Tremor, a prevalent movement disorder encountered in 2–6% of neurological consultations, presents significant diagnostic challenges, particularly in differentiating Parkinsonian tremor (PD) from essential tremor (ET) due to overlapping clinical features. Existing methods, including clinical observation and neuroimaging, are limited by subjectivity, cost, and accessibility.
This study introduces a novel, objective, and efficient electrophysiological framework based on electromyographic (EMG) signal analysis to enhance the differentiation between PD and ET. EMG signals were retrospectively collected from two cohorts (France and Taiwan), comprising 285 recordings across 77 patients.
A custom signal processing pipeline was developed, involving empirical mode decomposition (EMD), signal rectification, envelope extraction, and a quantile-based segmentation algorithm to identify tremor bursts and interburst intervals. Temporal and amplitude-based features—such as burst duration, interburst duration, and burst amplitude variability—were extracted and statistically analyzed using linear mixed models to account for repeated measures and class imbalance.
PD patients exhibited longer interburst intervals and higher variability in burst duration, whereas ET signals were characterized by shorter, more consistent bursts. Frequency-based analyses (2–15 Hz) did not significantly distinguish the groups.
A neural network classifier trained on the French dataset was evaluated by predicting on the independent Taiwan cohort, achieving 80% accuracy and an AUC of 0.89. These results demonstrate the generalizability and clinical relevance of this approach across populations.
The proposed EMG-based method offers a robust, non-invasive, and time-efficient alternative to current diagnostic tools, reducing misclassification and improving early-stage tremor assessment.