Human Motion Recognition for Adaptive Deep Brain Stimulation Using a Head-mounted Triaxial Accelerometer.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Machine Learning

Background:

• Adaptive deep brain stimulation (aDBS) offers personalized Parkinson's disease (PD) treatment.
• Real-time motion detection is crucial for effective aDBS system adaptability.
• Head-mounted sensors present a potential low-power solution for motion detection.

Purpose of the Study:

• To assess the feasibility of using a head-mounted triaxial accelerometer for human motion detection.
• To develop and evaluate machine learning algorithms for classifying daily activities using head acceleration data.
• To establish a framework for feature analysis and algorithm selection for embedded aDBS systems.

Main Methods:

• Recruited 32 healthy participants performing activities: sitting, standing, walking, falling.
• Utilized a head-mounted triaxial accelerometer to capture motion data.
• Employed Decision Tree, Random Forest, and Support Vector Machine (SVM) algorithms for motion classification, including a "pre-subsequent motion" feature.

Main Results:

• Radial Basis Function (RBF) kernel SVM achieved over 80% classification accuracy.
• Distinguishing between sitting and standing remained challenging.
• Incorporating a "pre-subsequent motion" feature improved accuracy to over 96%.

Conclusions:

• Head-mounted accelerometers are feasible for detecting human motion for aDBS systems.
• Machine learning, particularly SVM with enhanced features, shows high potential for motion recognition.
• This framework supports the development of low-power, body-embedded adaptive deep brain stimulation devices for Parkinson's disease.