An Epileptic EEG Detection Method Based on Data Augmentation and Lightweight Neural Network.

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Epilepsy affects 65 million globally, with traditional detection methods being inefficient.
• Deep learning for brain signal detection faces challenges with data quality and computational demands.
• Advancing clinical applications requires efficient and accurate seizure detection models.

Purpose of the Study:

• To develop a highly accurate and computationally efficient deep learning model for epilepsy detection.
• To address limitations of traditional methods and current deep learning approaches in terms of data and resource requirements.
• To create a lightweight model suitable for deployment on resource-constrained hardware.

Main Methods:

• Merged Bonn University and CHB-MIT datasets for robust training.
• Utilized small window segmentation and Synthetic Minority Over-sampling Technique (SMOTE) to manage dataset size and class imbalance.
• Proposed a streamlined neural network architecture with significantly reduced training parameters.

Main Results:

• Achieved 98.52% accuracy, 97.99% sensitivity, 99.35% specificity, and 98.44% precision on a three-classification task.
• Developed a model with only 9,371 parameters, demonstrating substantial parameter reduction.
• The proposed method outperformed existing approaches in both model size and accuracy.

Conclusions:

• The developed lightweight neural network offers superior performance for epilepsy detection.
• The model's efficiency makes it ideal for deployment on low-cost hardware, including wearable technology.
• Enables real-time epileptic electroencephalogram (EEG) detection for clinical applications.