Investigation of Epileptic Seizures and Sleep Disturbance

Introduction

In the field of neuroscience, the need to study the electrical activity of the brain is an important aspect. A wide range of tools and technologies have been introduced day by day. The most prominent tool which is used in a wide range is EEG. Electroencephalography (EEG) proves helpful in studying human brain activity and epileptic processes. EEG signals provide crucial information about epileptogenic networks that must be reviewed and analyzed before the commencement of therapeutic procedures. EEG signals are non-stationary signals. The most advantageous, beneficial, and cost-effective method for the study of epilepsy and sleep apnea is the use of EEG signals. The disorder functioning of the brain can lead to epileptic seizures which can affect the patient’s health. The minute discrepancies in EEG signals show a definite type of brain abnormality. The ultimate goal is to design and develop signal processing algorithms that extract this information and use it for diagnosis and treating patients with epilepsy.

Sleep apnea is a serious sleep disorder that can interrupt a person’s breathing during their sleep. This occurs when the throat muscles get more relaxed which can cause the airway to get blocked causing difficulty in breathing. Patients with epilepsy were reported to have concomitant sleep apnea. It is observed that a third or more of patients suffering from epilepsy have obstructive sleep apnea. Sleep apnea is known to disrupt sleep which can bring on epileptic seizures in patients already having epilepsy.

EEG measures variations in electric potentials caused by a large number of electric dipoles which are formed during neural excitations. An EEG signal comprises different brain waves which reflect the electrical activity of the brain based on the placement of the electrode. When measured on the scalp, the EEG signal amplitude is about 100 µV and it is about 1-2 mV when measured on the surface of the brain. The bandwidth of the signal falls under 1 Hz to about 50 Hz.

In both the functional and numerical analysis, a discrete wavelet transforms (DWT) is any wavelet transform in which the wavelets are discretely sampled. In comparison with other wavelet transforms, a major advantage of DWT over Fourier transforms is temporal resolution wherein it captures both frequency and location information (location in time).In the EEG signals, time-frequency signal-processing algorithms such as discrete wavelet transform(DWT) analysis are done. In order to describe the time and frequency domain of the EEG signal, It is important to address the different behavior of the EEG signal. It should also be understood that the DWT is used for the analysis of non-stationary signals, and this shows a good advantage over spectral analysis. Also, the location of the transient events is done by DWT. Such transient events such as spikes can occur during epileptic seizures.

The dimensionality reduction method which is used to reduce the dimensions of large data sets is Kernel Principal component analysis (K-PCA) is used majorly in data analysis and for creating predictive models. KPCA is used to obtain lower-dimensional data while preserving as much of the data's variation as possible. EEG signals were decomposed into frequency sub-bands using DWT. Then a group or set of features was extracted from these sub-bands to represent the distribution of wavelet coefficients. PCA is used in reducing the dimension of the data. Then these features were given as an input to an artificial neural network (ANN classifier) with two distinct outputs: epileptic or non-epileptic seizure. The accuracy and performance of the classifier will be analyzed and compared.

Another method of EEG signal analysis is the Hilbert–Huang Transform (HHT) and empirical mode decomposition (EMD) which is developed mainly for non-stationary signals. This coupling is done mainly done to extract the intrinsic functions of the EEG signal. This involves decomposing the original signal into a series of intrinsic mode functions (IMFs) through empirical mode decomposition (EMD) and then performing Hilbert–Huang Transform (HHT) on each intrinsic mode function. Finally, the corresponding energy and marginal spectrum are calculated in order to classify the features. The plot or visual mapping of the spectrum of frequencies of a signal along the time is called a spectrogram. This plot has a relation between time and frequency resolution that is a large window that has less localization in time and more discrimination in frequency. Spectrogram depends on the windowing. Hence selection of proper and appropriate window length is needed.