Memory-Efficient Intrinsic Gating Adaptation for Enhanced On-Device Epilepsy Diagnosis

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Epilepsy diagnosis relies on Electroencephalogram (EEG) but faces challenges due to patient-specific biomarker variability and resource constraints on edge devices.
• Conventional deep learning models struggle to adapt to changing biomarkers over time, leading to decreased diagnostic performance.
• On-device learning for epilepsy diagnosis is hindered by computational and memory limitations.

Purpose of the Study:

• To introduce a novel framework, Memory-Efficient Intrinsic Gating Adaptation (MEIGA), for efficient and accurate epilepsy diagnosis on resource-constrained edge devices.
• To address session-to-session variability in EEG biomarkers through lightweight adapter networks for on-device tuning.
• To reduce memory usage and computational overhead while maintaining high classification accuracy in seizure detection and prediction.

Main Methods:

• Pre-training a model on historical EEG data and employing lightweight adapter networks for efficient on-device tuning.
• Utilizing Direct Feedback Alignment (DFA) to minimize memory and computational requirements.
• Evaluating the framework on the CHB-MIT and AES epilepsy datasets.

Main Results:

• MEIGA significantly improved seizure prediction accuracy from 47.88% to 86.77% with minimal tunable parameters (5.05% of the backbone).
• Seizure detection accuracy increased from 85.06% to 96.29% by adapting only 17.40% of the base architecture.
• The framework demonstrated consistent performance across subjects and scalability on the AES dataset.

Conclusions:

• MEIGA offers an effective solution for real-world epilepsy diagnosis on edge devices, overcoming limitations of traditional deep learning models.
• The proposed method enhances diagnostic accuracy and efficiency by adapting to individual patient biomarker changes.
• MEIGA presents a promising approach for advancing accessible and reliable epilepsy monitoring systems.