Background: Essential tremor (ET) is among the most common neurological disorders, requiring precise diagnosis and severity assessment for personalized and effective management.
Methods: This study explores an innovative approach to evaluate ET severity using the gold-standard Archimedes spiral test. The family-based dataset covers the entire range of tremor severity, from very mild (level 1) to advanced stages, offering a valuable resource for studying early diagnosis and tracking disease progression. The proposed method introduces a machine learning pipeline that combines Principal Component Analysis (PCA), linear discriminant analysis (LDA), and support vector machines (SVMs) to classify ET severity based on Archimedean spiral radius data.
Results: By incorporating the Fahn-Tolosa-Marin Tremor Rating Scale (FMT-TRS), the pipeline effectively distinguishes between tremor presence and severity. Its robustness was demonstrated through rigorous cross-validation and tests involving Gaussian noise perturbations.
Conclusions: These results underscore the machine learning-based pipeline's potential as a non-invasive and trustworthy diagnostic tool for clinical use and telemedicine applications. Moreover, the combination of geometric features, FMT-TRS scores, clinically oriented evaluation metrics, and classical statistical and machine learning models offers a robust, interpretable, explainable, and clinically meaningful analytical framework.
Keywords: classification algorithms; essential tremor; handwriting analysis; linear discriminant analysis; machine learning; personalized medicine; principal component analysis; support vector machines.