Proceedings of the Fourth International Conference on Smart Systems and Inventive Technology (ICSSIT-2022)

IEEE Xplore Part Number: CFP22P17-ART; ISBN: 978-1-6654-0118-0

A Review on EEG based Epileptic Seizures

Detection using Deep Learning Techniques
Srikanth Cherukuvada Kayalvizhi R
Department of Networking and Communications, Department of Networking and Communications,
School of Computing, SRM Institute of Science and Technology School of Computing, SRM Institute of Science and Technology
Kattankulathur,Chennai, India Kattankulathur,Chennai, India
[email protected] [email protected]

Abstract—Magnetic Resonance Imaging (MRI) and [20], [21]. Figure 1 represents the different DL tools used
Electroencephalography (EEG) modalities have been used in for detecting seizures in epilepsy.
several screening procedures to diagnose epileptic seizures with
a high-level accuracy. O ver the last two decades; significant TensorFlow
contributions have been made to automate the procedure of 11
2022 4th International Conference on Smart Systems and Inventive Technology (ICSSIT) | 978-1-6654-0118-0/22/$31.00 ©2022 IEEE | DOI: 10.1109/ICSSIT53264.2022.9716374

22
EEG tests to diagnose epilepsy and seizures detection. Because 15 Keras
manual seizure detection will take more time for a doctor as 7
well as for technician to put their efforts into recognizing the Matlab
type of seizure within less time. Deep Learning (DL) helps in 7 Py Torch
automatic feature extraction and classification techniques. In
the Medical field, adopting these approaches has resulted in 14 Theano
the introduction of these approaches has led to a significant 14
breakthrough in substantial advancements including the
Octave
identification of epileptic seizures. This paper presents a
comprehensive summary of each section in depth and methods Fig. 1. Diagnosis of seizures in epilepsy using DL tools repeatedly in the
used to diagnose epilepsy from a variety of methods. In following reviewed papers.
addition, they are highlighted because they are best suited for
Since 2016, Convolutional Neural Network with
applied medicine, hardware execution, and cloud-based tasks.
This paper provides a review of novel seizure detection
Autoencoder (CNN-A E), Convolutional Neural Network-
applications and pre vious studies. The research concludes with
Recurrent Neural Network (CNN-RNN), Recurrent Neural
some fresh concepts for seizure detection utilizing DL Network (RNN), Convolutional Neural Network (CNN),
techniques which are becoming increasingly popular. Autoencoders (AE), and Deep belief network (DBN) are
Keywords—Classification, Deep Learning, Epileptic seizures, used for epilepsy detection [22], [23], [24], [25]. Figure 2
Electroencephalography, Feature Extraction, Magnetic shows the number of research done in identifying epileptic
Resonance Imaging. seizures applying several models of DL fro m 2014 to 2021.
Various DL models have been used to diagnose epileptic
I. INT RODUCT ION episodes, as shown in Figure 2.
Sudden attacks of seizures seem to be the most common
symptom of epilepsy, which would be a long-term Non- 5 22 2D-CNN
Co mmunicab le Disease (NCD) in hu mans [1]. Seizu res are 9
34
a type of epilepsy that affects the entire body and is caused 1D-CNN
by a quick and early disruption in the brain’s electrical 15 AE
activities [2]. Around 60 million people worldwide suffer
RNN
fro m various types of epileptic seizures [3]. Magnetic
resonance imag ing (MRI) [4], Magnetoencephalography 17
24 CNN-RNN
(MEG) [5], EEG [6], and functional imaging techniques
such as Positron emission tomography (PET) [7], Single - Fig. 2. Reviewed works carried out using several DL models from 2014 to
Photon Emission Computerized To mography (SPECT) [8], till2021.
Electrocorticography (ECo G) [9], are some of the screening
techniques used to diagnose epileptic seizu res. EEG signals 3
9 IEEE
are extensively used because they are inexpensive, portable,
Elsevier
and exhib it distinct frequency domain patterns [10]. EEG 13 38
signals provide the variation in supply voltage passes Springer
through neurons using cell memb rane in the brain, Others
representing that bio signaling takes place over here [11]. 14 Preprint
There’s also the chance of hu man mistakes, thus
establishing a computer-based diagnostic might help solve 16 Frontiers
these issues. To overcome the challenges, many researchers
have used Artificial Intelligence (AI) techniques with MRI Fig. 3. Various research strategies are applied to epileptic seizure detection.
for epileptic seizure detection [12], [13]. Many machine One Dimensional-Convolutional Neural Network (1D-CNN)
learning methods have been created to identify epileptic and two Dimensional-Convolutional Neural Network (2D-
seizures utilizing statistical, nonlinear parameters, and CNN) are widely used deep learning techniques. 2D-CNN is
frequency domain [14], [15], [16], [17], [18], [19]. For little widely used by researchers to detect epilepsy. First EEG data
algorith ms may not perform as effectively as they formerly is converted into Two Dimensional (2D) images in the pre-
did as data availability has increased. As a result, DL processing stage by using Fourier-related transform which is
techniques, which are contemporary procedures, were used a Fourier spectrum on each shorter segment such as Short

978-1-6654-0118-0/22/$31.00 ©2022 IEEE 966

Authorized licensed use limited to: NORTHWESTERN POLYTECHNICAL UNIVERSITY. Downloaded on May 22,2023 at 05:02:23 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Fourth International Conference on Smart Systems and Inventive Technology (ICSSIT-2022)
IEEE Xplore Part Number: CFP22P17-ART; ISBN: 978-1-6654-0118-0

Time Fourier Transform (STFT). By utilizing the converted episodes to detect aberrant patterns. The centrotemporal
images into a 2D-CNNN network is created. The second spikes and their properties were defined by Juozapaviciuset
group includes various 1D-CNN models, which are also used al. [39] to detect abnormalities in EEG signals. For autis m
for detecting epileptic seizures. Figure 3 depicts the various spectrum disease; Cociu et al. [40] outlined a mu ltimodal
ways of detecting epileptic seizures. The IEEE citation functional and structural brain connection strategy. Qian et
database appears to have the most widely recognized articles. al. [41] created a drowsiness monitoring system to reduce a
The major objectives of this research comprise of the large number of traffic accidents by analyzing EEG signal
following: records to predict tiredness levels. Yang et al. [42]
developed a hierarchical network structure containing
 Available datasets related to EEG are provided. subnetwork nodes that might be used to distinguish between
 Based on EEG, various DL models are d iscussed for three human emotions: positive, neutral, and negative. Costa
studying the epileptic seizures detection with the help et al. [43] investigated a trustworthy lead field mat rix to
of various modality signals;
overcome the Magnetoencephalography (MEG) source
 Advancing forthcoming challenges on EEG in localization problem. Giri et al. [44] used an 1D-CNN to
epilepsy to detect seizures.
identify ischemic stroke based on EEG and
 Investigation of the best-performing model in DL for
Electrooculogram (EOG).
several modality signals of data.
In section II, a Literature survey is performed on various DL III. SEIZURE DET ECT ION UT ILIZING DL TECHNIQUES BASED
techniques used in detecting epileptic seizures. Section III, ON CAD SYST EM
exp lains how a Co mputer-Aided Design (CAD) system is The DL model may emp loy EEG, Functional Near-
used in detecting epileptic seizures. In section IV, explains Infrared Spectroscopy-scan (fNIRS-scan), M EG-test, PET-
the design of the hardware approach applied to seizures scan, SPECT-scan, Electrocorticography, and MRI-scan as
detection using epilepsy. Section V shows that various inputs. Figure 4 describes the diagnosis of seizures using the
challenges that occurred during the specific process of CAD method using DL techniques. summarizing all the
epileptic seizure detection. Finally, Section VI represents the works accomplished carried out the epileptic seizure
conclusion. detection using DL methods are discussed in table 6.
II. LIT ERAT URE REVIEW A. Epileptic Datasets
Le et al. [26] According to the author, the most efficient The creation of an accurate and trustworthy CADS relies
model is the DBN, which builds the framework for deep heavily on datasets. Bern-Barcelona, Zenodo, CHB-MIT,
learning to detect seizures in the brain using EEG data. Yao Flint-Hills, Hauz Khas, Kaggle, Freiburg, and Bonn are only
et al. [27] EEG are evaluated and analyzed by pro fessional a few of the EEG datasets which recognize and distinguish
neurologists in contemporary clinical procedures to provide the seizures automatically. Several datasets that were utilized
support for medicinal reco mmendations. Iesmantas et al. to detect epileptic episodes using DL methods are
[28] provide a method for detecting and classifying seizures represented in Figure 5.
based on medical data fro m EEG tests and a trained B. EEG Signal Preprocessing
convolutional neural network through its brain
In the DL-based CAD approach, Electroencephalogram
synchronization in addition to the power spectrum of t ime -
signal pre-treatment for deep learning network applications
series properties. Liang et al. [29] constructed a model
termed an 18-layer Long-Term Recurrent Convolutional generally consists of three steps: Reduction of Noisy data,
normalizing the signals, and preparing the signal for giving
Network (LRCN) to recognize and localize seizure zones
input to the system. During this step reduction of
automatically. Tho mas et al. [30] presented an epileptic
EEG classificat ion method based on Interictal Ep ileptiform unnecessary noise in brain signals such as EEG. Finally,
various time-frequency approaches such as tempora l
Discharges (IED) detection. Wang et al. [31] emp loyed
frequency methods are utilized to imp rove signals in each
mu lticlass Support Vector Machines (SVM) to detect
epileptic and epilepsy seizures using EEG recordings. Jiang rapid classification of Deep Neural Networks (DNN).
et al. [32] exp lore ep ileptic seizure classification using C. Different Deep Learning Techniques
learning algorith ms and a fuzzy system. Abualsaudet al. [33]
discusses how the decomposition levels imp lement the CNN’s: In DL, Convolutional neural networks are one of the
development of mult iple wavelets. Tsiouris et al. [34] most prominent network models, accounting for the bulk of
describes how to reduce the effect and time which are machine learning research [22]. They were originally
required to process a huge volume of EEG info rmation. O designed for image processing but are now being utilized in
Solaija et al. [35] show how Dynamic mode deco mposition one and two-dimensional structures to identify and forecast
(DM D) is mostly utilized for fluid flow analysis. Saidi et al. disease using biological signals [51]. The identification of
[36] show how wavelet analysis in EEG signals has epileptic seizures using EEG data is a common application
specified sub-bands frequency for wavelet filters using for this type of deep learning network.
Discrete Wavelet Transform (DWT). Yuan et al. [37] 2D –CNN: Nowadays, deep 2D networks such as X- ray
discuss the use of encodes for feature extraction to learn [52], [53], tests of CT and M RI modalities [27] are utilized
channel-specific and global viewpoints. Fan et al. [38] in different therapeutic applications. Image
discuss the synchronization of EEG signals for epileptic

978-1-6654-0118-0/22/$31.00 ©2022 IEEE 967

Authorized licensed use limited to: NORTHWESTERN POLYTECHNICAL UNIVERSITY. Downloaded on May 22,2023 at 05:02:23 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Fourth International Conference on Smart Systems and Inventive Technology (ICSSIT-2022)
IEEE Xplore Part Number: CFP22P17-ART; ISBN: 978-1-6654-0118-0

T ABLE I. LIST OF P OP ULAR DATASETS UTILIZED FOR DETECTION OF EP ILEP TIC EP ISODES USING EEG RECORDINGS.
Se iz ure Subject Duration Frequency Types of
Datase ts Description
Count count (hours) (Hz) Recordings
Intracranial The database contains 21 patients’ invasive long-term EEG
Freiburg
87 21 708 256 Electroencephal recordings and has a sampling rate of 256 Hz during invasive
[45]
ography (IEEG) process.
Continues
Flint-Hills The dataset consists of continues electrocardiography signals
59 10 1419 249 intracranially
[17] with whole period of 1419 h and sampling rate of 249 Hz.
term ECoG
In Kaggle dataset, considering the intracranial EEG signals
Kaggle
48 2 627 400 IEEG from two subjects with count 48 as seizures and 627 h whole
[46]
period.
The CHB-MIT corpus contains duration of 844 hours related
CHB-MIT
163 22 844 256 Seeg to continuous scalp EEG signal recording, and also collected
[47]
seizure data of 163 from 23 paediatric subjects.
Bonn dataset contains five files, A, B, C, D, and E, each
Surface and
Bonn [48] NA 10 39 173.61 consists of 100 single channel EEG recordings with duration
IEEG
of 23.6 sec, sampling rate of 173.61 Hz.
Barcelona Each dataset includes 3750 pairs of recorded signals with
3750 5 83 512 IEEG
[49] duration of 20-sec, and sampling rate of 512-Hz.
The database describes about scalp EEG recordings from ten
Hauz Khas Scalp EEG
NA 10 NA 200 subjects, recorded from AS40 system with sampling
[17] (sEEG)
frequency at 200 Hz.
This dataset contains EEG data which are interpreted by three
Zenodo
460 79 74 256 Seeg clinicians, and each clinician is dealt with 460 seizures and
[50]
seizure free are 22, 39 neonates.
b) Visual Geometry Group(VGGNet): In 2014, a
group of Oxford researchers presented the VGG CNN
model [62]. VGC-16 was the name given to the model since
it had 16 layers.
c) GOOGLENET: GoogLeNet in 2014 won the
challenge with the top five of ImageNet benchmark with
93.3 percent accuracy, [64]. The variant of the inception
Network is a 22-layer network represented as GoogLeNet
d) ResNet (RESNET): Microsoft’s ResNet achieved
the competition using the ImageNet model with 96.4 percent
precision by employing a deep network consisting of 152-
layer with a ResNet model [65]. Two ResNet and DenseNet
designs were effectively employed by [66] to diagnose
epileptic episodes.
1D-CNN’S: 1D-CNNs are well-suited to processing
biological data like EEG for seizure detection.

D. Recurrent Neural Networks (RNNs)

Text, signals, and videos are examp les of sequential data
with varying and lengthy durations, rendering simp le DL
techniques appropriate [72].
a) Long Short-Term Memory (LSTM) Networks: The
Fig. 4. Seizure Detection using DL techniques based on the CAD system. vanishing gradient problem is another disadvantage of RNN
segmentation [54], medical image classification [55], and [25], [23], [22], [24], As it back-propagates, the issue occurs
facial recognition [56] are just a few of the computer vision due to the decreasing of gradients. LSTM gates were
issues that deep 2D networks are being utilized to solve developed to tackle the short- memory problem [22].
proposed a DL based model known as SeizNet that operates Information flow may be controlled using gates.
convolution layers with dense relationships. [58] introduced b) Gated Recurrent Unit (GRU): GRU [24], [23],
a novel DL model with five arch itectures of 14, 18, 22, 23, [25], [22] is an LSTM variation that merges the input and
and 26 layers, the Temporal Graph Convolutional network forgets gates into a single update gate. The number of gating
(TGCN). signals has been reduced to two. [73] co mb ined a GRU five-
a) ALEXNET: ImageNet [60] was established by Fei layer network with a neural network-based classifier called a
Li, a Stanford University professor, who generated a SoftMax classifier to get excellent output. Table II provides
an overview of the research conducted in this area.
collection of annotated images of real-world items. Alex
Krizhevsky’s algorithm, A lexNet, revolutionized image E. (AE): STANDARD AUTOENCODERS
categorization by winning the competition of ImageNet and An unsupervised neural network deep network has the same
come out fro m the deep learn ing stage in 2012 [61]. input and output [23], [25], [24], [22]. Before being utilized
AlexNet took first place in the co mpetit ion, with an
accuracy of 84.6 percent in the top five tests.

978-1-6654-0118-0/22/$31.00 ©2022 IEEE 968

Authorized licensed use limited to: NORTHWESTERN POLYTECHNICAL UNIVERSITY. Downloaded on May 22,2023 at 05:02:23 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Fourth International Conference on Smart Systems and Inventive Technology (ICSSIT-2022)
IEEE Xplore Part Number: CFP22P17-ART; ISBN: 978-1-6654-0118-0

T ABLE II. REVIEW OF ASSOCIATED WORKS MADE USING 2D CNNS MODEL .

Re fe re nce Me thods Layers Performance Classifier Merits Demerits

98.3(2-class),
[67] 2D-CNN 14 Softmax High classification accuracy No feature extraction
90.31(3-class)
From scalp EEG recordings, DNN can
[68] 2D-CNN 18 NA Softmax NA
accurately detect epileptic episodes.
DNN are used to learn meaningful
[69] 2D-CNN 9 86.31 Softmax representations automatically by using NA
this model.
Detection of epileptic episodes
High accuracy in certain datasets, due
and classification of EEG
[28] 2D-CNN 6 74 Softmax to clinical references and also
recording clinically is still a
respective algorithms.
challenging problem.
2D-CNN
Support of trained CNN models by
combining
[70] 12 NA Softmax using handheld devices and reports NA
with 1D-
streaming iEEG data
CNN
Experts usually performs the diagnosis Huge dataset is used, so that it
[59] 2D-CNN 8 99.48 Softmax of epileptic episodes on the findings of may not classify all signals of
EEG recordings. EEG accurately.
GoogleNet, Focal and non-focal epileptic seizures
Standard
[71] AlexNetLe 100 Softmax are utilized for signal classification NA
Networks
Net using EEG by applying DNN methods.
T ABLE III. SURVEY OF SIMILAR WORKS P ERFORMED UTILIZING DBNS MODEL .

Re fe rences Classifier Layers Performance Methods Merits

For detecting the seizures, a high resolution, multichannel EEG recordings
[74] LR 3 NA DBN
are taken and applied machine learning algorithms.
3 (hidden In this work, a Deep Belief Network (DBN) is presented to identify epileptic
[26] NA 96.87 DBN
layers) episodes using EEG

F. Deep Belief and Boltzmann Networks

Clinical (EEG)
Bonn Restricted Bolt zmann Machines (RBM) is a revised version
CHB-MIT of DBM [22], which is an undirected graphical model
2111111 TUH
42 Bern-Barcelona
(UGM). These are generative-adversarial deep learning
30
7
MRI Clinical) algorith ms with several layers of no minal and rando m
PET (Clinical) variables that are unsupervised and probabilistic [23], [22].
Freiburg Table III has further info rmation on DBN architecture fo r
22
UCI epileptic episodes.
24 FNIRS (Clinical)
Kaggle-MRI G. CNN - RNN
FMRI (Clinical) The CNN-RNN arch itecture is a very effective mix of deep
ECOG (Clinical)
neural networks. Detecting the spatial naturalness of nearby
Fig. 5. Various datasets were employed for detecting epileptic seizures in the patterns is improved by adding convolutional layers to
previous studies.
RNN, although RNN properties are better suited for time
to create the output, the input should be flattened into latent- series data processing. [75] In this work, in itiating with the
space representation. first article, many preprocessing schemas were used. The
T ABLE I V. REVIEW OF RELEVANT STUDIES CREATED BY AP P LYING RNNS MODEL .

Re ferences Me thods Performance Classifier No. of Layers Merits

In proposed method, examining a heterogeneity of hybrid DL
[75] LST M NA sigmoid 3,4
methods comprise of CNN and LS-T MN.
The high-level EEG representations are used effectively Inorder
[77] LST M 100 softmax 4
to distinguish among the seizure and non-seizure conditions.
Spatial features are extracted and applied to propose IndRNN
[27] 15-IndRNN 87 NA 48
technique and it is long neural network and deep also.
Classifying multichannel EEG signal recordings for detection of
[78] LST M 97.75 softmax 3
seizures utilized novel methods such as 2D-SDCAE.
LST M, The proposed method recognizes the segments of epileptic
[27] 84.35, 88.7 NA 4, 31
ADIndRNN-(3,3) seizures effectively.

Inception-V3 network was utilized by [76]. A summary of

important RNN-based research is presented in Table IV. IV. SURVIVALSYST EMS FOR EPILEPT IC SEIZURES
H. CNN – A Es: Yuan et al., 2018 [37] introduced a DET ECT ION
unique CNN-A E-based method in this part. Using the The initial research describes about detecting seizures in
SoftMax classifier, wh ich had an accuracy of 94.37 fast manner [86] by analy zing Brain-Co mputer Interface
percent. Table V contains general informat ion regard ing (BCI) system that was generated, utilizing the AE. In the
the CNN-AE studies examined. next study, [87] meant some pragmatic results for the

978-1-6654-0118-0/22/$31.00 ©2022 IEEE 969

Authorized licensed use limited to: NORTHWESTERN POLYTECHNICAL UNIVERSITY. Downloaded on May 22,2023 at 05:02:23 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Fourth International Conference on Smart Systems and Inventive Technology (ICSSIT-2022)
IEEE Xplore Part Number: CFP22P17-ART; ISBN: 978-1-6654-0118-0

T ABLE V. SUMMARY OF RELEVANT RESEARCH P ERFORMED RELATING CNN-AES MODEL .

Convolutional
Re ferences Classifier Performance Methods Merits
Layers
By using the CNN-AE, it recognizes the instances
[37] SoftMax 10 94.37 CNN-AE
rather than discriminating among classes.
For extracting the features in this framework, and
MLP/LST M/Bi LST M + 1D
[79] SoftMax 16 + 3/1/1 99.33/ 100 also reduces the dimensionality in it. operates
CNN-AE
directly on whole EEG recordings collected.
Different captured low dimensional features from recordings
[80] 15 92 CNN-AE
Classifiers of EEG by using epilepsy.

limited number of joint scalp-intracranial

[81] LR 8 66 CNN-ASAE
recordings

T ABLE VI. OUTLINE OF DL TECHNIQUES UTILISIZED FOR AUTOMATED DETECTION OF EP ILEP TIC EP ISODES.

Ref erence
Year Dataset Preprocessing Model Limitations
s
As future work, the use o f deep auto-encoder with various signal processing methods
[82] 2018 Bonn DWT SAE
may produce new results.
Down Sampling,
[30] 2018 Clinical 1D-CNN In future, planning to combine the rejection of artifacts, to overcome the false detections.
Filtering
In this overfitting of deep neural network takes place and also in future, intends to
[73] 2018 TUH EEG Distinct methods 1D-CNN-GRU
produce customized RNNs to undertake this problem.
IndRNN approach is more difficult and extensive. Finally requires more training samples
[27] 2019 CHB-MIT Windowing IndRNN
and is limited to processing imbalanced data.
CHB-MIT, This work focuses on what way it utilizes the dense IndRNN model with attention on
[83] 2019 Windowing ADIndRNN
Bonn seizure detection in real time.
Extraction, Working more hours on trained data regarding patient may not be always fe asible to use
[74] 2014 Clinical DBN
Standardization this as a model.
Augmentation, In this work, patient’ s frequency domain fe atures can be mor e biased than the feature
[84] 2019 Clinical Downsampling, SeizNet extraction fro m ti me domain and this model triggers false alar ms for the respective
Z-normalization subjects.
In 2d-CNN evaluating the neonatal subject recordings is the limitation and not
[85] 2018 Clinical Normalization 2D-CNN
represented for general scenario of EMU monitoring.
Down Sampling 2D-CNN
In this DL approaches were discussed which result in a prima ry shift o f clinical analysis
[68] 2018 Clinical Down Sampling NA
based on EEG recordings.
NA NA
TUH EEG The basic e rror modalities mentioned were false alar ms generated through the shorter
[75] 2017 Filtering method CNN-RNN
(Clinical) duration of the delta range, which results in irregular, slow patterns.
CHB-MIT
[80], 2018 Channel Selection CNN-AE Lower the feature dimensions will effects the performance of a model
Bonn
AlexNet
Bern
[71] 2017 NA GoogleNet Complex model, high-level classification, dimensionality features.
Barcelona
LeNet
Marginal increase of AUC values based on inputs of sensitivity by applying techniques
[28] 2019 TUH EEG Feature Extraction 2D-CNN
of dimensionality reduction.
In this TGCN model, si milar datasets are used but not configured identically and to train
[58] 2019 Clinical STFT TGCN the model utilizing the pooled dataset fro m various hospitals with different scalp EEG
sessions.
In future, the real time alar m based wearable device are available to detect before seizure
[67] 2019 Bonn Filtering 2D-CNN onset using single channel EEG detection. This may save the one’s life, not to occur
seizure furtherly.
Down Sampling,
Over conventional methods, AE has some i mportant benefits such as detecting the
[89] 2019 Clinical Filtering, Auto Encoder
seizures of any subject universally without lacking.
Normalization
GRU is a light weight model contain order of trainable parameters such as 100,000. In
[90] 2017 Bonn Auto-Correlation GRU
future this model will focus on huge multivariate datasets of EEG.
Seizure detection can be detected automatically by using this fra mework inorder to
[91] 2019 TUH EEG Different Methods SDA
classify various tasks.
1D-CNN
Limited time might be available inorder to investigate huge amount of EEG recordings
[79] 2018 Bonn Signal windowing with Bi
without errors. Hence automation may require in future.
LSTM
ASAE-CNN,
[81] 2018 Clinical Mapping In future, the use of deep subdural electrodes as an alternative source of iEEG data.
AAE-CNN

analysis of seizu re ep isodes. Singh et al. presented the imaging modalities and DL methods. Firstly, availab le
schematic layout of survival system for seizure detection datasets contain only selected EEG record ings, so that it is
within Figure 6. Kiral-Ko rnek et al. [88] described the not suitable to apply these signals in real world applicat ions,
specific hardware used for wearable devices which supports because the clinical dataset may not be available to public
long term performance and patient specific. Table VI and it’s hard to detect real time signals. All authors can be
indicates the detailed outline about works reviewed in this utilized this model to diagnose the seizures of epileptic
survey. patient with the help of CAD system by using DL
approaches. Secondly, these huge volumes of datasets are
V. CHALLENGES cannot be joined simply due to non-identical frequencies,
Currently, several challenges are faced by the researchers to sampling rate and also to train a model with s mall usable
data. Researchers acting in the domain of ep ileptic seizures
diagnose the seizures of epileptic patient using brain

978-1-6654-0118-0/22/$31.00 ©2022 IEEE 970

Authorized licensed use limited to: NORTHWESTERN POLYTECHNICAL UNIVERSITY. Downloaded on May 22,2023 at 05:02:23 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Fourth International Conference on Smart Systems and Inventive Technology (ICSSIT-2022)
IEEE Xplore Part Number: CFP22P17-ART; ISBN: 978-1-6654-0118-0

cannot have acquired always high-power hard ware is still members, with the help of wearable devices, such that the
challenging to perform novel DL approaches. This leads to patients can get proper treatment on time.
develop the computerized development of hardware which
is expensive and not effortlessly available to clinicians. So,
the researchers are required to work in this direction to
hinder with the epileptic episodes with effective prototypes,
to help the patients at any moment, any place.

VI. CONCLUSION
Over the last few years, several types of DL techniques like
AE, RNN, CNN are used for detecting epileptic seizures. In
this study, different screening techniques such as MRI and
EEG are used. Hardware and deep learn ing-based practical
methods are used in detecting epileptic seizures detection. In
the future, a detailed and carefu l study will be perfo rmed on
hardware and practical applications to get precise results.
Hence, various applications s uch as wearable devices and
mobiles have been used along with the models, cloud
servers for computational purposes. Diagnosis of seizures Fig. 6. Schematic layout of the proposed method utilizing DL approaches
through epilepsy using some alert notificat ions, where these for detecting seizures using EEG recordings.
alerts will send to doctors, relatives, hospitals, family
[11] Sharma, Rajeev, and Ram Bilas Pachori. "Classification of
REFERENCES epileptic seizures in EEG signals based on phase space
[1] Ghassemi, Navid, Afshin Shoeibi, Modjtaba Rouhani, and Hossein representation of intrinsic mode functions." Expert Systems with
Hosseini-Nejad. "Epileptic seizures detection in EEG signals Applications 42, no. 3 (2015): 1106-1117.
using T QWT and ensemble learning." In 2019 9th International [12] Romaine, James Brian, Mario Pereira Martín, José Ramón
Conference on Computer and Knowledge Engineering (ICCKE), Salvador Ortiz, and José María Manzano Crespo. "EEG—Single-
pp. 403-408. IEEE, 2019. Channel Envelope Synchronisation and Classification for Seizure
[2] Gupta, Vipin, Abhijit Bhattacharyya, and Ram Bilas Pachori. Detection and Prediction." Brain Sciences 11, no. 4 (2021): 516.
"Classification of seizure and non-seizure EEG signals based on [13] Perez-Sanchez, Andrea V., Carlos A. Perez-Ramirez, Martin
EMD-TQWT method." In 2017 22nd International Conference on Valtierra-Rodriguez, Aurelio Dominguez-Gonzalez, and Juan P.
Digital Signal Processing (DSP), pp. 1-5. IEEE, 2017. Amezquita-Sanchez. "Wavelet T ransform-Statistical T ime
[3] Bhattacharyya, Abhijit, Ram Bilas Pachori, Abhay Upadhyay, and Features-Based Methodology for Epileptic Seizure Prediction
U. Rajendra Acharya. "T unable-Q wavelet transform based Using Electrocardiogram Signals." Mathematics 8, no. 12 (2020):
multiscale entropy measure for automated classification of 2125.
epileptic EEG signals." Applied Sciences 7, no. 4 (2017): 385. [14] Alickovic, Emina, Jasmin Kevric, and Abdulhamit Subasi.
[4] Kulaseharan, Sancgeetha, Azad Aminpour, Mehran Ebrahimi, and "Performance evaluation of empirical mode decomposition,
ElysaWidjaja. "Identifying lesions in paediatric epilepsy using discrete wavelet transform, and wavelet packed decomposition for
morphometric and textural analysis of magnetic resonance automated epileptic seizure detection and prediction." Biomedical
images." NeuroImage: Clinical 21 (2019): 101663. signal processing and control 39 (2018): 94-102.
[5] van Klink, Nicole, Anne Mooij, Geertjan Huiskamp, Cyrille Ferrier, [15] Sharma, Manish, Ankit A. Bhurane, and U. Rajendra Acharya.
Kees Braun, Arjan Hillebrand, and Maeike Zijlmans. "MMSFL-OWFB: A novel class of orthogonal wavelet filters for
"Simultaneous MEG and EEG to detect ripples in people with epileptic seizure detection." Knowledge-Based Systems 160
focal epilepsy." Clinical Neurophysiology 130, no. 7 (2019): (2018): 265-277.
1175-1183. [16] Sharma, Manish, Ram Bilas Pachori, and U. Rajendra Acharya.
[6] Zazzaro, Gaetano, Salvatore Cuomo, Angelo Martone, R. Valentino "A new approach to characterize epileptic seizures using analytic
Montaquila, Gerardo Toraldo, and Luigi Pavone. "Eeg signal time-frequency flexible wavelet transform and fractal dimension."
analysis for epileptic seizures detection by applying data mining Pattern Recognition Letters 94 (2017): 172-179.
techniques." Internet of Things (2019): 100048. [17] Assi, Elie Bou, Dang K. Nguyen, Sandy Rihana, and Mohamad
[7] Pianou, Nikoletta, and Sofia Chatziioannou. "Imaging with PET/CT Sawan. "Towards accurate prediction of epileptic seizures: A
in Patients with Epilepsy." In Epilepsy Surgery and Intrinsic review." Biomedical Signal Processing and Control 34 (2017):
Brain T umor Surgery, pp. 45-50. Springer, Cham, 2019. 144-157.
[8] Lauretani, Fulvio, YariLongobucco, Giulia Ravazzoni, Elena [18] Acharya, U. Rajendra, Yuki Hagiwara, and HojjatAdeli.
Gallini, Marco Salvi, and Marcello Maggio. "Imaging the "Automated seizure prediction." Epilepsy & Behavior 88 (2018):
Functional Neuroanatomy of Parkinson’s Disease: Clinical 251-261.
Applications and Future Directions." International Journal of [19] Acharya, U. Rajendra, Yuki Hagiwara, Sunny Nitin Deshpande,
Environmental Research and Public Health 18, no. 5 (2021): S. Suren, Joel En Wei Koh, Shu Lih Oh, N. Arunkumar, Edward
2356. J. Ciaccio, and Choo Min Lim. "Characterization of focal EEG
[9] CHACON, Lilia Maria MORALES, Karla Batista García-Ramo, signals: a review." Future Generation Computer Systems 91
Judith González González, Juan E. Bender del Busto, Margarita (2019): 290-299.
MinouBáez Martin, Barbara Ofelia Estupiñan Diaz, Lourdes del [20] Khodatars, Marjane, Afshin Shoeibi, NavidGhassemi,
Carmen LorigadosPedre et al. "Estrategia para la evaluacióny Mahboobeh Jafari, Ali Khadem, Delaram Sadeghi, Parisa
tratamientoquirúrgico de las epilepsiasfarmacorresistentesen Moridian et al. "Deep learning for neuroimaging-based diagnosis
Cuba." Anales de la Academia de Ciencias de Cuba 11, no. 1 and rehabilitation of autism spectrum disorder: A review." arXiv
(2021): 809. preprint arXiv:2007.01285 (2020).
[10] Subasi, Abdulhamit, Jasmin Kevric, and M. Abdullah Canbaz. [21] Sadeghi, Delaram, Afshin Shoeibi, NavidGhassemi, Parisa
"Epileptic seizure detection using hybrid machine learning Moridian, Ali Khadem, RoohallahAlizadehsani, Mohammad
methods." Neural Computing and Applicatepilepions 31, no. 1 Teshnehlab, J. Manuel Gorriz, and SaeidNahavandi. "An
(2019): 317-325. Overview on Artificial Intelligence T echniques for Diagnosis of

978-1-6654-0118-0/22/$31.00 ©2022 IEEE 971

Authorized licensed use limited to: NORTHWESTERN POLYTECHNICAL UNIVERSITY. Downloaded on May 22,2023 at 05:02:23 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Fourth International Conference on Smart Systems and Inventive Technology (ICSSIT-2022)
IEEE Xplore Part Number: CFP22P17-ART; ISBN: 978-1-6654-0118-0

Schizophrenia Based on Magnetic Resonance Imaging IEEE Transactions on Biomedical Engineering., 66(3): 601-608.
Modalities: Methods, Challenges, and Future Works." arXiv http://dx.doi.org/ 10.1109/TBME.2018.2850959 .
preprint arXiv:2103.03081 (2021). [39] Juozapavicius,A,Bacevivius,GBugelskis, D,Samaitiene, R 2011,
[22] LeCun, Yann, YoshuaBengio, and Geoffrey Hinton. "Deep „EEG analysis – Automatic spike detection‟, Nonlinear Analyis:
learning”, nature, vol. 521, no. 7553. DOI 10 (2015): 436. Modeling and Control, vol, 16, no. 4, pp.375-386.
[23] Pedrycz, Witold, and Shyi-Ming Chen, eds. “Development and [40] Cociu, BA, Das, S, Billeci, L, Jamal, W, Maharatna, K,
Analysis of Deep Learning Architectures”. Springer, 2020. Calderoni, S, Narzisi, A &Muratori, F 2017, 'Multimodal
[24] Srivastava, Nitish, and Ruslan Salakhutdinov. "Multimodal Functional and Structural Brain Connectivity Analysis in Autism:
Learning with Deep Boltzmann Machines." In NIPS, vol. 1, p. 2. A Preliminary Integrated Approach with EEG, fMRI and DTI',
2012. IEEE Transactions on Cognitive and Developmental Systems, pp.
1-1
[25] Yu, Dong, and Li Deng. "Deep learning and its applications to
signal and information processing [exploratory dsp]." IEEE [41] Qian, D, Wang, B, Qing, X, Zhang, T, Zhang, Y, Wang, X &
Signal Processing Magazine 28, no. 1 (2010): 145-154. Nakamura, M 2017, 'Drowsiness Detection by Bayesian-Copula
Discriminant Classifier Based on EEG Signals during Daytime
[26] Le, Thanh Xuyen, Trung Thanh Le, Van Viet Dinh, Quoc Long
Short Nap', IEEE Transactions on Biomedical Engineering, vol.
Tran, Linh Trung Nguyen, and Duc Thuan Nguyen. "Deep
64, no. 4, pp. 743-754.
learning for epileptic spike detection." VNU Journal of Science:
Computer Science and Communication Engineering 33, no. 2 [42] Yang, Y, Wu, QMJ, Zheng, W-L & Lu, B-L 2017, 'EEG-based
(2018): 1-13. emotion recognition using hierarchical network with subnetwork
nodes', IEEE Transactions on Cognitive and Developmental
[27] Yao, Xinghua, Qiang Cheng, and Guo-Qiang Zhang. "A novel Systems, pp. 1-1.
independent RNN approach to classification of seizures against
non-seizures." arXiv preprint arXiv:1903.09326 (2019). [43] Costa, F, Batatia, H, Oberlin, T &Tourneret, JY 2017, 'Skull
Conductivity Estimation for EEG Source Localization', IEEE
[28] Iesmantas, Tomas, and RobertasAlzbutas. "Convolutional neural Signal Processing Letters, vol. 24, no. 4, pp. 422-426.
network for detection and classification of seizures in clinical
data." Medical & Biological Engineering & Computing 58, no. 9 [44] Giri, EP, Fanany, MI &Arymurthy, AM 2016, 'Ischemic Stroke
(2020): 1919-1932. Identification Based on EEG and EOG using 1D Convolutional
Neural Network and Batch Normalization', „arVix.org‟, pp.1-13.
[29] Liang, Weixia, Haijun Pei, Qingling Cai, and Yonghua Wang.
"Scalp EEG epileptogenic zone recognition and localization based [45] M. Ihle, H. Feldwisch-Drentrup, C. A. Teixeira, A. Witon, B.
on long-term recurrent convolutional network." Neurocomputing Schelter, J. Timmer, and A. Schulze-Bonhage, “Epilepsiae–a
396 (2020): 569-576. european epilepsy database,” Computer methods and programs in
biomedicine, vol. 106, no. 3, pp. 127–138, 2012.
[30] Thomas, John, Luca Comoretto, Jing Jin, Justin Dauwels, Sydney
S. Cash, and M. Brandon Westover. "EEG classification via [46] A. E. Society. (2015) Seizure prediction challenge. [Online].
convolutional neural network-based interictal epileptiform event Available: https://www.kaggle.com/c/seizure-prediction
detection." In 2018 40th Annual International Conference of the [47] A. H. Shoeb, “Application of machine learning to epileptic
IEEE Engineering in Medicine and Biology Society (EMBC), pp. seizure onset detection and treatment,” Ph.D. dissertation,
3148-3151. IEEE, 2018. Massachusetts Institute of Technology, 2009.
[31] Wang, Y., Li, Z., Feng, L., Bai, H. and Wang, C. (2017). [48] R. G. Andrzejak, K. Lehnertz, F. Mormann, C. Rieke, P. David,
Hardware design of multiclass SVM classification for epilepsy and C. E. Elger, “Indications of nonlinear deterministic and
and epileptic seizure detection. IET Circuits, Devices & Systems, finitedimensional structures in time series of brain electrical
12(1): 108-115. http://dx.doi.org/ 10.1049/iet-cds.2017.0216. activity: Dependence on recording region and brain state,”
[32] Jiang, Y., Wu, D., Deng, Z., Qian, P., Wang, J., Wang, G. and Physical Review E, vol. 64, no. 6, p. 061907, 2001.
Wang, S. (2017). Seizure Classification From EEG Signals Using [49] R. G. Andrzejak, K. Schindler, and C. Rummel,
Transfer Learning. Semi-Supervised Learning and T SK Fuzzy “Nonrandomness, nonlinear dependence, and nonstationarity of
System, IEEE Transactions on Neural Systems and Rehabilit ation electroencephalographic recordings from epilepsy patients,”
Engineering, 25(12): 2270-2284. http://dx.doi.org/ Physical Review E, vol. 86, no. 4, p. 046206, 2012.
10.1109/TNSRE.2017.2748388. [50] N. Stevenson, K. T apani, L. Lauronen, and S. Vanhatalo, “A
[33] Abualsaud, K., Mohamed, A., Khattab, T ., Yaacoub, E., Hasna, dataset of neonatal eeg recordings with seizure annotations,”
M., and Guizani, M. (2018). Classification for Imperfect EEG Scientific data, vol. 6, p. 190039, 2019.
Epileptic Seizure in IoT applications: A Comparat ive Study. In [51] Faust, Oliver, Yuki Hagiwara, Tan Jen Hong, Oh Shu Lih, and U.
2018 14th International Wireless Communications and Mobile Rajendra Acharya. "Deep learning for healthcare applications
Computing Conference (IWCMC) IEEE., 364- based on physiological signals: A review." Computer methods
369.http://dx.doi.org/ 10.1109/IWCMC.2018.8450279. and programs in biomedicine 161 (2018): 1-13.
[34] T siouris, K. M., Konitsiotis, S., Markoula, S., Rigas, G., [52] Alizadehsani, Roohallah, Mohamad Roshanzamir, Sadiq Hussain,
Koutsouris, D. D., and Fotiadis, D. I. (2018). Unsupervised Abbas Khosravi, AfsanehKoohestani, Mohammad Hossein
detection of epileptic seizures from EEG signals: A channel- Zangooei, MoloudAbdar et al. "Handling of uncertainty in
specific analysis of long-term recordings. In Biomedical & Health medical data using machine learning and probability theory
Informatics (BHI), 2018 IEEE EMBS International Conference techniques: A review of 30 years (1991–2020)." Annals of
on IEEE., 92-95. 199 http://dx.doi.org/ Operations Research (2021): 1-42.
10.1109/BHI.2018.8333377.
[53] Alizadehsani, Roohallah, Danial Sharifrazi, NavidHoseini Izadi,
[35] Solaija, M. S. J., Saleem, S., Khurshid, K., Hassan, S. A., and Javad HassannatajJoloudari, Afshin Shoeibi, Juan M. Gorriz,
Kamboh, A. M. (2018). Dynamic Mode Decomposition Based Sadiq Hussain et al. "Uncertainty-aware semi-supervised method
Epileptic Seizure Detection from Scalp EEG. IEEE Access., 6: using large unlabelled and limited labeled COVID-19 data."
38683-3869 http://dx.doi.org/10.1109/ACCESS.2018.2853125. arXiv preprint arXiv:2102.06388 (2021).
[36] Saidi, A., Othman, S. B., Kacem, W., and Saoud, S. B. (2018). [54] Pereira, Sérgio, Adriano Pinto, Victor Alves, and Carlos A. Silva.
FPGA implementation of EEG signal analysis system for the "Brain tumor segmentation using convolutional neural networks
detection of epileptic seizure. In 2018 International Conference on in MRI images." IEEE transactions on medical imaging 35, no. 5
Advanced Systems and Electric Technologies (IC_ASET) (2016): 1240-1251.
IEEE.,415-420.http://dx.doi.org/10.1109/ASET.2018.8379892.
[55] Ghassemi, Navid, Afshin Shoeibi, and Modjtaba Rouhani. "Deep
[37] Yuan, Y., Xun, G., Jia, K., and Zhang, A. (2018). A Multi-view neural network with generative adversarial networks pre-training
Deep Learning Framework for EEG Seizure Detection. IEEE for brain tumor classification based on MR images." Biomedical
Journal of Biomedical and Health Informatics, 23(1): 83-94. Signal Processing and Control 57 (2020): 101678.
http://dx.doi.org
[56] Masi, Iacopo, Yue Wu, Tal Hassner, and Prem Natarajan. "Deep
[38] Fan, M., and Chou, C. A. (2018). Detecting Abnormal Pattern of face recognition: A survey." In 2018 31st SIBGRAPI conference
Epileptic Seizures via Temporal Synchronization of EEG Signals, on graphics, patterns and images (SIBGRAPI), pp. 471-478.
IEEE, 2018.

978-1-6654-0118-0/22/$31.00 ©2022 IEEE 972

Authorized licensed use limited to: NORTHWESTERN POLYTECHNICAL UNIVERSITY. Downloaded on May 22,2023 at 05:02:23 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Fourth International Conference on Smart Systems and Inventive Technology (ICSSIT-2022)
IEEE Xplore Part Number: CFP22P17-ART; ISBN: 978-1-6654-0118-0

[57] Asif, Umar, Subhrajit Roy, Jianbin Tang, and Stefan Harrer. [74] J. T urner, A. Page, T. Mohsenin, and T. Oates, “Deep belief
"SeizureNet: Multi-spectral deep feature learning for seizure type networks used on high resolution multichannel
classification." In Machine Learning in Clinical Neuroimaging electroencephalography data for seizure detection,” in 2014
and Radiogenomics in Neuro-oncology, pp. 77-87. Springer, AAAI Spring Symposium Series, 2014
Cham, 2020. [75] Golmohammadi, Meysam, SaeedehZiyabari, Vinit Shah, Silvia
[58] Covert, Ian C., Balu Krishnan, Imad Najm, Jiening Zhan, Lopez de Diego, Iyad Obeid, and Joseph Picone. "Deep
Matthew Shore, John Hixson, and Ming Jack Po. "Temporal architectures for automated seizure detection in scalp EEGs."
graph convolutional networks for automatic seizure detection." In arXiv preprint arXiv:1712.09776 (2017).
Machine Learning for Healthcare Conference, pp. 160-180. [76] Fang, Zhou, Howan Leung, and Chiu Sing Choy. "Spatial
PMLR, 2019. temporal GRU convnets for vision-based real time epileptic
[59] Bouaziz, Bassem, LotfiChaari, Hadj Batatia, and Antonio seizure detection." In 2018 IEEE 15th International Symposium
Quintero-Rincón. "Epileptic seizure detection using a on Biomedical Imaging (ISBI 2018), pp. 1026-1029. IEEE, 2018.
convolutional neural network." In Digital Health Approach for [77] R. Hussein, H. Palangi, Z. J. Wang, and R. Ward, “ Robust
Predictive, Preventive, Personalised and Participatory Medicine, detection of epileptic seizures using deep neural networks,” in
pp. 79-86. Springer, Cham, 2019. 2018 IEEE International Conference on Acoustics, Speech and
[60] Deng, Jia, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Signal Processing (ICASSP). IEEE, 2018, pp. 2546–2550.
Fei-Fei. "Imagenet: A large-scale hierarchical image database." In [78] S. T. Jaafar and M. Mohammadi, “Epileptic seizure detection
2009 IEEE conference on computer vision and pattern using deep learning approach,” UHD Journal of Science and
recognition, pp. 248-255. Ieee, 2009. T echnology, vol. 3, no. 2, pp. 41–50, 2019.
[61] Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton. [79] A. M. Abdelhameed, H. G. Daoud, and M. Bayoumi, “ Epileptic
"Imagenet classification with deep convolutional neural seizure detection using deep convolutional autoencoder,” in 2018
networks." Advances in neural information processing systems 25 IEEE International Workshop on Signal Processing Systems
(2012): 1097-1105. (SiPS). IEEE, 2018, pp. 223–228
[62] Simonyan, Karen, and Andrew Zisserman. "Very deep [80] T. Wen and Z. Zhang, “Deep convolution neural network and
convolutional networks for large-scale image recognition." arXiv autoencoders-based unsupervised feature learning of eeg signals,”
preprint arXiv:1409.1556 (2014). IEEE Access, vol. 6, pp. 25 399–25 410, 2018.
[63] Ahmedt-Aristizabal, David, Clinton Fookes, Kien Nguyen, Simon [81] A. Antoniades, L. Spyrou, D. Martin-Lopez, A. Valentin, G.
Denman, Sridha Sridharan, and Sasha Dionisio. "Deep facial Alarcon, S. Sanei, and C. C. Took, “Deep neural architectures for
analysis: A new phase I epilepsy evaluation using computer mapping scalp to intracranial eeg,” International journal of neural
vision." Epilepsy & Behavior 82 (2018): 17-24. systems, vol. 28, no. 08, p. 1850009, 2018
[64] Szegedy, Christian, Wei Liu, Yangqing Jia, Pierre Sermanet, [82] A. M. Karim, O. Karal, and F. C¸ elebi, “A new automatic
Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent epilepsy ¨ serious detection method by using deep learning based
Vanhoucke, and Andrew Rabinovich. "Going deeper with on discrete wavelet transform,” no, vol. 4, pp. 15–18, 2018.
convolutions." In Proceedings of the IEEE conference on
computer vision and pattern recognition, pp. 1-9. 2015. [83] X. Yao, Q. Cheng, and G.-Q. Zhang, “Automated classification of
seizures against nonseizures: A deep learning approach,” arXiv
[65] He, Kaiming, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. preprint arXiv:1906.02745, 2019.
"Deep residual learning for image recognition." In Proceedings of
the IEEE conference on computer vision and pattern recognition, [84] M. T. Avcu, Z. Zhang, and D. W. S. Chan, “Seizure detection
pp. 770-778. 2016. using least eeg channels by deep convolutional neural network,”
in ICASSP 2019-2019 IEEE International Conference on
[66] Bizopoulos, Paschalis, George I. Lambrou, and Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2019,
DimitriosKoutsouris. "Signal2image modules in deep neural pp. 1120–1124.
networks for eeg classification." In 2019 41st Annual
[85] F. Achilles, F. Tombari, V. Belagiannis, A. M. Loesch, S.
International Conference of the IEEE Engineering in Medicine
Noachtar, and N. Navab, “Convolutional neural networks for real-
and Biology Society (EMBC), pp. 702-705. IEEE, 2019.
time epileptic seizure detection,” Computer Methods in
[67] Liu, Jianguo, and Blake Woodson. "Deep learning classification Biomechanics and Biomedical Engineering: Imaging &
for epilepsy detection using a single channel Visualization, vol. 6, no. 3, pp. 264–269, 2018.
electroencephalography (EEG)." In Proceedings of the 2019 3rd
International Conference on Deep Learning Technologies, pp. 23- [86] Hosseini, Mohammad-Parsa, Hamid Soltanian-Zadeh, Kost
26. 2019. Elisevich, and Dario Pompili. "Cloud-based deep learning of big
EEG data for epileptic seizure prediction." In 2016 IEEE global
[68] Tjepkema-Cloostermans, Marleen C., Rafael CV de Carvalho, conference on signal and information processing (GlobalSIP), pp.
and Michel JAM van Putten. "Deep learning for detection of focal 1151-1155. IEEE, 2016.
epileptiform discharges from scalp EEG recordings." Clinical
neurophysiology 129, no. 10 (2018): 2191-2196. [87] Singh, Kuldeep, and Jyoteesh Malhotra. "Stacked autoencoders
based deep learning approach for automatic epileptic seizure
[69] fRoy, Subhrajit, Isabell Kiral-Kornek, and Stefan Harrer. "Deep detection." In 2018 First International Conference on Secure
learning enabled automatic abnormal EEG identification." In Cyber Computing and Communication (ICSCCC), pp. 249-254.
2018 40th Annual International Conference of the IEEE IEEE, 2018.
Engineering in Medicine and Biology Society (EMBC), pp. 2756-
[88] Kiral-Kornek, Isabell, Subhrajit Roy, Ewan Nurse, Benjamin
2759. IEEE, 2018.
Mashford, Philippa Karoly, Thomas Carroll, Daniel Payne et al.
[70] P. Nejedly, V. Kremen, V. Sladky, M. Nasseri, H. Guragain, P. "Epileptic seizure prediction using big data and deep learning:
Klimes, J. Cimbalnik, Y. Varatharajah, B. H. Brinkmann, and G. toward a mobile system." EBioMedicine 27 (2018): 103-111.
A. Worrell, “Deep-learning for seizure forecasting in canines with
epilepsy,” Journal of neural engineering, vol. 16, no. 3, p. 036031, [89] Emami, Ali, Naoto Kunii, Takeshi Matsuo, Takashi Shinozaki,
2019. Kensuke Kawai, and Hirokazu Takahashi. "Autoencoding of
long-term scalp electroencephalogram to detect epileptic seizure
[71] A. M. Taqi, F. Al-Azzo, M. Mariofanna, and J. M. Al-Saadi, for diagnosis support system." Computers in biology and
“ Classification and discrimination of focal and non-focal eeg medicine 110 (2019): 227-233.
signals based on deep neural network,” in 2017 International
Conference on Current Research in Computer Science and [90] Talathi, Sachin S. "Deep Recurrent Neural Networks for seizure
Information Technology (ICCIT). IEEE, 2017, pp. 86–92 detection and early seizure detection systems." arXiv preprint
arXiv:1706.03283 (2017).
[72] Goodfellow, Ian, YoshuaBengio, and Aaron Courville. Deep
[91] Golmohammadi, Meysam, Amir Hossein Harati Nejad Torbati,
learning. MIT press, 2016.
Silvia Lopez de Diego, Iyad Obeid, and Joseph Picone.
[73] Roy, Subhrajit, Isabell Kiral-Kornek, and Stefan Harrer. "Automatic analysis of EEGs using big data and hybrid deep
"ChronoNet: a deep recurrent neural network for abnormal EEG learning architectures." Frontiers in human neuroscience 13
identification." In Conference on Artificial Intelligence in (2019): 76.
Medicine in Europe, pp. 47-56. Springer, Cham, 2019.

978-1-6654-0118-0/22/$31.00 ©2022 IEEE 973

Authorized licensed use limited to: NORTHWESTERN POLYTECHNICAL UNIVERSITY. Downloaded on May 22,2023 at 05:02:23 UTC from IEEE Xplore. Restrictions apply.