, the experimental apparatus.

Independent Component Analysis (ICA) is a mathematical technique for extracting components, where

the extracted components describe temporally independent activities from spatially fixed overlapping

sources. Sources corresponding to artifacts can then be removed from the signal mixtures to facilitate

further analysis of the EEG data.

The big advantage of using ICA over other methods is that, it helps in parallelizing the analysis, that is,

individually, neuroanalysts can analyse individual channel, as each channel is independent of each other.

By combining the high-resolution data provided by dense- array EEG with sophisticated data analysis

techniques such as ICA, researchers are developing ways to ‘see’ into the neurophysiological and

Page 2151

www.rsisinternational.org

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026

cognitive processes within the brain. Section 2 describes the methods used to extract the EEG signal s

from the brain and display EEG on computer. Component Analysis. Section 3 shows a table of 22

channels. Section 4 shows the block diagram and work done by the author. Section 6 describes the

software to be used. Section 7 describes what are the signs of brain abnormalities, and how they are

detected using ICA methods. Section 8 describes the applications and future scope.

EEG data extraction

In conventional scalp EEG, the recording is obtained by placing electrodes on the scalp with a conductive

gel or paste, usually after preparing the scalp area by light abrasion to reduce impedance due to dead skin

cells.The sensor is known as ENOBIO.Some systems use caps or nets into which electrodes are

embedded; this is particularly common when high-density arrays of electrodes are needed.Each electrode

is connected to one input of a differential amplifier (one amplifier per pair of electrodes); a common

system reference electrode is connected to the other input of each differential amplifier.

The amplification is about 1,000–100,000 times, or 60–100 dB of voltage gain. Most EEG systems ever,

are digital, and the amplified signal is digitized via an analog-to-digital converter, after being passed

through an anti-aliasing filter. Analog-to-digital sampling typically occurs at 256-512 HZ in clinical scalp

EEG; TI-MSP 430 is the processor used for digital conversion and filtering. The digital EEG signal is stored

electronically and can be filtered for display. Typical settings for the high-pass filter and a low-pass filter

are 0.5-1 HZ and 35–70 HZ, respectively. The high-pass filter typically filters out slow artifact, such as

electrogalvanic signals and movement artifact, whereas the low-pass filter filters out high- frequency

artifacts, such as electromyographic signals. An additional notch filter is typically used to remove artifact

caused by electrical power lines (60 HZ in the United States .This EEG data is then transmitted wirelessly

to a computer,using BLUETOOTH Technology. A typical adult human EEG signal is about 10µ V to 100

µV in amplitude when measured from the scalp and is about 10– 20 mV when measured on computer.

Fig.1.Figure shows the extraction and Wireless transmission of EEG

ICA (Independent Component Analysis)

Initially, EEG Neuroimaging was performed using clustering methods. Now, clustering could not handle

large datasets, and also, the individual channel data was dependant on each other.

Cross Validation method, divided he data into 2 subsets, and the 2 subsets were independent of each other,

but this method too was very time taking.

Independent Component Analysis (ICA)[2] is a mathematical technique for extracting components, where

the extracted components describe temporally independent activities from spatially fixed overlapping

sources. Sources corresponding to artifacts can then be removed from the signal mixtures to facilitate

further analysis of the EEG data.

One form of ICA is encapsulated in the both ICA methods, first described in and implemented in

MATLAB. FastICA works by searching for a weight matrix that maximizes the non-gaussianity of the

resultant components[4]. Any mixture of non-gaussian random variables will be more gaussian than the

variables themselves (Central Limit Theorem of Statistics). Therefore, it is possible to use non-gaussianity

as a measure of statistical independence.

Page 2152

www.rsisinternational.org

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026

The phases of execution of FAST ICA are:

Start processes on head and remote nodes

Load parameters and replicate to workers

Load data and distribute amongst workers

Optionally whiten the data

Load parameters and replicate to workers

Output components, weights, sphering matrix

Channels Used in ICA

For the first study, on the applicability of ICA on electroencephalogram, the EEG consists of 22 signals,

recorded by unipolar electrodes at a sampling rate of 250Hz[5].

For the second part of the study the polysomnographic data used in the study were collected at the

Asklepios Clinic, Gauting, Munich, Germany. The sampling rates of the channels are presented below.

Signal

Unit

Sampling

Freq.

EEG --- C4/A1

µV

125 Hz

EEG --- C3/A2

µV

125 Hz

EOG RIGHT

F4/A1

µV

125 Hz

BODYPOS

HANDY F3/A2

125 Hz

EMG CHIN P4/P3

µV

250 Hz

ECG --- Ecg

µV

125 Hz

RESP FLOW Fp2/Cz

25 Hz

RESP FLOW Fp1/Cz

25 Hz

RESP FLOW Fz/Cz

25 Hz

RESP THORAX Thrx

µV

25 Hz

RESP ABDOMEN Abdm

µV

25 Hz

SOUND RAW Snd

µV

250 Hz

SAO2 SCHWARZ Oxi

25 Hz

EMG TIBIALR T4/T6

µV

250 Hz

EMG TIBIALL T3/T5

µV

250 Hz

BODYPOS SCHWARZ Pos

25 Hz

CPAP SCHWARZ Prss

mBa

25 Hz

TABLE 1: 22 Eeg Channels

Page 2153

www.rsisinternational.org

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026

Work Done

Proposed Block Diagram

Fig.2.Block Diagram

Objective of Work

The objective of work is to collect EEG reports and use Infomax and Fast ICA methods for analyzing a

large set of EEG data samples and separate most mixed components that explain the observed data. The

reduction in execution time provided by Fast ICA was one of the primary motivators for this research.

Also, raw EEG data, which looks similar for all persons, is analyzed using both methods, and a better

output is obtained which shows abnormalities clearly.

Proposed Experimental Work

The tasks of ICA are:

Obtain input data, in the form of a matrix x.

Search for a weight matrix W such that will unmix the whitened signal mixtures x into a set of

independent source signals s.

Compute covariance matrix, sphering matrix, to make data uncorrelated with each other.

Apply the transfer Functions to an Inverse system, to obtain the reconstructed sensor signals.

Infomax uses Linear Transformation methods, where the direction of data vectors, and the angle between

the previous and new weights, is taken for calculation. Fast ICA is a fixed point algorithm, which involves

simple matrix operations, thus making it faster than infomax. Depending upon the configurations of the

system, Fast Ica takes less than half the time taken by Fast ICA.

The execution time of both methods was measured after implementing them, and proved that, Fast ICA is faster.

But, in some cases, after observing the waveforms produced by both methods, Infomax output looks more

accurate than Fast ICA.

Page 2154

www.rsisinternational.org

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026

Software Design

MATLAB programs were developed for both Infomax and Fast ICA , using commands in MATLAB. Using

tic and toc commands, the time taken for analyzing has been calculated. Also, the codes for removing baseline

wander and power line interference, have been developed in MATLAB. Using MATLAB plot and subplot

commands, individual patient’s data can be observed. 325 patients’ data is uploaded on MATLAB directory.

EEGLAB[3] is a Matlab toolbox distributed under the free GNU GPL license for processing data from

electroencephalography(EEG), magnetoencephalography MEG and other electrophysiological device.

Along with all the basic processing tools, EEGLAB implements independent component analysis (ICA),

time/frequency analysis, artifact rejection, and several modes of data visualization. EEGLAB allows

users to import their electrophysiological data in about 20 binary file format from standard EEG

manufacturer, preprocess it, visualize activity in single trials, and perform ICA. Artifactual ICA

components may be subtracted from the data. Alternatively, ICA components representing brain activity

may be further processed and analyzed. EEGLAB also allow users to group data from several subjects,

and to cluster their independent components. EEGLAB may be downloaded at

http://sccn.ucsd.edu/eeglab.

MATLAB programs were developed for both Infomax and Fast ICA , using commands in MATLAB.

Detection of Abnormalities.

ICA extracts the independent components from the data and makes it artifact free. Soon, the raw data,

which looks similar for most persons, looks accurate and proper for each patient.

Spikes and Sharp waves are the main indications of any abnormality. The presence of spikes indicates

the presence of idiopathic (primary) epilepsy in a patient. The presence of sharp waves indicates that the

patient is suffering from Symptomatic epilepsy, which is not genetic.

Spikes are narrow waves with high amplitude. Their sudden occurrence in an otherwise smooth EEG

waveform indicates that the person is abnormal.

Sharp waves are broader waves, with negative amplitude; they also indicate that the person is suffering

from epilepsy.

Likewise, using the frequency of occurrence of spikes and sharp waves, a person’s neurological condition

can be detected.

Applications

ICA is a mathematical technique for extracting independent components. Sources corresponding to

artifacts are then removed from the signal mixtures to facilitate further analysis of the EEG data. So, the

ways to observe the neurophysiological processes within the brain become simple because of ICA.

Fast ICA provides a significant advantage over the existing MATLAB-based algorithms in two ways:

execution time and data size. The reduction in wall-clock time provided by ICA was one of the primary

motivators for the ICA research. However, an equally significant feature of ICA is its ability to handle

much larger data sets than MATLAB can handle.

Work Done and Results

EEG data from 325 patients were collected from [6].This has been loaded on MATLAB directory, for further

processing. The data consists of 2500 columns, that is, the number of samples. Also,

The data were subdivided the entire data into 6 subplots, for the important 6 channels, and represented them.

Page 2155

www.rsisinternational.org

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026

Each patient’s independent data can be observed. Also, baseline wander and power line interference has been

removed.

The execution time of both algorithms was measured. The average execution time of Infomax ICA was 2.18 s,

whereas FastICA required 0.53 s, corresponding to a computational speedup of approximately 4.1×. FastICA

therefore provided substantially faster execution while preserving effective component separation. The

processed EEG signals were further analyzed to identify epileptic abnormalities based on spikes and sharp

waves.

Also, the author has worked on detecting the type of epilepsy that some patients are suffering from, by

observing the pattern of spikes and sharp waves.

CONCLUSION

The ICA class of techniques and parallel implementations of FAST ICA Algorithm and INFOMAX has

been described by the author. Indeed, one active area of neuroscience research is the comparison of

different ICA algorithms as applied to EEG. Different algorithms are appropriate for different assumptions

about the EEG data and the underlying signals.

INFOMAX uses fractional values for the whitening and sphering matrices. FAST ICA uses only decimal

values, which makes computations easy, and less time taking, but the resolution is little reduced.

It is being planned to increase the usability of HiPerSAT by increasing the performance of the existing

algorithms, and by implementing additional ICA methods.

Also, to make online extraction of EEG possible, the author are working on coding the programs in TURBO

C, C++

REFERENCES

1. Kohavi, R. (1985). A study of cross-validation and bootstrap for accuracy estimation and model

selection. In Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence (pp.

1137–1143).

2. Ungureanu, M., Bigan, C., Strungaru, R., & Lazarescu, V. Independent Component Analysis Applied in

EEGLAB. (Publication details not provided.)

3. EEGLAB Homepage. (2005).

4. Keith, D. B., Hoge, C. C., Frank, R. M., & Malony, A. D. (2006). ICA Methods for EEG Neuroimaging.

Neuroinformatics Center, University of Oregon, Eugene, Oregon, USA. In Proceedings of the 20th IEEE

International Parallel and Distributed Processing Symposium (IPDPS 2006), April 25–29, 2006.

5. Xu, L., Cheung, C. C., Yang, H. H., & Amari, S. (2006). Independent Component Analysis by the

Information-Theoretic Approach with Mixture of Densities. IEEE Transactions on Neural Networks, 5,

1821–1826.

6. EEGMIT. www.eegmit.com

7. eMed Journals. www.emedjournals.com

8. Epilepsy Foundation. www.epilepsyfoundation.org

9. Neuroinformatics Centre. NIC-Transaction-01.

10. Neuroinformatics Centre. NIC-Transaction-17.