Efficient EEG Feature Learning Model Combining Random Convolutional Kernel with Wavelet Scattering for Seizure Detection | JoVE Visualize

Area of Science:

Biomedical Engineering
Computational Neuroscience
Medical Informatics

Background:

Epilepsy diagnosis and treatment rely heavily on accurate seizure detection.
Current deep learning models for electroencephalography (EEG) seizure detection face challenges in generalization and computational efficiency.
Developing lightweight and effective EEG feature learning models is crucial for practical clinical application.

Purpose of the Study:

To develop a novel, lightweight model for EEG feature learning for improved seizure detection.
To enhance the generalization performance and reduce the computational burden of automated seizure detection systems.
To create a comprehensive seizure detection system that is accurate and efficient for clinical use.

Main Methods:

A lightweight model integrating random convolutional kernel transform (ROCKET) with a wavelet scattering network was developed for EEG feature learning.
Significant EEG features were selected using analysis of variance (ANOVA) and minimum redundancy-maximum relevance (MRMR) methods.
The developed feature learning model was combined with an extreme gradient boosting (XGBoost) classifier for seizure detection.

Main Results:

The proposed model achieved over 90% sensitivity and specificity for epoch-based seizure detection on both scalp and intracranial EEG datasets.
The method demonstrated superior performance in both cross-patient and patient-specific seizure detection compared to state-of-the-art approaches.
The model effectively identified significant EEG features and retained only relevant channels, contributing to its efficiency and interpretability.

Conclusions:

The novel lightweight ROCKET-based model offers a promising solution for automated EEG seizure detection, addressing limitations of existing deep learning methods.
The developed system provides high accuracy and efficiency, making it suitable for practical epilepsy diagnosis and treatment.
This approach advances the field of automated seizure detection by offering a computationally efficient and highly effective tool.