Adaptive regression model for Parkinson's disease diagnosis from speech signals using Box-Cox-based clustering and...

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Data Science

Background:

• Parkinson's Disease (PD) is a neurodegenerative disorder impacting motor and cognitive functions, with speech impairments serving as early biomarkers.
• Current speech-based PD monitoring faces challenges like feature redundancy, non-Gaussian data, and inadequate feature grouping, limiting diagnostic accuracy.
• Voice analysis offers a promising, accessible method for remote PD assessment and tracking disease progression.

Purpose of the Study:

• To develop an adaptive framework for precise PD diagnosis using biomedical voice measurements.
• To predict Motor Unified Parkinson's Disease Rating Scale (UPDRS) and Total-UPDRS scores from voice data.
• To overcome limitations in existing speech-based PD prediction methods through advanced feature engineering and machine learning.

Main Methods:

• Implemented a three-component framework: Box-Cox transformation for data normalization, K-Means clustering with mutual information for feature selection, and an Extra Trees Regressor (ETR) for prediction.
• Employed a subject-independent data splitting strategy and k-fold cross-validation for robust model evaluation.
• Compared the proposed method against various feature selection techniques and regression models.

Main Results:

• The proposed clustering-based feature selection combined with ETR achieved high performance, with R-squared scores of 0.999 for Motor-UPDRS and 0.997 for Total-UPDRS on the test set.
• The framework effectively addressed feature redundancy and non-Gaussian data distributions, enhancing diagnostic precision.
• Cross-validation and feature importance analyses confirmed the robustness and effectiveness of the developed approach.

Conclusions:

• The study demonstrates a highly effective and robust framework for speech-based Parkinson's Disease telemonitoring.
• The proposed method significantly improves PD diagnostic precision by optimizing feature selection and utilizing advanced regression techniques.
• This approach holds potential for accurate, remote, and early detection of Parkinson's Disease progression.