Method for automatic detection of wheezing in lung sounds

Area of Science:

• Biomedical Engineering
• Respiratory Medicine
• Signal Processing

Background:

• Wheezing is a key indicator of respiratory conditions like asthma and COPD.
• Accurate and automated wheezing detection from lung sounds is crucial for timely diagnosis and management.
• Current methods may lack efficiency or require specialized expertise for interpretation.

Purpose of the Study:

• To develop and validate an automated technique for wheezing recognition in digital lung sound recordings.
• To leverage spectral information processing for enhanced wheeze detection accuracy.
• To provide a user-friendly system with visual feedback for clinical application.

Main Methods:

• Digital lung sounds were recorded and pre-processed to normalize spectral information.
• Spectrograms were computed and enhanced using 2D convolution and thresholding.
• Spectral projections were fed into a multi-layer perceptron artificial neural network for wheeze classification.

Main Results:

• The automated system achieved 84.82% accuracy for isolated respiratory cycles.
• Accuracy increased to 92.86% when analyzing groups of respiratory cycles from the same individual.
• The system provides raw sound and processed spectrograms for user verification.

Conclusions:

• The developed spectral analysis and artificial neural network technique offers a robust method for automatic wheezing detection.
• The system demonstrates high accuracy, particularly when analyzing multiple respiratory cycles.
• This automated approach can support clinicians in diagnosing and monitoring respiratory diseases.