Deep Learning-Based Classification of Epileptic Electroencephalography Signals Using a Concentrated Time-Frequency...

Area of Science:

• Biomedical Signal Processing
• Neurology
• Machine Learning

Background:

• Epilepsy affects millions globally, causing unpredictable seizures.
• Electroencephalography (EEG) signals monitor brain activity and can potentially predict seizures.
• Accurate time-frequency (TF) analysis is crucial for interpreting complex biomedical signals like EEG.

Purpose of the Study:

• To evaluate the performance and robustness of the ConceFT (concentration of frequency and time) method for analyzing epileptic EEG signals.
• To present a signal classification algorithm utilizing ConceFT-derived TF images for epilepsy detection.
• To demonstrate the utility of ConceFT in conjunction with deep learning for biomedical applications.

Main Methods:

• Developed ConceFT, a novel TF analysis technique combining multitaper and synchrosqueezing transform (SST).
• Generated TF images from EEG signals using ConceFT.
• Employed GoogLeNet, a deep learning model, for classifying the TF images.

Main Results:

• ConceFT produced highly concentrated TF representations with excellent time and frequency resolutions.
• The classification algorithm achieved high accuracies, ranging from 95.83% to 99.58% for two- and three-class scenarios.
• Classification performance was directly correlated with the accuracy of ConceFT's TF representations.

Conclusions:

• ConceFT is a successful and promising TF analysis method for non-stationary biomedical signals like EEG.
• The proposed method demonstrates significant potential for accurate epilepsy detection through EEG signal classification.
• The integration of ConceFT with deep learning offers a robust approach for analyzing complex neurological data.