Fast processing and classification of epileptic seizures based on compressed EEG signals

Area of Science:

• Biomedical Engineering
• Signal Processing
• Neurology

Background:

• Epilepsy diagnosis from electroencephalogram (EEG) signals is complex and error-prone due to signal variability and volume.
• Developing portable, real-time epilepsy diagnostic systems faces challenges in signal processing and accurate classification.

Purpose of the Study:

• To propose and evaluate compressive sensing (CS) for efficient EEG signal condensation and seizure classification.
• To demonstrate the feasibility of a portable, embedded system for real-time epilepsy diagnosis.

Main Methods:

• Compressive sensing (CS) using discrete cosine transform (DCT) and random matrix multiplication for EEG signal compression (5-70% ratios).
• Feature selection based on mutual information and correlation matrix.
• XGBoost machine learning model for seizure classification.
• Implementation on STM32 microcontroller and Raspberry Pi for embedded system demonstration.

Main Results:

• Achieved 98.78% classification accuracy with XGBoost.
• At 70% compression: 70% file size reduction, 84% decrease in transmission time, and substantial energy savings.
• Maintained signal quality with PSNR of 16.15±3.98 and Pearson correlation coefficient of 0.68±0.15.

Conclusions:

• Compressive sensing offers an efficient method for reducing EEG data for portable, real-time epilepsy diagnosis.
• The proposed system enables precise, automated seizure classification with significant improvements in efficiency and energy consumption.