Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

JETIR2405449 Published

Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

© 2024 JETIR May 2024, Volume 11, Issue 5 www.jetir.

org (ISSN-2349-5162)

CLASSIFICATION OF EEG SIGNALS USING


DEEP LEARNING TECHNIQUE
H K Shreedhar, Abhishek Gowda K,
Abhishek, Ashwini P
Global Academy of Technology, Bengaluru - 560098

Abstract: When the brain experiences repeated seizures, epilepsy results. The human brain is harmed by frequent epileptic
seizures, which can lead to memory loss, mental illness, and other problems. The electroencephalogram (EEG) test is a significant
tool for learning about brain activity and for diagnosing neurological disorders, such as epilepsy. An automated seizure detection
approach has been effectively introduced in this study. The investigation uses LSTM, CNN and KNN to analyze the EEG signals
from an online database for binary categorization. Seizures are detected using two of the five sets from set A (normal) to set E
(abnormal). Using MATLAB Tool, the simulations are run with a range of activation functions, optimizers, and loss models to
assess the performance. Future suggestions will be formulated with the assistance of the proposed study, which will also assist
academics in discovering how deep learning algorithms are implemented for the classification of EEG signals.

I. INTRODUCTION:

The cerebral cortex of the human brain has a remarkable and rich spatiotemporal dynamics that is specific to humans. In the brain,
chemical and electrical impulses are exchanged between millions of neurons. Seizures are abnormal electrical disturbances that
occur in the brain. When the brain experiences repeated seizures within short duration of time, epilepsy results. Proper examination
of the electroencephalogram (EEG) can reveal important details about brain activity and be helpful in the diagnosis of brain
disorders, particularly epilepsy. An abnormal brain condition can cause an EEG to show unusual electrical discharge. Meaningful
communication is provided by measuring the frontal pole (Fp), frontal (F), parietal (P), temporal (T), and occipital (O) regions of
the brain. Treating epileptic patients well is more important than diagnosing them accurately, so future research must focus on
automated seizure detection. Machine learning is a key component of expert systems, enabling the creation of systems that support
autonomous decision-making.
Just as the human brain learns via information and study, so too can a machine learn from data. In two steps, machine learning
performs a specific task to classify the data. The machine is initially trained using pertinent features and the matching categorical
labels of the data in the training stage. In the testing stage, the machine is examined using unknown data that has similar features.
The unknown data must be classified by the classifier into the proper class. The percentage of test data that the classifier correctly
classifies is used to evaluate the classifier's performance. Effective use of machine learning algorithms such as LSTM, CNN, and
KNN allows for the making of judgment’s based on input.

II. METHODOLOGY

The current investigation used the EEGs of five patients, all of whom had successfully undergone resection of one of the
hippocampal formations, which allowed for the accurate identification of the epileptogenic zone. 100 single channel EEG segments,
each lasting 23.6 seconds, were arranged into five sets (designated A–E) for the study. Segments from surface EEG recordings,
made on five healthy participants utilizing a standardized electrode placement scheme, made up Sets A and B.The volunteers were
awake and calm, with their eyes open in (A) and closed in (B).Segments from the hippocampal formation of the opposing
hemisphere of the brain were recorded in set C, whereas segments from the epileptogenic zone were recorded in set D.

Set E solely included seizure activity, whereas sets C and D only included activity recorded during seizure-free intervals. Here, a
selection of segments.
Displaying ictal activity were made from all recording sites. After then, the recordings were put through a number of classification
methods.

JETIR2405449 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org e432
© 2024 JETIR May 2024, Volume 11, Issue 5 www.jetir.org (ISSN-2349-5162)

1) Discrete Wavelet Transform:


By breaking the signal down into approximation and detail information, discrete WT (DWT) with the fourth-order daubechies (db4)
and sixth-order daubechies (db6) wavelet functions analyzes the signal at different frequency bands, with varying resolutions. DWT
is employed in both preprocessing and decomposition of the EEG signals. The raw EEG has been cleaned up of noise by using
filters. For seizure detection, the noise-free signals are examined and trained.

2) Time Domain Analysis:


For all real numbers at different discrete times, or in the case of continuous time, the value of the signal or function is understood
in the time domain. Using time domain analysis, the six statistical features—mean, RMS, peak value, standard deviation, SNR, and
THD—are retrieved from raw EEG data in this study in order to identify distinctive patterns from the original data for consistent
classification.

3) frequency Domain Analysis:

The analysis of computations or signals in relation to frequency is known as frequency domain. About a range of frequencies, the
frequency domain shows the percentage of the signal that is present inside a specific frequency band. In order to obtain distinctive
patterns for reliable frequency domain analysis classification of EEG signals, four statistical features—Median frequency, Mean
frequency, Band power, and Power bandwidth—are extracted here at a sample rate of 256.

Fig 1: EEG Signals Classification

JETIR2405449 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org e433
© 2024 JETIR May 2024, Volume 11, Issue 5 www.jetir.org (ISSN-2349-5162)

III. FLOWCHART:

Fig 2: Flowchart of Classification

The dataset for this work is downloaded from Universitats Klinikumbonn[Ukb][1]. Throughout the entire experiment, 20% of the
data is designated as the test set and 80% of the data as the training set. A total of 50 training epochs allocated to the deep neural
network (DNN), LSTM, CNN, and KNN.A distinct type that has been employed in machine learning for categorization is feature
extraction, which lowers the dimension of raw EEG data. Six statistical characteristics are collected from the raw EEG data for
the project: mean, RMS, Peak value, standard deviation, SNR, and THD.A crucial phase in the classification process is feature
selection. To achieve the best outcome, it is recommended to feed the chosen features. Redundant features can overload the
system, which would prevent the best outcome from happening. As a result, reducing the number of features will aid in improving
the classifier's learning and performance. Deep Learning Algorithms have been used to train the system with the database and the
corresponding class labels during the learning stage. The system is tested using test data for categorization in the second stage.
The system's efficiency is assessed based on the proportion of test data that the classifier correctly classifies. Confusion matrix
with accuracy as a parameter is used to evaluate the performance of LSTM, BiLSTM, CNN, and KNN.

IV. RESULTS AND DISCUSSION

Here is an illustration of the study's experimentation, analysis, and satisfactory performance for automated seizure identification
using pertinent features as an input to compare and choose the best model among the LSTM, CNN, Bi-LSTM, and KNN that
classifies with high accuracy.
The classification results by choosing DWT with CNN is indicated in figure 3 and 4.

JETIR2405449 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org e434
© 2024 JETIR May 2024, Volume 11, Issue 5 www.jetir.org (ISSN-2349-5162)

DWT with CNN


No. of Epochs (M)=50

Fig 3: Training Plot

Fig 4: Confusion Matrix

JETIR2405449 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org e435
© 2024 JETIR May 2024, Volume 11, Issue 5 www.jetir.org (ISSN-2349-5162)

TIME DOMAIN ANALYSIS: Accuracy of classification in time domain with different models is summarized in table 1.

Table 1. Time Domain Analysis with CNN, LSTM, KNN and Bi LSTM models

Classes Accuracy With Accuracy Accuracy Accuracy


CNN With With KNN With Bi LSTM
LSTM

A_E 100% 100% 100% 100%

B_E 97.5% 95% 97.5 % 97.5 %


C_E
96.0% 100% 97.5% 97.5%
D_E
97.6% 97.5% 100% 100%
AB_E
96.2% 97.5% 100% 95.0%
CD_E
100% 100% 97.0%
95.0%
ABC_E
95% 100% 99% 98%

ABCD_E 95.6% 92.23% 79% 83%

FREQUENCY DOMAIN ANALYSIS: Accuracy of classification in frequency domain with different models is summarized
in table 2.

Table 2. Frequency Domain Analysis with CNN, LSTM, KNN and Bi LSTM models

Accuracy Accuracy Accuracy


Classes Accuracy With With KNN With Bi LSTM
With CNN LSTM

A_E 97.5% 62.5% 92.5% 92.5%

B_E 92.5% 80.0% 87.5% 87.5%

C_E 92.5% 92.5% 97.5% 97.5%

D_E 90% 85.0% 95.0 % 95.0 %

AB_E 100% 97.5% 92.5% 92.5%


CD_E
90% 90% 100% 100%

AB _CD_E 80% 74% 86% 86%

ABCD_E 95% 78% 96.0% 96.0%

AC_E 81% 73% 81.7% 81.7%

DISCRETE WAVELET TRANSFORM: Classification accuracy by using DWT and different models is summarized in table 3.
Table 3. Analysis by using DWT, with CNN, LSTM, KNN and Bi LSTM models
Accuracy Accuracy Accuracy Accuracy
Classes DWT With With With With Bi
CNN LSTM KNN LSTM
db_4 95% 97.5% 95% 97.5%
A_E db_6 95% 97.5% 95% 97.5%
db_4 95% 100% 95% 90%
B_E db_6 96.0% 92.5% 96.0% 90.0%
db_4 97.5 95.5% 97.5 95.0%
C_E db_6 92.5% 97.5% 92.5% 95.0%
db_4 95% 95% 95% 97.5%
D_E db_6 97.5% 90.5% 97.5% 92.5%
db_4 95% 97.5% 95% 97.5%
AB_E db_6 92.5% 97.5% 92.5% 97.5%
db_4 95% 95% 95% 95%
CD_E db_6 95% 95% 95% 95%

JETIR2405449 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org e436
© 2024 JETIR May 2024, Volume 11, Issue 5 www.jetir.org (ISSN-2349-5162)

CONCLUSION:
In the field of EEG signal classification, deep learning algorithms have shown to yield promising results. These models can be
trained to generate an automated seizure detection mechanism with labeled sensor data, and they are highly valuable in fields like
neuromarketing, clinical and psychiatric studies, and brain-computer interfaces (BCIs).

REFERENCES:
[1]Indications of nonlinear deterministic and finite-dimensional structures in time series of brain electrical activity: Dependence on
recording region and brain state by Ralph G. Andrzejak, Klaus Lehnertz, Florian Mormann, Christoph Rieke, Peter David, and
Christian E. Elger published on 20 November 2001. DOI:10.1103/PhysRevE.64.061907

[2] Kusumika Krori Dutta Department, of Electrical and Electronics Engineering, Ramaiah Institute of Technology, Bangalore,
Karnataka, INDIA. Published in 2019 9th International Conference on Cloud Computing, Data Science & Engineering,
DOI 10.1109/CONFLUENCE.2019.8776889

[3] EEG signal classification using LSTM and improved neural network algorithms by P. Nagabushanam, S. Thomas George, and
S. Radha published on 13 November 2001, Doi.org/10.3390/e18090272.

[4]A One-Dimensional CNN-LSTM Model for Epileptic Seizure Recognition Using EEG Signal Analysis by Gaowei Tianhe Ren,
Yu Chen, and Wenliang Che

[5]A study of deep learning approach for the classification of electroencephalogram (EEG) brain signals Author links open overlay
panel Dharmendra Pathak, Ramgopal Kashyap, Surendra Rahamatkar. Doi.org/10.1016/B978-0-12-824054-0.00009-5

[6]A comparative analysis of signal processing and classification methods for different applications based on EEG signals by
Ashima Khosla, Padmavati Khandnor, Trilok Chand, Computer Science and Engineering Department, Punjab Engineering College
published on 21 February 2020.

[7] Status of deep learning for EEG-based brain-computer interface applications by Khondoker Murad Hossain, Md. Ariful Islam
Shahera Hossain, Anton Nijholt, and Md Atiqur Rahman Ahad published on 16 January 2020,
Doi.org/10.3389/fncom.2022.1006763.

[8] Long Short-Term Memory deep learning network for the prediction of epileptic seizures using EEG signals by Κostas Μ.
Tsiouris a b, Vasileios C. Pezoulas b, Michalis Zervakis c, Spiros Konitsiotis d, Dimitrios D. Koutsouris a, Dimitrios I. Fotiadis be
published in Volume 99, 1 August 2018, Doi.org/10.1016/j.compbiomed.2018.05.019.

[9]Supervised Machine Learning and Deep Learning Techniques for Epileptic Seizure Recognition Using EEG Signals by
Mohamed Sami Nafea 1,2ORCID andZool Hilmi Ismail 2, ORCID Published in 8 December 2022,
Doi.org/10.3390/bioengineering9120781.

JETIR2405449 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org e437

You might also like