Cepstrum Coefficient Analysis from Low-Frequency to High-Frequency Applied to Automatic Epileptic Seizure Detection with Bio-Electrical Signals

Abstract

This study analyzes bioelectrical signals to achieve automatic epileptic seizure detection. Electroencephalographic (EEG) signals were recorded with electrodes on healthy, epileptic seizure-free, and epileptic seizure patients. The challenges in this field are generally regarded to be the impacts of non-stationarity and nonlinearity in EEG signals. To address these challenges, this study attempts to recognize different brain statuses. The idea originated from a novel hypothesis that considers EEG signals as convolution signals and regards itself as the generation mechanism of EEG signals, to some extent. Based on this hypothesis, the nonlinear problem can be viewed as a deconvolution procedure. As such, the method can be simplified into three parts: eliminating non-stationary is used to catch high-frequency to low-frequency signals, which is followed by a local mean decomposition (LMD) algorithm; these signals are deconvoluted to form ultra-high-dimensional feature sets, which is completely terminated by the mel-frequency cepstrum coefficients (MFCC) algorithm; and several classifiers are combined to achieve highly accurate recognition results and to verify the superiority and reasonableness of this method. The publicly available EEG database from the University of Bonn, Germany is employed to demonstrate the effectiveness and outstanding performance of this method. According to the results, the method has the ability to attain a higher average classification accuracy than other methods in all of the four following cases: healthy (datasets A and B) versus epileptic seizure (dataset E), epileptic seizure-free (datasets C and D) versus epileptic seizure (dataset E), healthy (datasets A and B) versus epileptic seizure-free (datasets C and D) versus epileptic seizure (dataset E), and healthy (dataset A) versus healthy (dataset B) versus epileptic seizure-free (dataset C) versus epileptic seizure-free (dataset D) versus epileptic seizure (dataset E).

Short-term epileptic seizures prediction based on cepstrum analysis and signal morphology

“…According to the hypothesis of [ 15 ], EEG signals can be regarded as the result of the convolution of two functions generated by motivation signals and the common effect of the neuronal structures. This reference, proposes the deconvolution method for separating the EEG from different components.…”

“…Therefore, we utilize the first coefficient by Cepstrum calculation. As a result, the number of outputs of this step is the same as the previous steps [ 15 , 24 ].…”

“…In [ 13 , 14 ], the Cepstrum method is proposed to diagnose seizures, to be used on EEG and Intracranial EEG (IEEG) multi-channel separately in short term, whose artifacts have been removed and which includes different patient states. According to the hypothesis [ 15 ], the nonlinear problem of EEG caused by signal convolution is solved by Mel-frequency Cepstrum for improving the detection efficiency. Method of [ 14 ] was improved in 2014 by calculating the signals Teager energy Cepstrum and then following the same procedure as before[ 13 ].…”

Interhemispheric Brain Switching Correlates with Severity of Sleep-Disordered Breathing for Obstructive Sleep Apnea Patients

“…The standard approach to diagnose OSA involves in-laboratory polysomnography (PSG), in which EEG remains the best technique for the functional imaging of the brain during sleep [8,9]. EEG is the most common tool used in sleep research [10][11][12][13][14][15][16]. Recently, EEG coherence analysis has garnered considerable interests to study the functional asymmetry of the brain [17][18][19][20][21][22].…”