Loading…
FV 6 Phenotypical characterization of tremor syndromes using unbiased time-series feature analysis
Introduction: The reliable differentiation of tremor disorders remains a challenge and clinical diagnosis often depends on the subjective interpretation of subtle signs and symptoms. So far electrophysiological tests fail to capture the entirety of information in tremor data. Massive time series fea...
Saved in:
Published in: | Clinical neurophysiology 2022-05, Vol.137, p.e4-e4 |
---|---|
Main Authors: | , , , |
Format: | Article |
Language: | English |
Online Access: | Get full text |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | Introduction: The reliable differentiation of tremor disorders remains a challenge and clinical diagnosis often depends on the subjective interpretation of subtle signs and symptoms. So far electrophysiological tests fail to capture the entirety of information in tremor data. Massive time series feature extraction is a powerful tool to examine oscillating biological signals, such as tremor disorders, in an un-biased way. We previously showed that this method is feasible and can predict the response to non-invasive stimulation (1). This study is aiming to validate and extend our analytical pipeline examining higher-dimensional mathematical features extracted from accelerometer time-signal recordings from tremor patients. This approach will be used to differentiate tremor aetiologies, response to treatment and explore disease mechanisms.
Materials and Methods: Accelerometer recordings and corresponding phenotypical information from n=400 patients suffering from essential tremor (ET), dystonic tremor (DT), Parkinsońs disease (PD), functional tremor and monogenic tremor from six different European centres with specialist movement disorder units have been integrated into a singular data set. Sub-cohorts include repeat measurements and measurements under different treatments. Massive higher-order feature extraction is applied to perform supervised and unsupervised statistical learning. We are developing techniques for batch correction to control for centre-specific differences in the data.
Results: Raw accelerometer recordings from involved centres were cleaned, aligned, screened for movement artefacts and harmonised. After plausibility checks and quality control, we are establishing the robustness of feature extraction with regards to sampling frequency, applied filters and recording length. Data analysis involves the comparison of different machine learning algorithms for supervised and unsupervised tremor exploration.
Conclusions: With this work we aim to explore communalities and differences in tremor syndromes using unbiased and feature-based statistical learning in a clinically and methodologically robust way. To this end, the integration of accelerometer-based movement characteristics from different tremor entities and centres will allow a first objective glimpse at the tremor landscape. This aims to improve clinical diagnostic accuracy and expand our understanding of tremor aetiologies and the underlying mechanisms.
References
1. S. R. Schreglmann, D. Wa |
---|---|
ISSN: | 1388-2457 1872-8952 |
DOI: | 10.1016/j.clinph.2022.01.014 |