Machine Learning Approach for Quantifying Hereditary Cerebellar Ataxia Severity and Evaluating Rehabilitation...

Area of Science:

• Neurology
• Biomedical Engineering
• Rehabilitation Science

Background:

• Ataxia diagnosis and treatment monitoring traditionally rely on subjective clinical assessments.
• Objective, quantitative methods are needed for precise evaluation of movement coordination disorders like Hereditary Cerebellar Ataxia (HCA).

Purpose of the Study:

• To establish an objective framework for measuring HCA severity using motion data.
• To assess the efficacy of rehabilitation interventions through quantitative analysis of patient movement.
• To develop and validate machine learning models for ataxia assessment.

Main Methods:

• Collected motion data from 69 individuals with ataxia using Inertial Measurement Unit (IMU) sensors during truncal sitting and Romberg standing tests.
• Extracted clinically relevant features guided by established scales (BBS, FIST).
• Developed and evaluated ten supervised machine learning models, comparing their performance against clinical scores.

Main Results:

• The KNeighbors regression model achieved a significant correlation (0.7613, p=0.01) with the FIST score for sitting balance using back sensor data.
• The Lasso Regression model showed a significant correlation (0.5641, p=0.04) with the Berg Balance Scale (BBS) score during standing using back sensor data.
• The framework successfully distinguished between rehabilitation and control groups in both sitting and standing tasks (p≤0.05).

Conclusions:

• Motion data combined with machine learning offers a reliable and objective method for assessing ataxia severity and rehabilitation outcomes.
• This data-driven approach can complement or potentially replace subjective clinical evaluations, improving treatment precision.
• The developed framework demonstrates potential for widespread clinical application in neurological rehabilitation.