Detection and clustering analysis of crackles in respiratory sounds

In this work, a method for crackle detection is proposed using two-level thresholding of the energy envelopes of the respiratory sounds. The results of the clustering analysis based on the agglomerative hierarchical clustering of crackle features are given. Detection was performed on the respiratory sound waveforms recorded from nine patients having obstructive and restrictive pulmonary diseases. Four parameters, initial deflection width (IDW), two cycle duration (2CD), total duration (TD), and peak frequency (PF), were then extracted from the detected crackles and were used as features in the clustering process.