Characterizing Brain Signals for Epileptic Pre-ictal Signal Classification

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Epilepsy is a neurological disorder defined by recurrent seizures.
• Characterizing pre-ictal brain activity is vital for early detection but computationally challenging.
• Advances in deep learning offer new possibilities for analyzing complex brain signals.

Purpose of the Study:

• To classify pre-ictal signals in epilepsy patients.
• To characterize brain waveforms during the pre-ictal period for early seizure detection.
• To develop a robust machine learning model mitigating patient-to-patient variability.

Main Methods:

• Developed a novel machine learning model named Pre-ictal Signal Classification (PiSC).
• Implemented a unique preprocessing procedure for stereo-electroencephalography (sEEG) signals.
• Utilized a deep learning framework integrating deep neural networks and meta-learning.

Main Results:

• PiSC demonstrated improved accuracy and F1 score by 10% compared to existing models.
• The model effectively mitigated patient-to-patient variances and allowed for fine-tuning.
• Identified two novel sEEG patterns associated with seizure development in nocturnal epilepsy.

Conclusions:

• PiSC offers a promising solution for computationally challenging pre-ictal signal classification.
• The developed deep learning framework enhances model stability and generalization for sEEG data.
• The identified sEEG patterns contribute to a better understanding of seizure onset mechanisms.