Heart and lung sounds based events classification

The acoustic signatures reflecting cardiopulmonary system operation contain complex but quantifiable essential diagnostic signs. Overall, these indicators encompass inspiratory and expiratory lung sounds, S1, S2, and other heart sounds with embedded peculiar signals of diverse disease conditions. Therefore, different pattern recognition techniques including among them Gaussian Mixed Models (GMM) and Hidden Markov Models (HMM), and representation of the data in terms of acoustic features vectors, which were utilized in this project, offer useful tool for these events classification. In particular, this paper demonstrates a set of classification schemes for lung and heart sounds with the aim of increasing recognition efficiency. The results achieved in the events classification (inspiration, expiration) with HMM models were 65%, and further improved by applying Principal Component Analysis (PCA) on Mel Frequency Cepstral Coefficients (MFCC), up to 88%. In Quartiles case, the results were improved from 78% to 82%. Thus, PCA improves the results when the noise is present, but Linear Discriminant Analysis (LDA) did not show this advantage. These classification outcomes drive the implementation of events quantification techniques, and also the implementation of automated auscultation for enhanced medical diagnosis.