CMEpiNet: Complex-Valued Multimodal Epilepsy Detection Network Model.

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Artificial Intelligence in Medicine

Background:

• Existing seizure detection methods struggle to leverage spatiotemporal features from multimodal signals.
• Current approaches fail to capture deep cross-modal feature associations, limiting unified representation learning.
• This hinders the effective modeling of complex spatiotemporal dependencies in epilepsy.

Purpose of the Study:

• To propose CMEpiNet (Complex-valued Multimodal Epilepsy detection Network model) for improved epilepsy detection.
• To address limitations in exploiting spatiotemporal features and cross-modal associations.
• To develop a unified representation of spatiotemporal dependencies using multimodal signals.

Main Methods:

• Utilized complex-valued convolutions for feature extraction, modeling phase synchronization and cross-frequency coupling.
• Represented EEG, ECG, and EMG features in the complex-valued domain.
• Employed a two-level semantic alignment-based fusion method with cross-modal and distribution-level alignment.
• Developed a spatial attention-guided 3D convolutional classifier for joint temporal, feature, and modality modeling.

Main Results:

• Demonstrated improved seizure detection sensitivity on the SeizeIT2 dataset.
• Significantly reduced the false alarm rate compared to existing methods.
• Maintained stable performance even under signal perturbations.

Conclusions:

• CMEpiNet effectively integrates multimodal signals for enhanced epilepsy detection.
• The complex-valued approach and advanced fusion techniques improve spatiotemporal dependency modeling.
• The proposed model offers a robust and accurate solution for clinical seizure detection.