Phase-Amplitude Coupling Features Accurately Classify Multiple Sub-States Within a Seizure Episode.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Technology

Background:

• Epilepsy seizures often lead to a postictal EEG suppression state (PGES).
• Automated detection of seizure and postictal states is crucial for early warning, intervention, and identifying complications like status epilepticus and sudden unexpected death in epilepsy (SUDEP).

Purpose of the Study:

• To determine if ictal and postictal states in epilepsy can be reliably differentiated using EEG.
• To develop a machine learning model for classifying seizure progression and postictal states.

Main Methods:

• Analyzed 52 intracranial and scalp EEG seizure records from 19 patients.
• Calculated phase-amplitude cross-frequency coupling for each EEG recording.
• Utilized a convolutional neural network model trained on frequency coupling data.

Main Results:

• The convolutional neural network model achieved a mean accuracy of 0.89±0.09 in differentiating ictal and postictal states.
• The model correctly classified seizure recordings from sudden unexpected death in epilepsy (SUDEP) patients as primarily interictal (70%) and postictal EEG suppression state-like (26%).
• Results indicate that SUDEP seizures predominantly occur in postictal states without typical ictal sub-state evolution.

Conclusions:

• Frequency coupling markers combined with machine learning reliably identify ictal and postictal sub-states in epilepsy.
• This approach offers potential for novel monitoring and management strategies in epilepsy care.
• The findings highlight distinct seizure characteristics in sudden unexpected death in epilepsy (SUDEP) patients.