A multi-dimensional CNN-Bi-GRU for IoT-based brain-computer interface in early epileptic seizure detection

Area of Science:

• Biomedical Engineering
• Neuroscience
• Artificial Intelligence

Background:

• Epilepsy affects millions globally, necessitating advanced seizure detection methods.
• Current systems often lack real-time capabilities and personalized support.
• Electroencephalogram (EEG) data offers a rich source for analyzing brain activity during seizures.

Purpose of the Study:

• To develop and evaluate a high-accuracy deep learning model for early seizure detection.
• To design an integrated Internet of Things-Brain-Computer Interface (IoT-BCI) system for real-time patient support.
• To compare the proposed model's performance against existing deep learning and machine learning approaches.

Main Methods:

• Utilized EEG data from Mendeley for model training and validation.
• Developed a Multi-Dimensional CNN-Bi-GRU (MDCBG) hybrid deep learning architecture.
• Implemented and simulated an IoT-based automation system for triggering micro-devices based on seizure detection.
• Performed ablation studies and SHAP analysis to assess model contributions and feature importance.

Main Results:

• The MDCBG model achieved a peak accuracy of 97.43% in seizure detection.
• Outperformed baseline models such as EEGNet (92.17%) and CTNET (85.11%).
• Demonstrated strong performance in classifying various seizure types and identified Channel 5 as a key predictive feature.

Conclusions:

• The MDCBG model represents a significant advancement in automated, real-time seizure detection.
• The integrated IoT-BCI system shows promise for early warning and home-automation strategies to assist epilepsy patients.
• This approach offers a pathway to improved quality of life and safety for individuals with epilepsy.