Comparative Motor Pre-clinical Assessment in Parkinson's Disease Using Supervised Machine Learning Approaches

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Parkinson's disease (PD) significantly impacts quality of life, with diagnosis relying on motor symptoms.
• Non-motor symptoms, like idiopathic hyposmia (IH), are increasingly recognized for their association with PD risk.
• IH in healthy adults is linked to a higher likelihood of developing PD.

Purpose of the Study:

• To evaluate a wearable inertial device (SensFoot V2) for objective motor data acquisition.
• To differentiate between healthy individuals, individuals with IH, and PD patients using machine learning.
• To explore a non-invasive, two-step approach for early PD detection.

Main Methods:

• Collected motor data from 90 participants (30 healthy, 30 IH, 30 PD) using SensFoot V2 during MDS-UPDRS III lower limb tasks.
• Extracted significant, non-correlated motor parameters into a feature array.
• Applied three supervised machine learning algorithms for comparative classification analysis.

Main Results:

• The system achieved high accuracy (specificity and recall = 0.967) in distinguishing healthy individuals from PD patients.
• The machine learning model successfully identified IH as a distinct class with 0.78 accuracy.
• Identified key motor parameters that differentiate the three study groups.

Conclusions:

• The SensFoot V2 system shows potential as an objective tool to support clinical PD assessment.
• Combining IH identification with motor parameter analysis offers a promising non-invasive strategy for early PD detection.
• Wearable sensor technology can aid in the early identification of neurodegenerative conditions.