A novel expert classifier approach to pre-screening obstructive sleep apnea during wakefulness

摘要

Obstructive sleep apnea (OSA) is a widespread disorder that is cumbersome to diagnose using the goldstandard, overnight polysomnography (PSG). This paper highlights further development of our Awake-OSA method for predicting whether someone has severe sleep apnea using breath sounds recorded during wakefulness. We propose the use of an expert classification approach that consists of individual majority-voting classifiers. Each classifier is trained to distinguish one class of subject from all other classes. The outcomes of these classifiers are, in turn, combined using a truth matrix to determine the final outcome. Using the breath sound features of 249 subjects, the classifiers attempted to classify 180 subjects as either non-OSA (AHI less than 5) or severe-OSA (AHI greater than 30). 79% and 75% of OSA and non-OSA subjects, respectively, could be classified. Of those classified, the resultant testing sensitivity and specificity were found to be 78% and 86%, respectively. The consistency of the testing to training accuracies indicates the robustness and generalizability of using multiple expert classifiers on the dataset. This technique has the potential to be used in a doctor's office to rapidly and cheaply pre-screen for OSA, so that physicians may be better able to determine which patients are in need of overnight PSG.