Tremor Detection Techniques in People with Parkinson's Disease

Abstract

Accurate and scalable tremor measurement is needed to complement clinical scale assessments, which are often episodic and subjective. This study investigated whether wrist inertial measurement units (IMUs) can provide objective tremor features during task-based movements. A MATLAB algorithm was developed to extract features in both time- and frequency domains, apply sensor fusion to minimise orientation dependence, and evaluate robustness under different walking speeds (slow, normal, fast) and reduced sampling frequencies (from 100 Hz to 30 Hz in 5 Hz steps). Analyses of laboratory data from Parkinson’s disease patients (PD, n=20) and healthy adults (HA, n=20) used cohort-level summaries and percent-change metrics relative to 100 Hz for down-sampling and to normal walking speed for speed comparisons. Across sampling rates, Dominant Frequency remained nearly stable and was consistently lower in PD than in HA, while amplitude- and energy-related features showed the largest distortions at lower sampling rates. At 100 Hz, walking speed modulated both amplitude and frequency features, with PD showing larger speed-dependent increases in amplitude and variability, while Power Dispersion remained generally higher in PD across speeds. These findings suggest that frequency features can be reliably obtained at moderately reduced sampling rates compatible with smartwatch limitations, whereas amplitude-based measures are more sensitive to both sampling rate and walking speed. Future research should include rest and postural tasks, clinical annotations, and bilateral sensing to improve clinical interpretability and reproducibility.