Edge-Enabled Pre-Ictal Activity Prediction Framework Using Geometric Deep Learning.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Technology

Background:

• Epileptic seizures pose significant challenges for patient management and quality of life.
• Accurate and timely prediction of pre-ictal states is crucial for effective intervention.
• Existing deep learning techniques for seizure prediction often lack interpretability and real-time edge deployment capabilities.

Purpose of the Study:

• To introduce an edge-enabled pre-ictal phase prediction framework for epileptic seizures.
• To evaluate the performance of the Brain Network Transformer (BNT) for analyzing EEG-based brain connectivity.
• To demonstrate the suitability of BNT for real-time deployment on edge devices.

Main Methods:

• Utilized the Brain Network Transformer (BNT) model for analyzing electroencephalogram (EEG) signals.
• Employed geometric deep learning techniques focusing on EEG-based brain connectivity.
• Compared BNT performance against various existing deep learning techniques on the CHB-MIT dataset.
• Benchmarked computational efficiency on an Nvidia Jetson Xavier NX edge device.

Main Results:

• BNT achieved high accuracy (average 97.17%, median 98.51%) and an AUC score of 0.99 on the CHB-MIT dataset.
• The model demonstrated a mean inference time of 9.978 ms on an edge device, indicating real-time capability.
• BNT outperformed existing techniques in pre-ictal activity prediction.

Conclusions:

• The BNT framework provides an accurate, interpretable, and computationally efficient solution for real-time epileptic seizure prediction.
• Edge-device compatibility makes the system suitable for practical deployment in clinical and wearable monitoring systems.
• This technology has the potential to significantly improve patient outcomes through timely intervention and enhanced quality of life.