Development of a Tremor Detection Algorithm for Use in an Academic Movement Disorders Center.

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Tremor is a key neurological symptom, often assessed visually, lacking standardized objective quantification.
• Current objective tremor quantification methods show promise but lack standardization across clinical centers.

Purpose of the Study:

• To evaluate machine learning pipelines for objective tremor detection in kinematic data.
• To compare algorithm performance against expert clinical ratings for movement disorders.

Main Methods:

• Utilized a database of 2272 kinematic recordings from patients with movement disorders.
• Compared six distinct processing pipelines using metrics like F1 score, accuracy, precision, and recall.
• A hybrid Support Vector Machine classifier incorporating engineered features was developed.

Main Results:

• All evaluated algorithms demonstrated comparable performance in tremor detection.
• The average F1 score across pipelines was 0.84 ± 0.02.
• A hybrid Support Vector Machine achieved a cross-validated F1 score of 0.87, indicating high accuracy.

Conclusions:

• Machine learning classifiers can enhance the performance of clinical decision support systems for tremor assessment.
• Modern machine learning offers potential for updating and improving objective tremor quantification tools.
• This study highlights the feasibility of integrating advanced computational methods into clinical practice for neurological movement disorders.