Addressing voice recording replications for tracking Parkinson's disease progression

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Speech Science

Background:

• Tracking Parkinson's disease (PD) symptom severity using voice analysis is challenging due to within-subject variability in replicated voice recordings.
• Existing pattern recognition methods struggle with non-identical replicated features caused by technological imperfections and biological variations.
• The temporal aspect of symptom progression is crucial for accurate PD monitoring.

Purpose of the Study:

• To propose a novel Bayesian linear regression approach for analyzing voice recordings to track Parkinson's disease symptom severity.
• To address the challenge of within-subject variability in replicated voice measurements.
• To incorporate the temporal dynamics of symptom progression into the analysis.

Main Methods:

• Automatic extraction of voice features from multiple recordings of the same subjects.
• Development of a Bayesian linear regression model to handle replicated measurements and time-dependent features.
• Implementation of Gibbs sampling-based approaches to overcome computational difficulties.
• Introduction of a penalized regression version to select optimal predictors.

Main Results:

• The proposed Bayesian linear regression approach effectively handles within-subject variability in voice features.
• The model successfully incorporates the time factor for tracking symptom severity.
• The penalized version aids in identifying the most informative voice features for PD monitoring.
• Gibbs sampling provided a computationally feasible solution for the complex model.

Conclusions:

• A novel Bayesian linear regression method is presented for tracking Parkinson's disease symptom severity from voice recordings.
• The approach accounts for within-subject variability and temporal changes, offering a more robust analysis.
• This methodology provides a promising tool for objective and continuous monitoring of PD progression.