MultiEpilepsyNet: An EEG and MRI data based multimodal seizure detection model using hybrid deep learning model

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Imaging

Background:

• Epilepsy diagnosis requires accurate, privacy-preserving solutions.
• Current methods struggle with centralized data, generalizability, feature extraction, and noise.

Purpose of the Study:

• To introduce MultiEpilepsyNet, a multimodal seizure detection framework.
• To enhance privacy, accuracy, and generalizability in epilepsy diagnostics.

Main Methods:

• Federated learning (SeizureFed-Net) for privacy-preserving collaborative learning from EEG and MRI data.
• Hybrid deep learning (SeizureShieldNet) fusing BBIDNet and FD-TMS for robust seizure detection.
• Jackal-Wolf Hybrid Optimizer (JWHO) for optimal feature selection.
• EpiSkullNet++ for enhanced MRI brain segmentation.

Main Results:

• Achieved 99.36% accuracy on CHB-MIT EEG dataset and 99.38% on an epilepsy MRI dataset.
• Demonstrated improved robustness to missing modalities and reduced training overhead.
• Showcased enhanced privacy preservation compared to centralized models.

Conclusions:

• MultiEpilepsyNet effectively addresses critical barriers in clinical epilepsy diagnosis.
• The framework offers a scalable, privacy-preserving, and clinically viable solution.
• Highlights the potential of multimodal federated learning in neurological disorder management.