Contrastive Learning Model for Wearable-based Ataxia Assessment.

Area of Science:

• Biomedical Engineering
• Neurology
• Machine Learning

Background:

• Objective assessment of ataxia severity is crucial for disease management and treatment evaluation.
• Current wearable-based methods often rely on subjective clinical scales, introducing potential biases.
• There is a need for more flexible and objective feature extraction from wearable inertial data.

Purpose of the Study:

• To introduce a novel contrastive learning-based model for ataxia severity assessment using wearable inertial data.
• To leverage motor severity differences to learn robust and relevant features.
• To reduce reliance on imperfect clinical rating scales for feature development.

Main Methods:

• Trained a contrastive learning model on inertial data from 87 ataxia patients and 44 healthy controls performing the finger-to-nose task.
• Utilized a pairwise contrastive loss function to learn representations of relative ataxia severity.
• Evaluated learned features through downstream regression, classification, and reliability analyses.

Main Results:

• Learned features showed strong cross-sectional (r=0.84) and longitudinal (r=0.68) correlations with clinical scores.
• Demonstrated robust measurement reliability (ICC=0.96).
• Achieved high accuracy in distinguishing between ataxia and healthy groups (AUC=0.95).

Conclusions:

• The contrastive model effectively captures ataxia severity, outperforming methods reliant on direct clinical score feature extraction.
• This approach offers a more objective and scalable method for ataxia assessment.
• The findings suggest potential for enhanced patient monitoring and clinical trial efficiency.