A deep learning framework for epileptic seizure detection based on neonatal EEG signals

Area of Science:

• Neuroscience
• Medical Informatics
• Computational Biology

Background:

• Epilepsy diagnosis relies heavily on electroencephalogram (EEG) interpretation, a process typically manual and time-consuming.
• Automating epileptic activity detection using conventional and Deep Learning (DL) methods is an active research area.
• A lack of detailed, reproducible methodologies hinders the advancement of automated EEG analysis.

Purpose of the Study:

• To develop and share a complete, reproducible framework for automated seizure detection in neonatal EEG using Deep Learning.
• To address the gap in detailed methodology for DL-based EEG analysis.
• To provide accessible R and Python code for the entire seizure detection pipeline.

Main Methods:

• Utilized 79 neonatal EEG recordings for framework development.
• Implemented a Deep Learning approach, specifically a Convolutional Neural Network (CNN).
• Developed code for data ingestion (European Data Format), annotation reading, data splitting, model creation, training, evaluation, and result saving.

Main Results:

• A comprehensive framework for DL-based seizure detection in neonatal EEG was successfully developed.
• Ready-to-use R and Python scripts are provided, enabling reproducibility.
• The framework covers all stages from raw data processing to model performance assessment.

Conclusions:

• The developed framework offers a reproducible solution for automated seizure detection in neonatal EEG.
• Sharing the code promotes transparency and facilitates further research in DL for epilepsy diagnosis.
• This work contributes to improving the efficiency and accuracy of epilepsy diagnosis through automated analysis.