Distinguishing obstructive from central sleep apneas and hypopneas using linear SVM and acoustic features

Sleep Apnea (SA) is a very common but underdiagnosed respiratory disorder. SA has 2 main types, obstructive and central sleep apnea (OSA and CSA, respectively). The distinction between the 2 types is important for proper clinical management. Our aim in this study was to deploy acoustic analysis of breath sounds to distinguish central from obstructive events. We recorded breath sounds from 29 patients from which 45 segments with obstructive only and 40 segments with central only respiratory events were isolated. Subsequently, 10 acoustic features were extracted and used to identify basic breath sounds: inspiration, expiration, and snoring. A 2
^nd
set of 6 sound-specific features were extracted from the basic sounds, designed based on SA pathophysiology. These 6 features were used to train and test a linear SVM classifier using a leave-one-out cross validation scheme. We achieved an excellent accuracy of 91.8%. In conclusion, this is the first study to demonstrate the ability to distinguish CSA from OSA with high reliability from breath sound recordings during sleep.