Epilepsy is a common neurological illness marked by repeated and uncontrolled seizures that primarily impact sufferers' lives. In many circumstances, electroencephalogram signals provide essential physiological information about human brain activity that can be utilized for an epilepsy diagnosis. Visual evaluation of a large number of electroencephalogram data, on the other hand, takes a long time and often results in discrepancies in doctors' findings. The electroencephalogram (EEG) signal is very important in the diagnosis of epilepsy because it can show if there are brain disorders or illnesses.
A team of researchers from Islamic Azad University in Najafabad, Iran, led by Mohammad Reza Yousef, Saina Golnejad, and Melika Mohammad Hosseini, developed a single instruction to diagnose epilepsy that comprises two phases. In the initial phase, a low-pass filter was used to preprocess the data, then three independent mid-pass filters for various frequency bands and a multilayer neural network were built. The wavelet transform method was utilized to handle data in the second stage. The findings showed a somewhat equal effect factor on the use or non-use of wavelet transform on the development of epilepsy data functions. Still, it was found that using a perceptron multilayer neural network may be better for experts.
3D graphic of human with brain
The researchers utilized epilepsy data from the Bonn University database, which included five distinct registration models divided into 500 pieces, each with 100 points. The data was preprocessed using a 70 Hz low-pass filter. The alpha, beta, and gamma frequency bands were then isolated by building a separate mid-pass filter in the data processing stage. The categorization was constructed using a multilayer perceptron neural network (MLP). After this classifier was tested and the results were good enough, the wavelet transform method was used on its own.
Feature extraction is used to create an epilepsy detection model using standard epilepsy data and perform epilepsy detection using EEG signal data. A signal is decomposed by the wavelet transform into scaled and translated copies of a mother wavelet and a scaling function. The discrete wavelet transform (DWT) has been widely employed in recent epileptic spike and seizure detection methods that have shown encouraging results. The impact of feature extraction is strongly connected to the accuracy of epilepsy diagnosis; hence, improving feature extraction is critical. The EEG recording takes around 20 minutes to capture the epilepsy signal separately and includes photostimulation and rest intervals with open and closed eyes. Four essential things about the frequency and time of data for epileptic and non-epileptic data were taken from each alpha, gamma, and beta frequency band.
The median and mean frequencies rise because of the greater brain frequency in a person with epilepsy, but the frequency ratio falls. The changes are measured, and the absolute value of the following sample decreases from the previous model in the waveform length characteristics. If the signal is smooth with few changes, this attribute becomes zero. As a result, in epileptic illness, when the disorder's entropy is low, and the brain oscillations are the same, the waveform length drops. The entropy was calculated using the Shannon entropy and the wavelet transform, which estimates the entropy based on bit changes. A multilayer neural network with ten input layers and 5% validation was used to categorize the epilepsy signal, and this classifier obtained 95.5% accuracy in the multilayer neural network. The perceptron multilayer neural network's validation value and accuracy The entropy was then calculated using the wavelet transform with level 8 and decomposition level 4 and the Shannon entropy. This approach yielded 91% accuracy in the diagnosis of epilepsy.
A professional physician evaluates an epileptic episode by a visual assessment of long-term EEG data. The recording process is both expensive and time-consuming. The present work developed a novel seizure detection method based on EEG data to solve these issues. The study's fundamental strength in epilepsy is the description of the given sample using a multilayer perceptron neural network classifier and comparison with a wavelet transform function.
Future research with bigger sample sizes is required to confirm and extend these findings. The diagnostic significance of physiological signals recorded using EEG was assessed in this research. This study also successfully extracted features, which are the median and mean frequencies, which are obtained by using the Fourier transform of the desired signal. These two characteristics increase in a person with epilepsy because the brain frequency is higher than in the average person. The frequency rate refers to the brain's high-to-low frequency ratio, and since a person with epilepsy has a greater brain frequency than the average person, the frequency rate falls. However, waveform length, a temporal feature, subtracts the absolute magnitude of subsequent samples of the signal from the previous models and adds them together, actually measuring the changes and complexity of the movement, and waveform length is reduced in a person with epilepsy due to decreased entropy and similar fluctuations in the brain. Calculating these attributes can help doctors figure out how to apply this procedure to people who have epilepsy.