A hierarchical structure for human behavior classification using STN local field potentials.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Human behavior classification from brain signals is crucial for adaptive deep brain stimulation (DBS) systems.
• Local field potential (LFP) signals recorded from the subthalamic nucleus (STN) offer a promising avenue for this classification.

Purpose of the Study:

• To develop and evaluate a hierarchical classification framework for human behavior using STN LFP signals.
• To explore computational cost reduction strategies without compromising classification accuracy.

Main Methods:

• A multi-level hierarchical classification structure was implemented using LFP signals from DBS leads.
• Time-frequency representations were combined with a Multiple Kernel Learning-based Support Vector Machine (MKL-SVM) classifier.
• An inter-hemispheric synchronization approach was utilized to reduce computational load.

Main Results:

• The hierarchical approach using all six LFPs significantly improved classification performance.
• The synchronization-based method substantially reduced computational burden while maintaining classification accuracy.
• The proposed methods outperformed existing approaches in behavior classification based on LFP signals across two datasets and nine subjects.

Conclusions:

• STN LFP signals contain valuable information for human behavior recognition.
• This research provides a foundation for developing next-generation closed-loop DBS systems capable of real-time behavioral adaptation.