Contrastive Learning Model for Wearable-based Ataxia Assessment

medRxiv (Cold Spring Harbor Laboratory), Journal Year: 2025, Volume and Issue: unknown

Published: March 4, 2025

Abstract Objective Frequent and objective assessment of ataxia severity is essential for tracking disease progression evaluating the effectiveness potential treatments. Wearable-based assessments have emerged as a promising solution. However, existing methods rely on inertial data features directly correlated with subjective coarse clinician-evaluated rating scales, which serve imperfect gold standards. This approach may introduce biases restrict flexibility in feature design. To address these limitations, this study introduces novel contrastive learning-based model that leverages motor differences wearable to learn relevant features. Methods The was trained collected from 87 individuals diagnostically heterogeneous ataxias 44 healthy participants performing finger-to-nose task. A pairwise loss function proposed representations capturing relative severity, were evaluated through downstream regression classification tasks. Results learned demonstrated strong cross-sectional (r = 0.84) longitudinal 0.68) associations clinical scores robust measurement reliability (intraclass correlation coefficient 0.96). Additionally, exhibited known-group validity, distinguishing between phenotypes an area under receiver operating characteristic curve 0.95. Conclusion captures reduced reliance outperforming state-of-the-art derive scores. Significance Combining sensors learning enables more objective, scalable, frequent method assessing enhance patient monitoring improve trial efficiency.

Language: Английский

At-home wearables and machine learning capture motor impairment and progression in adult ataxias

medRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Oct. 29, 2024

A significant barrier to developing disease-modifying therapies for spinocerebellar ataxias (SCAs) and multiple system atrophy of the cerebellar type (MSA-C) is scarcity tools sensitively measure disease progression in clinical trials. Wearable sensors worn continuously during natural behavior at home have potential produce ecologically valid precise measures motor function by leveraging frequent numerous high-resolution samples behavior. Here we test whether movement-building block characteristics (i.e., submovements), obtained from wrist ankle home, can capture SCAs MSA-C, as recently shown amyotrophic lateral sclerosis (ALS) ataxia telangiectasia (A-T). Remotely collected cross-sectional (

Language: Английский