LFSP-DSM: A Lightweight Framework for Seizure Prediction Based on Deep Statistical Model

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Epilepsy is a chronic neurological disorder marked by recurrent seizures.
• Traditional machine learning for epileptic seizure prediction faces challenges due to inconsistent EEG signal labels and large data volumes, leading to complexity and long prediction cycles.

Purpose of the Study:

• To develop a lightweight and efficient framework for epileptic seizure prediction.
• To enhance the predictive capability of electroencephalogram (EEG) signal analysis for epilepsy.

Main Methods:

• Proposed LFSP-DSM framework integrating a hybrid enhancement model (HEM) and deep statistical models.
• HEM enhances EEG signal features in spatial and temporal dimensions.
• Deep statistical model comprises StaM for online labeling and LCNet (lightweight CNN) for multilevel feature learning.

Main Results:

• LFSP-DSM achieved high performance metrics: 91% for seizure frequency, 86% for seizure timing, and 93.24% for accuracy.
• Demonstrated effectiveness in handling complex epileptic sequence data and improving prediction performance.
• Validated the framework's ability to capture intricate signal patterns.

Conclusions:

• LFSP-DSM offers an effective solution for epileptic seizure prediction, overcoming limitations of traditional methods.
• The framework's lightweight design and enhanced feature extraction contribute to improved prediction accuracy and efficiency.
• Successfully addresses challenges in EEG signal analysis for epilepsy management.