Explainable machine learning for movement disorders - Classification of tremor and myoclonus.

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Essential tremor (ET) and cortical myoclonus (CM) share symptoms, making clinical distinction challenging.
• High inter- and intra-observer variability necessitates improved diagnostic tools for ET and CM.

Purpose of the Study:

• To develop a machine learning (ML) model utilizing accelerometry data for differentiating ET from CM.
• To provide clinicians with an objective tool to aid in the diagnosis of ET versus CM.

Main Methods:

• Collected upper body movement data from 19 ET and 19 CM patients across 21 tasks using eight accelerometry sensors.
• Applied explainable ML (Generalized Matrix Learning Vector Quantization - GMLVQ) to power spectrum analysis of accelerometry recordings for classification.
• GMLVQ identified relevant frequency patterns contributing to phenotype classification.

Main Results:

• Achieved excellent classification performance (AUROC approaching 1.0) for both static and dynamic tasks.
• GMLVQ identified key frequency bands (5-7 Hz, 3-4 Hz, 9-10 Hz) crucial for distinguishing ET and CM.
• These frequencies align with known tremor peaks and spectral characteristics in the literature.

Conclusions:

• Demonstrated proof of concept for using GMLVQ analysis of accelerometry power spectra to discriminate ET and CM.
• The developed ML approach shows potential to enhance diagnostic accuracy for clinicians treating ET and CM.