Transfer Learning-based Seizure Detection on Multiple Channels of Paediatric EEGs

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Epilepsy affects millions globally, with diagnosis often relying on time-consuming EEG analysis.
• Automated seizure detection in pediatric EEG is crucial but limited by deep learning's computational demands.
• Transfer learning offers a solution to develop efficient pediatric seizure detection models.

Purpose of the Study:

• To develop and evaluate a transfer learning-based method for automated seizure detection in multi-channel pediatric EEGs.
• To address the limitations of traditional deep learning methods in terms of computational cost and development time.
• To improve the efficiency and accuracy of seizure detection in pediatric EEG analysis.

Main Methods:

• Utilized the publicly available CHB-MIT EEG dataset, splitting it into training, validation, and testing sets.
• Employed spectrograms from 10s EEG signals with 5s overlap as input for three pre-trained transfer learning models: ResNet50, VGG16, and InceptionV3.
• Ensured test set independence by separating children into distinct training and testing groups.

Main Results:

• The transfer learning method achieved 85.41% accuracy, 85.94% recall, and 85.49% precision on the independent pediatric EEG test set.
• Demonstrated the effectiveness of transfer learning in analyzing complex pediatric EEG data.
• The model successfully processed multi-channel EEG data for seizure detection.

Conclusions:

• The developed transfer learning method shows significant potential for assisting researchers and clinicians in automated pediatric EEG seizure analysis.
• This approach can help reduce the time and resources required for expert EEG interpretation.
• Highlights the utility of transfer learning for computationally intensive tasks in neurodiagnostics.