Phase portrait for high fidelity feature extraction and classification: A surrogate approach

Area of Science:

• Medical Physics
• Biomedical Engineering
• Signal Processing

Background:

• Auscultation is crucial for diagnosing respiratory conditions.
• Classifying lung sounds like bronchial breath (BB) and pleural rub (PR) is challenging.
• Traditional methods often overlook the complex, nonlinear dynamics of breath sounds.

Purpose of the Study:

• To develop and validate a novel surrogate method for classifying breath sound signals.
• To utilize phase portrait analysis and principal component analysis (PCA) for enhanced lung sound classification.
• To compare the efficacy of phase portrait features versus power spectral density (PSD) features in PCA-based classification.

Main Methods:

• Extracted nonlinear parameters (Lyapunov exponent, sample entropy, fractal dimension, Hurst exponent) from phase portraits of breath sound signals.
• Analyzed 39 breath sound signals (BB and PR) using spectral, fractal, and phase portrait techniques.
• Classified signals using PCA, comparing feature sets derived from phase portraits and PSD data.

Main Results:

• Phase portrait analysis revealed bronchial breath sounds exhibit higher complexity and randomness (higher fractal dimension and sample entropy) than pleural rub sounds.
• PCA based on phase portrait features achieved 89.6% classification accuracy for BB and 80.5% for PR.
• Phase portrait-based PCA demonstrated higher fidelity by capturing temporal correlations, outperforming PSD-based PCA.

Conclusions:

• Phase portrait analysis provides a more comprehensive understanding of breath sound complexity.
• The proposed PCA method leveraging phase portrait features offers a superior approach for classifying bronchial breath and pleural rub sounds.
• This advanced classification technique holds promise for improving diagnostic accuracy in respiratory auscultation.