Crackles detection using a time-variant autoregressive model

Area of Science:

• Respiratory acoustics
• Signal processing
• Biomedical engineering

Background:

• Automatic detection of fine and coarse crackles in lung sounds remains a challenge.
• Existing methods for crackle detection have limitations in accuracy and robustness.

Purpose of the Study:

• To evaluate the effectiveness of the time-variant autoregressive (TVAR) model for detecting and quantifying fine and coarse crackles.
• To compare the TVAR model's performance against expert-based time-expanded waveform analysis.

Main Methods:

• The TVAR model was applied to simulated and real lung sound recordings containing crackles.
• Adaptive filtering with recursive least squares (forgetting factor 0.97) and a model order of four were used to obtain TVAR coefficients.
• Performance was assessed by comparing TVAR detection with expert analysis.

Main Results:

• The TVAR model demonstrated over 90% efficiency in detecting crackles, outperforming expert analysis (around 30% efficiency).
• The TVAR model successfully detected crackles even when they were overlapping or had amplitudes as low as 1.5 times the standard deviation of background lung sounds.
• Time-expanded waveform analysis showed significant limitations under these challenging conditions.

Conclusions:

• The TVAR model is a suitable alternative for accurate detection and estimation of fine and coarse crackles in lung sounds.
• TVAR model performance is robust in the presence of overlapping crackles and low-amplitude signals.
• TVAR offers a significant improvement over traditional methods for crackle detection in complex respiratory sounds.