A Comparison of Machine Learning Classifiers for Energy-Efficient Implementation of Seizure Detection

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Closed-loop electrical stimulation is a novel epilepsy treatment.
• Energy-efficient seizure detection is crucial for implantable devices.
• Current detection methods require improvement.

Purpose of the Study:

• Evaluate machine learning algorithms for seizure detection.
• Compare algorithm performance against existing methods.
• Assess the energy efficiency of a Random Forest classifier for implantable devices.

Main Methods:

• Evaluated Random Forest and Support Vector Machine (SVM) classifiers.
• Utilized time and frequency domain features.
• Implemented Random Forest on a microcontroller for energy efficiency testing.

Main Results:

• Random Forest classifier outperformed SVM and the reference approach.
• The Random Forest implementation demonstrated superior detection sensitivity and specificity.
• Low detection delay was maintained.

Conclusions:

• The Random Forest classifier with selected features offers an energy-efficient solution.
• This approach enhances closed-loop epilepsy treatment.
• Improved seizure detection capabilities are achieved for implantable neurostimulation devices.