Automatic breath phase detection using only tracheal breath sounds

Abstract

Current flow estimation methods use tracheal sounds in all except one step of the process: ‘breath phase detection’, is done by assuming alternating breath phases or using a second acoustic channel. The alternating assumption is unreliable in long recordings; non-breathing events (apnea, swallow or cough) change the alternating pattern. Although phases can be detected using lung sounds intensity, the additional channel and associated labor is clinically impractical. We present a method using breath sound parameters to differentiate between the two respiratory phases. The novel method is independent of flow level, requiring only one prior- and one post- breath segment to identify the phase. This was tested on data from 93 healthy individuals, without any history of pulmonary diseases, at 4 different flow levels. The most prominent features were duration, volume and shape of the sound envelope. This method showed accuracy of 95.6%, 95.5% sensitivity and 95.6% specificity.