Spectrogram-based detection of crackles from lung sounds

Analyzing lung sound helps in identifying the associated pathological condition. Auscultation is the common tool used for categorizing the lung sounds. Computer based analysis of lung sounds are reliable as the auscultation is prone to subjectivity. This work deals with identifying the presence of crackles in human lungs utilizing spectrogram analysis. Crackles are intermittent, momentary, non-musical adventitious respiratory sounds. Crackles signifies sudden rupture of air within bronchioles. The duration of crackles is within 20ms, and has a varied frequency range is from 100 to 2000 Hz or above. The breath sound recorded using respiratory sensor and microphone kept close to chest acts as the input signal. The method adopted involves recognition of crackles from recorded lung sounds by extracting and examining the spectral features. Continuous wavelet transform is applied to the recorded lung sounds to remove the unwanted noise and other moving artifacts. Spectrogram and Scalogram is plotted for these transformed signals. Various mathematical morphology features such as energy, entropy, variance and waveform length are extracted from spectrogram of acquired signal. These features further help in demarcation of lung sound into normal and abnormal. This work thus aids the clinician in automated detection of abnormal pulmonary sound for those suffering from respiratory disorders.