Discriminating Between Patients With Parkinson's and Neurological Diseases Using Cepstral Analysis

Area of Science:

• Neurology
• Biomedical Engineering
• Signal Processing

Background:

• Parkinson's disease (PD) diagnosis can be challenging.
• Objective diagnostic tools are needed to complement clinical assessments.
• Voice analysis offers a non-invasive method for neurological disorder detection.

Purpose of the Study:

• To discriminate between patients with Parkinson's disease and those with other neurological disorders using voice features.
• To evaluate the effectiveness of different Cepstral techniques for voice analysis in neurological conditions.
• To identify optimal feature sets and classifiers for accurate PD detection.

Main Methods:

• Collected voice samples from 50 subjects (PD and other neurological disorders).
• Extracted acoustic features using Mel frequency cepstral coefficients (MFCC), perceptual linear prediction (PLP), and ReAlitive SpecTrAl PLP (RASTA-PLP).
• Employed leave-one-subject-out cross-validation with five supervised learning classifiers.

Main Results:

• The best classification performance achieved 90% accuracy.
• Optimal results were obtained using the first 11 coefficients of Perceptual Linear Prediction (PLP).
• Linear Support Vector Machine (SVM) kernels demonstrated high efficacy.

Conclusions:

• Voice analysis, particularly using PLP features, is a promising method for discriminating Parkinson's disease.
• The findings support the development of objective, non-invasive diagnostic tools for neurological disorders.
• Further research can refine these methods for broader clinical application.