Page 121 - 2024-Vol20-Issue2

117 | Assim & Mahmood

Table I describes the key frequencies and amplitudes of human
EEG waves. EEG recordings can be obtained through stan-
dard EEG, sleep EEG, ambulatory EEG, and video telemetry.
However, it’s important to note that EEG has limitations in

TABLE I.
BASIC BRAIN WAVES CHARACTERISTICS

Freq. Frq. HZ Details states Fig. 2. Stages of epileptic diagnosis by deep learning.
Band More than 30 Amp. mv Concentration
Gamma 1) Data Preprocessing:
15-30 5-10 Anxiety is • Clean the EEG data to remove noise and artifacts that
Beta prevalent, might interfere with the analysis.
9-14 2-20 energetic,
Alpha focused on • Segment EEG signals into smaller epochs for analysis.
4-8 20-60
Theta 1-3 others, • Convert EEG signals into a format suitable for input
Delta 2-100 and calm. into deep learning models, such as numerical arrays.
20-200
very 2) Feature Extraction:
calm, • Extract relevant features from EEG signals that indicate
unresponsive different types of epileptic activity.
focus
internally • Common features include spectral features, statistical
concentrated measures, and time-domain features.
and deeply
relaxed 3) Data Augmentation (Optional):
Sleep Augment the dataset by applying transformations like rota-
tion, scaling, or adding noise. This step helps generate more
precisely recording deep brain cortex layers. diverse training samples, especially when the original dataset
is limited.
B. Magnetic Radiographic Imaging MRI: Secondly, the Deep learning model needs the following:
MRI is a radiographic imaging technique used to study the
structural and functional problems of epilepsy [18, 19]. Func- 1. Model selection:
tional MRI (fMRI) observes the brain’s reaction to stimuli and
helps identify epilepsy etiology. MRI maps the brain’s white • Choose an appropriate deep-learning architecture
and grey matter distribution and blood flow rate. While MRI for the task.
provides detailed images, it is a costly procedure requiring
advanced instruments and expertise. [20–23]. • Experiment with various architectures to find the
most suitable one for the specific EEG classifica-
C. Modern Techniques for Epileptic Diagnosis: tion task.
Technology advancements have led to the integration of Ar-
tificial Intelligence (AI) in health systems [24–29]. Machine 2. Model Training:
learning (ML) and deep learning (DL) methodologies are used
for epileptic diagnosis [30–35]. ML models require iterative • Train the selected deep learning model using the
processes of feature selection and classification. In contrast, pre-processed and augmented data.
DL models, such as Convolutional Neural Networks (CNNs)
and Recurrent Neural Networks (RNNs), demand a large • Utilize appropriate loss functions and optimiza-
amount of data for effective training [35–39]. Epileptic diag- tion techniques for training the model.
nosis using deep learning involves several stages, as shown
in Fig. 2. Here’s an outline of the typical stages involved in • Monitor the training process, validating the mod-
using deep learning techniques for diagnosing epilepsy: First els performance on a separate validation dataset
of all, there may be a need for some process for the dataset, to prevent over-fitting. the specific EEG classifi-
Like : cation task.