Fine Motor Assessment With Unsupervised Learning For Personalized Rehabilitation in Parkinson Disease

Area of Science:

• Biomedical Engineering
• Neuroscience
• Rehabilitation Technology

Background:

• Parkinson disease (PD) presents significant motor and non-motor impairments, diminishing patient quality of life.
• Accurate assessment of PD severity is critical for tailoring effective, personalized rehabilitation and treatment approaches.
• Current assessment methods may lack the precision needed for individualized care planning.

Purpose of the Study:

• To develop and validate a wearable system for objective motor performance assessment in Parkinson disease patients.
• To utilize unsupervised learning algorithms for classifying PD severity based on kinematic features.
• To enable precise evaluation of residual motor capabilities for personalized rehabilitation.

Main Methods:

• Two inertial measurement units captured upper and lower limb motion during six distinct motor tasks.
• Forty-one kinematic features were extracted from inertial sensor data to quantify motor performance.
• Unsupervised learning algorithms, including k-Means, Self-Organizing Maps (SOM), and hierarchical clustering, were employed for performance grouping.

Main Results:

• Self-Organizing Maps (SOM) demonstrated superior classification accuracy, achieving 0.950 for two severity classes (mild vs. advanced).
• SOM also achieved 0.817 accuracy when classifying three severity levels (mild vs. moderate vs. severe).
• The system successfully grouped motor performance data, reflecting objective PD severity.

Conclusions:

• The proposed wearable system offers an objective method for assessing Parkinson disease severity through motion analysis.
• This technology facilitates the evaluation of remaining motor functions, crucial for personalized PD care.
• The findings support the development of tailored rehabilitation and assistance strategies for individuals with Parkinson disease.