A two-stage variable selection and classification approach for Parkinson's disease detection by using voice recording...

Area of Science:

• Biomedical Engineering
• Computational Statistics
• Speech Science

Background:

• Lack of variable selection and classification methods for replicated data in scientific literature.
• Need for improved discrimination between Parkinson's disease (PD) patients and healthy individuals using voice analysis.

Purpose of the Study:

• To develop and evaluate a novel two-stage variable selection and classification approach for replicated voice recordings.
• To discriminate individuals with Parkinson's disease (PD) from healthy subjects based on acoustic features.

Main Methods:

• A two-stage statistical approach designed for replication-based experimental data.
• Utilized a Gibbs sampling algorithm for efficient computation and variable selection.
• Applied to acoustic features extracted from replicated voice recordings.

Main Results:

• Achieved an acceptable predictive capacity for PD discrimination with a small sample size.
• Reported accuracy rate of 86.2%, sensitivity of 82.5%, and specificity of 90.0%.
• Demonstrated improved result interpretability, better computational efficiency (chain mixing, computation time) compared to existing methods.

Conclusions:

• This represents the first approach to incorporate intra-subject variability in variable selection and classification for replicated data.
• The method, applied to PD discrimination, is adaptable to other research areas with similar experimental designs.