Multi-channel classification of respiratory sounds

Area of Science:

• Respiratory acoustics
• Biomedical signal processing
• Medical diagnostics

Background:

• Auscultation is a fundamental diagnostic technique.
• Objective analysis of respiratory sounds can enhance diagnostic accuracy.
• Distinguishing pathological from healthy respiratory patterns is clinically significant.

Purpose of the Study:

• To develop a diagnostic aid for pathological respiratory conditions using computational analysis of auscultation data.
• To compare the effectiveness of different signal processing techniques for respiratory sound classification.
• To build reference libraries for healthy and pathological respiratory sounds.

Main Methods:

• Respiratory sound data from 14 channels were collected from healthy and pathological subjects.
• Frequency spectrum and autoregressive (AR) model parameters were extracted.
• Reference libraries were created for inspiration and expiration phases.
• A K-nearest neighbor (k-NN) classification algorithm was implemented and evaluated.

Main Results:

• The study compared the performance of classifiers using spectral features (quantile frequencies) and AR model coefficients (6th order).
• Classification performance was assessed separately for inspiration and expiration phases.
• The K-nearest neighbor algorithm was utilized for multi-channel classification.

Conclusions:

• The developed approach using respiratory sound analysis and k-NN classification shows potential as a diagnostic aid.
• Comparing spectral features and AR model parameters provides insights into optimal feature selection for respiratory sound analysis.
• Further validation is needed to establish clinical utility.