Ultrasound-Enhanced Data-Driven Modeling for Characterizing Natural Wrist Tremor Dynamics

Area of Science:

• Biomedical Engineering
• Neurology
• Robotics

Background:

• Involuntary rhythmic wrist tremors present challenges in symptom management due to difficulties in identifying key characteristics like dominant frequency and amplitude.
• Existing methods often rely on inertial measurement units (IMUs) alone, which cannot capture internal muscle dynamics crucial for understanding tremor generation.

Purpose of the Study:

• To develop and validate an ultrasound (US)-enhanced data-driven modeling framework for improved wrist tremor characterization and tracking.
• To explore different modeling configurations, including a real-time adaptive model, for dynamic tremor behavior analysis.
• To establish a foundation for personalized tremor modeling and model-based control strategies.

Main Methods:

• A data-driven modeling framework was developed using wrist angle kinematics as state variables, augmented with muscle-specific ultrasound (US) data.
• Three models were explored: kinematics-only, fixed-parameter US-augmented, and a real-time adaptive US-enhanced model using a recursive least squares (RLS) algorithm.
• Comprehensive validation involved time- and frequency-domain analyses on experimental data from six patients with tremor, including eigenvalue analysis.

Main Results:

• Integrating US input significantly improved modeling accuracy, reducing time-domain normalized root mean square error (nRMSE) from 49.36% (baseline) to 24.53% (US-augmented).
• The adaptive US-enhanced model achieved a remarkable error reduction to 0.48%, demonstrating dynamic adaptation to tremor variability.
• The framework enabled accurate estimation of dominant tremor frequencies (nRMSE = 12.7%), quantification of state contributions, and reliable tremor event detection (F1-score = 0.796).

Conclusions:

• The US-enhanced data-driven framework accurately reproduces tremor trajectories and reveals evolving tremor behavior linked to neuromuscular activity.
• This approach offers a promising method for personalized tremor modeling, moving towards real-time tracking and model-based interventions like closed-loop stimulation.
• Leveraging ultrasound's sensing capabilities provides a novel pathway for advanced tremor management and control.