Unsupervised phase detection for respiratory sounds using improved scale-space features

Automatic respiratory sound (RS) analysis provides a possible solution for the minimization of inherent subjectivity caused by auscultation via stethoscope, and it allows a reproducible quantification of RS. As one of the crucial initial steps, reliable unsupervised respiratory phase detection plays an important role in automatic RS analysis. In this paper, a novel unsupervised phase detection scheme is proposed using improved triplet markov chain (TMC) based statistical technique. The problems of the commonly used unsupervised respiratory phase detection techniques and their improvement with the proposed discriminative features are explored. The feasibility and limitations of this advanced statistical approach for respiratory phase detection are also addressed.