An Efficient Epilepsy Prediction Model on European Dataset With Model Evaluation Considering Seizure Types.

Area of Science:

• Medical devices
• Computational neuroscience
• Biomedical engineering

Background:

• Epilepsy affects millions globally, necessitating advanced seizure prediction methods.
• Current prediction models often require significant computational resources, limiting their use in real-time applications.
• Intracranial electroencephalogram (iEEG) data offers high-resolution insights into neural activity preceding seizures.

Purpose of the Study:

• To develop a computationally efficient, patient-specific seizure prediction model.
• To adapt the model for potential use in wearable and implantable devices.
• To evaluate the model's performance on a large, diverse epilepsy dataset and analyze the impact of seizure types.

Main Methods:

• A two-layer Long Short-Term Memory (LSTM) network was employed for seizure prediction.
• Model efficiency was enhanced by reducing input size and the number of electrodes used.
• Automatic artifact removal using common average reference preprocessing was applied to the European iEEG dataset.
• The model was tested on data from 26 patients, analyzing performance across different seizure types.

Main Results:

• The developed LSTM model achieved a high average Area Under the Curve (AUC) of 0.885.
• The model's simplified structure and mean post-processing procedure yielded optimal performance.
• This research is the first to utilize the European iEEG database for epilepsy prediction.
• Seizure type was identified as a significant factor influencing prediction system performance.

Conclusions:

• The proposed LSTM model offers a computationally efficient solution for patient-specific seizure prediction.
• The model's design makes it suitable for integration into wearable and implantable epilepsy monitoring devices.
• Understanding the influence of seizure types is crucial for improving the accuracy and reliability of epilepsy prediction systems.