Acoustic thoracic image of crackle sounds using linear and nonlinear processing techniques

Area of Science:

• Biomedical Engineering
• Signal Processing
• Pulmonology

Background:

• Crackles are adventitious lung sounds indicative of various pulmonary diseases.
• Accurate detection and quantification of crackles are crucial for clinical diagnosis.
• Current methods for crackle analysis face challenges in spatial distribution imaging.

Purpose of the Study:

• To develop and evaluate novel signal processing techniques for imaging crackle sound distribution on the thorax.
• To compare the performance of normalized fractal dimension (NFD), univariate AR modeling with supervised neural network (UAR-SNN), and time-variant autoregressive (TVAR) models.
• To assess the robustness of these methods under various simulated crackle conditions.

Main Methods:

• Simulated crackles were introduced into normal lung sounds acquired from a multichannel system on the posterior thoracic surface.
• Processing schemes (NFD, UAR-SNN, TVAR) were tested by manipulating crackle number, type, spatial distribution, and signal-to-noise ratio (SNR).
• The TVAR model's performance was further validated against a human expert using both simulated and real acoustic data.

Main Results:

• The TVAR scheme demonstrated superior performance in detecting and counting simulated crackles compared to NFD and UAR-SNN.
• TVAR achieved an average specificity near 100% and an average sensitivity of 98 ± 7.5%.
• Effective crackle detection and counting were achieved even with overlapped crackles and low SNR (scaling factor as low as 1.5).

Conclusions:

• Confident imaging of crackle sound distribution via crackle counting using the TVAR model on the thoracic surface is feasible.
• Crackles imaging holds potential as an aid in the clinical evaluation of pulmonary diseases characterized by discontinuous lung sounds.
• The TVAR model offers a robust and accurate approach for analyzing lung sound acoustics.