Estimation of lung sound cycle span using spectro-temporal respiratory frequency evaluation

The common lung disease diagnostics by pulmonologists involve manual thorax auscultation using stethoscopes. Despite years of experience, this method is susceptible to human errors, which an automated system can alleviate to a large extent. An important step towards computerized lung disease detection involves efficient extraction of inspiration-expiration phases of complete lung sound cycles (LSCs), which mainly suffer from inter-observer variability when a manual segmentation process is employed. This work proposes automated respiratory cycle extraction by utilizing a joint spectro-temporal respiratory frequency identification approach applied to the lung sound signal envelope. Considering the dynamics of LSC over time and corresponding frequencies, the energy distribution related to modulating spectral bands of respiration is quantified to further optimize the cycle extraction process. We also compare the performance of single and multi-channel lung sound signals for precise identification of lung sound modulation frequency. Results show that the cycle demarcation provided by the proposed LSC algorithm exhibits lower error when evaluated using the ground truth values.