Classification of Seizure Termination Patterns using Deep Learning on intracranial EEG.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Technology

Background:

• Seizure termination is less understood than initiation and propagation.
• Studying termination can lead to better interventional epilepsy treatments.
• Identifying termination mechanisms in self-terminating seizures is crucial.

Purpose of the Study:

• To investigate temporal and spectral features of intracranial EEG (iEEG) during seizures.
• To identify time-frequency signatures that predict seizure termination patterns.
• To develop a deep learning model for classifying seizure termination patterns.

Main Methods:

• Decomposition of iEEG data into time-frequency maps using Morlet Wavelet Transform.
• Training a Convolutional Neural Network (CNN) on cross-patient time-frequency maps.
• Classification of seizure termination into burst suppression and continuous bursting patterns.

Main Results:

• The CNN achieved 90% accuracy and 92% precision in classifying seizure termination patterns.
• The proposed CNN model outperformed k-Nearest Neighbour (k-NN), which achieved 70% accuracy and 72% precision.
• The model effectively captured temporal and spatial patterns for high classification performance.

Conclusions:

• The developed deep learning model accurately classifies seizure termination patterns.
• This classification can predict seizure endings, informing epilepsy management and treatment strategies.
• The model offers a 90% accurate method to assist in the clinical treatment of epilepsy patients.