Objective: Generating a set of significant neurophysiological features to build a machine learning based diagnostic system for hyperkinetic movement disorders.
Background: Hyperkinetic movement disorders cause excessive, involuntary movements. It is difficult to phenotype these disorders as they have overlapping symptoms and patients may have multiple movement disorders at the same time. Phenotyping hyperkinetic movement disorders is based on the clinical presentation and on neurophysiological features extracted from sensor data. The literature defines a wide range of features extracted from sensor data that describe typical signs of movement disorders. Currently, it is unknown which feature-combinations best distinguish between different hyperkinetic movement disorders.
Method: We collected data from 27 essential tremor, 14 myoclonus, and 21 dystonia patients using accelerometers (ACC) measuring the acceleration of motion and electromyography (EMG) measuring muscle activity. Based on this data, we constructed a set of more than 45,000 phenotypic features including features previously considered in the literature and novel data-driven features that may help distinguishing between these disorders. Using relevance learning, we constructed machine learning models for disorder classification and analysed the contribution of each feature to the decision-making process. Our methods identify features that are relevant for the distinction of hyperkinetic movement disorders and reduce the feature set so that only significant features remain.
Results: We trained a classifier on a subset of frequency domain features extracted from the ACC data. Preliminary results show that the classifier can discriminate the two classes essential tremor and myoclonus with an AUC(ROC) of up to 0.9. In particular, features extracted from a postural task were significant in this setting.
Conclusion: The study leads to a comprehensible and transparent system providing new insights into which characteristic features define hyperkinetic movement disorders.
To cite this abstract in AMA style:E. V.D. Brandhof, S. V.D. Veen, G. Russo, I. Tuitert, J. Dalenberg, M. V.D. Stouwe, J. Elting, M. Biehl, M. de Koning-Tijssen. Towards a Machine Learning Based Classification System for Hyperkinetic Movement Disorders: Generating a Neurophysiological Feature Set [abstract]. Mov Disord. 2022; 37 (suppl 2). https://www.mdsabstracts.org/abstract/towards-a-machine-learning-based-classification-system-for-hyperkinetic-movement-disorders-generating-a-neurophysiological-feature-set/. Accessed September 23, 2023.
« Back to 2022 International Congress
MDS Abstracts - https://www.mdsabstracts.org/abstract/towards-a-machine-learning-based-classification-system-for-hyperkinetic-movement-disorders-generating-a-neurophysiological-feature-set/