Empirical Wavelet Transform Based Features for Classification of Parkinson's Disease Severity.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Neuroscience

Background:

• Parkinson's disease (PD) is a progressive neurodegenerative disorder with symptoms like tremor and rigidity.
• Current diagnostic methods often use binary classification, failing to distinguish PD severity levels.
• There is a need for advanced systems to accurately detect and classify PD stages.

Purpose of the Study:

• To propose a multiclass classification system for Parkinson's disease severity (mild, moderate, severe) and healthy controls.
• To utilize signals from wearable motion and audio sensors for PD detection and classification.
• To evaluate the effectiveness of Empirical Wavelet Transform (EWT) and Empirical Wavelet Packet Transform (EWPT) in PD analysis.

Main Methods:

• Applied EWT and EWPT to decompose speech and motion sensor data up to five levels.
• Extracted signal features using the Hilbert transform on decomposed signal coefficients.
• Analyzed performance using K-nearest neighbour (KNN), probabilistic neural network (PNN), and extreme learning machine (ELM) classifiers.

Main Results:

• Achieved over 90% classification accuracy using EWT/EWPT-ELM with individual motion or audio sensors.
• Demonstrated the ability to differentiate PD subjects from non-PD subjects, including PD severity levels.
• Attained over 95% classification accuracy by integrating information from both motion and audio sensors with EWT/EWPT-ELM.

Conclusions:

• The proposed EWT/EWPT-ELM approach effectively classifies Parkinson's disease severity using wearable sensor data.
• Integrating motion and audio sensor data significantly enhances classification accuracy for PD detection.
• This method offers a promising tool for early and accurate diagnosis of Parkinson's disease stages.