Prediction of Epilepsy Seizure Based on Cepstrum Analysis and Deep Learning

Area of Science:

• * Neuroscience
• * Signal Processing
• * Machine Learning

Background:

• * Electroencephalogram (EEG) signals exhibit distinct frequency component changes preceding epileptic seizures.
• * Analyzing EEG in the Mel frequency domain can enhance seizure-specific features for improved prediction.
• * Non-stationary nature of EEG necessitates adaptive analysis methods for accurate seizure prediction.

Purpose of the Study:

• * To propose an adaptive prediction model for epilepsy seizure using Mel-frequency analysis of EEG signals.
• * To investigate the integration of Mel-frequency cepstral coefficients (MFCC) and linear predictive coding cepstral coefficients (LPCC) for comprehensive EEG feature extraction.
• * To evaluate the model's performance against established machine learning algorithms.

Main Methods:

• * EEG signals processed using Mel-frequency cepstral coefficients (MFCC) and linear predictive coding cepstral coefficients (LPCC).
• * Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) integrated for advanced feature extraction from non-stationary EEG data.
• * Model validated on the CHB-MIT epilepsy EEG dataset and compared with Support Vector Machine, K-Nearest Neighbors, and other classifiers.

Main Results:

• * The proposed model achieved high prediction accuracy (94%), sensitivity (96%), and specificity (92%).
• * Integrated CNN-LSTM approach demonstrated superior performance in capturing seizure-related EEG dynamics compared to traditional methods.
• * Mel-frequency domain analysis effectively highlighted subtle changes indicative of impending seizures.

Conclusions:

• * The developed model demonstrates significant effectiveness in predicting epilepsy seizures.
• * Integrating MFCC, LPCC, CNN, and LSTM offers a robust approach for analyzing non-stationary EEG signals.
• * This method holds promise for improving clinical management and patient outcomes through early seizure detection.