Bioacoustic signal analysis through complex network features.

Area of Science:

• Bioacoustics
• Network Science
• Medical Diagnostics

Background:

• Traditional auscultation relies on subjective interpretation of respiratory sounds.
• Understanding airflow dynamics is crucial for diagnosing respiratory conditions.
• Bioacoustics signals, like vesicular (VE) and bronchial (BR) breath sounds, contain valuable diagnostic information.

Purpose of the Study:

• To propose a novel graph-theoretical approach for analyzing and classifying respiratory bioacoustics signals.
• To explore the potential of complex network analysis in understanding airflow dynamics during respiration.
• To develop a methodology for remote auscultation applicable in clinical settings, including during pandemics like COVID-19.

Main Methods:

• Complex network analysis was applied to bioacoustics signals (VE and BR breath sounds) from 48 healthy individuals.
• Machine learning techniques were used to classify breath sounds by extracting graph features: number of edges (E), graph density (D), transitivity (T), degree centrality (Dcg), and eigenvector centrality (Ecg).

Main Results:

• Higher values of E, D, and T in BR sounds correlate with temporally correlated airflow in wider airways, producing sustained low-frequencies.
• Lower values of E, D, and T in VE sounds indicate less correlated airflow through narrower airways, associated with higher frequencies.
• Lower degree and eigenvector centrality values support spectral and other graph parameter inferences.

Conclusions:

• Graph-theoretical analysis provides a quantitative method for classifying respiratory sounds.
• The distinct graph features of VE and BR sounds reflect underlying airflow dynamics.
• This methodology shows promise for developing objective, remote auscultation tools.