Article Preview
Top1. Introduction
Epilepsy is the neurological diseases over world-wide, which arise in brain, and defects nervous system. In other words, the epilepsy is the disease of neurological in brain that happened by the sudden abnormal changes of neurons, about 1% of world’s population suffers from the epilepsy (Alzami, et al., 2018). In short term, the epilepsy patients having the little influence, whereas the long-term attacks seriously happened defects on patient’s health (Li, et al., 2017). Consequently, seizure is happened when the abnormal hyper synchronous releases the neurons in brain. The occurrence in human brain related closely to entire characteristics. In addition, the seizure is classified into generalized and focal. In initial stage, the specific region of the brain gets damaged, and in secondary stage, whole brain region is affected by the activity of excessive neuronal (Parvez and Paul, 2015). Electroencephalography (EEG) signal (Valsalan, et al., 2020) plays the very important role for the epilepsy detection. Here, the EEG electrodes placement on scalp brain to obtain the EEG signals (George, et al., 2012). Thus, the epilepsy diagnosis from EEG recordings made by neurologist is the expensive and time consuming (Das, et al., 2020).
EEG is the non-invasive electrophysiological test for monitoring electrical activity of the brain. However, EEG signal is measure by the use of voltage fluctuations that results from ionic currents circulating brain neurons (Alzami, et al., 2018). Thus, it is utilized for diagnosing epilepsy that is characterized by the recurrent seizures resulting discharges of cells neuron (Li, et al., 2017). The long-term monitoring EEG signal needs stationary situations developments based on low, hence makes mobility. Recent energy wireless transmission and the sensor design leads to new wireless headsets for recording (Bhattacharyya & Pachori, 2017). Therefore, the signals captured from wireless headsets results to the motion-based artifacts (Schindler, et al., 2002) (Remmiya and Abisha, 2018). The conversion of EEG signals into images is the recent widely used approach, and the conversion is carried out using the changing points, such as local minimum and maximum, the traditional variational autoencoder (VAE) method based on LSTM, and the generative adversarial network (GAN), etc (Kavasidis, et al., 2017). The images generated in this methods are more realistic, semantically coherent, and less computationally complexity. Anyhow, the missing of fine details and noises are the major drawbacks of these methods (Hatipoglu, et al., 2018). The detection of the epileptic seizures requires large duration for patient’s signals monitoring. However, manual long duration monitoring of patient’s EEG signal is time-consuming and tedious. Additionally, the obtained EEG signals are contaminated with the neuronal symptomatology, and background noise (Usulkar, et al., 2016). Therefore, the automatic seizure detection facilitates real-time therapy and monitoring epileptic seizures (Eftekhar, et al., 2014).
The epileptic seizure detection has been paid great attention for medical epilepsy detection. In addition, EEG is utilized in several applications, like video quality assessment, emotion recognition, sleep stage detection, alcoholic consumption measurement, change brainwaves by smoking, and the mobile phone usages. Several signal processing approaches are introduced to extract appropriate distinctive features from EEG signals automatically for epileptic seizure prediction. Thus, features are partitioned into time-domain features, such as spiking rate (Schindler, et al., 2002), and counts of amplitude patterns (Eftekhar, et al., 2014), the frequency-domain features, such as hybrid time frequency “(t-f)”, and sub-band spectral powers (Bandarabadi, et al., 2015). In (Boubchir, et al., 2017), “t-f” analysis is performed for extracting the person identification features in order to detect the inter-patient variabilities. Ensemble learning is introduced in (Bandarabadi, et al., 2015) for classifying temporal and spectral features (Islam, et al., 2020). Recent detection systems considered several physiological parameters, likes temperature, Heart Rate (HR), skin conductance, 3D Gyroscope (GYR), and so on. Thus, the multivariate systems permit to increase the accuracy of detection, and to reduce FAR (Salem, et al., 2018).