Applying cybernetic technology to diagnose human pulmonary sounds

Area of Science:

• Medical Diagnostics
• Signal Processing
• Artificial Intelligence

Background:

• Chest auscultation is vital for lung disease diagnosis but is subjective and limited by human hearing sensitivity to low frequencies.
• Pulmonary sounds (PSs) above 120 Hz are difficult for the human ear to differentiate, hindering accurate classification.
• Existing diagnostic methods struggle with the complexity and frequency range of pulmonary sounds.

Purpose of the Study:

• To develop an automated system for accurate classification of six common pulmonary sound types.
• To overcome the limitations of subjective auscultation and human auditory sensitivity.
• To improve the diagnosis of lung diseases through advanced signal processing and artificial intelligence.

Main Methods:

• Acquisition of PS signals using a piezoelectric microphone and data acquisition card.
• Feature extraction via wavelet transform for decomposing PS signals into frequency subbands.
• Development of a two-stage classifier combining back-propagation (BP) and learning vector quantization (LVQ) neural networks with 17 statistical features.

Main Results:

• The proposed system successfully classified six types of pulmonary sounds: vesicular, bronchial, tracheal, crackles, wheezes, and stridor.
• The two-stage neural network classifier demonstrated high performance, verified by receiver operating characteristic (ROC) curve analysis.
• The system effectively addresses the limitations of human auditory perception in low-frequency sound analysis.

Conclusions:

• The developed PS recognition system offers an objective and accurate method for classifying common pulmonary sounds.
• This technology expands traditional auscultation, providing a valuable tool for lung disease diagnosis.
• The system's design, including characteristic values and spectral analysis, facilitates a clear human-machine interface.