Discriminating progressive supranuclear palsy from Parkinson's disease using wearable technology and machine learning

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Distinguishing progressive supranuclear palsy (PSP) from idiopathic Parkinson's disease (PD) is clinically challenging.
• Accurate and early discrimination is crucial for timely administration of disease-modifying therapies for PSP.
• Gait and related task analysis offers a potential method for differentiating these conditions.

Purpose of the Study:

• To investigate the efficacy of a wearable sensor array combined with machine learning for classifying PSP.
• To determine the ability of these methods to discriminate PSP from PD and healthy controls (HC).

Main Methods:

• Participants (21 PSP, 20 PD, 39 HC) wore six inertial measurement units.
• Tasks included a two-minute walk, static sway test, and timed up-and-go.
• Data were analyzed using Logistic Regression (LR) and Random Forest (RF) algorithms.

Main Results:

• Seventeen independent features were identified for discrimination.
• Random Forest (RF) achieved 86% sensitivity and 90% specificity for PSP vs. PD.
• RF achieved 90% sensitivity and 97% specificity for PSP vs. HC.
• Reduced sensor arrays showed modest accuracy loss, recoverable with 3 sensors; maximum specificity required the full six-sensor array.

Conclusions:

• Wearable sensors and machine learning accurately discriminate PSP from PD.
• The complexity of the sensor array should be context-dependent.
• Full arrays are advantageous for high specificity in diagnostics, while simpler systems may suffice for disease tracking.