Category: Tremor
Objective: Providing a proof of concept for the classification of movement disorders using explainable machine learning applied to power spectra derived from accelerometry recordings.
Background: The clinical discrimination of the hyperkinetic movement disorders essential tremor (ET) and cortical myoclonus (CM) poses challenges due to their overlapping symptomatology, resulting in a large inter- and intra-observer variability. Accurate phenotyping is crucial to provide patients with the correct treatment as the movement disorders have distinct underlying pathophysiology. This study presents a machine learning based approach to support clinicians with diagnosing movement disorders.
Method: We classified the movement disorders based on the upper body movement recordings from 19 ET and 19 CM patients, measured with eight accelerometry sensors. Power spectra extracted from these recordings served as input for the explainable machine learning approach generalized matrix learning vector quantization (GMLVQ). Alongside the general performance measure, this method provides insights into relevant characteristics for distinguishing between the groups and, as a result, the recognition of unique phenotypic patterns.
Results: We reached excellent classification results for all postural and dynamic tasks, providing a proof of concept for the distinction of ET and CM by applying GMLVQ to the power spectra obtained from accelerometry recordings. Classification performance approaching AUC 1.0 were achieved. The frequencies between 5-7 Hz as well as 3-4 and 9-10 Hz, corresponding to the tremor peak and local minima in the power spectrum, emerged as crucial for classification.
Conclusion: This study provides a proof of concept for the classification of ET and CM based on power spectra derived from accelerometry recordings using machine learning. This system could assist clinicians in enhancing diagnostic accuracy and shorten the time to treatment.
To cite this abstract in AMA style:
E. vd Brandhof, I. Tuitert, A M. vd Stouwe, JW. Elting, J. Dalenberg, M. Biehl, M. Tijssen. Classification of Tremor and Myoclonus: An Explainable Machine Learning Approach [abstract]. Mov Disord. 2024; 39 (suppl 1). https://www.mdsabstracts.org/abstract/classification-of-tremor-and-myoclonus-an-explainable-machine-learning-approach/. Accessed November 23, 2024.« Back to 2024 International Congress
MDS Abstracts - https://www.mdsabstracts.org/abstract/classification-of-tremor-and-myoclonus-an-explainable-machine-learning-approach/