Automatic on-line detection of apneas and hypopneas

Obstructive sleep apnea (OSA) is among the most common chronic disorders in humans with a prevalence of about 1% in the general population. The growing number of patients being examined for OSA is causing a strain on healthcare personnel and a need for technological improvements to increase efficiency. Diagnosis of OSA involves evaluation of the airflow signal recorded during sleep. This is a laborious task for the human analyst, who must scan the entire sleep record, manually scoring periods of absent or decreased airflow known as apneas and hypopneas. The repetitiveness and subjectiveness of the task also leads to inaccuracies and low interscorer agreement. The authors present a fast algorithm for the automatic on-line identification of apneic and hypopneic episodes. The approach involves preprocessing of the digitized airflow signal, identification of characteristic abnormal patterns, and application of a rule-based algorithm to classify the results. A preliminary study involving 30-minute records from 8 subjects with varying degrees of OSA shows that the algorithm saves considerable time and effort as compared to manual scoring. It also results in a significant reduction in the number of events missed and an improvement in the overall accuracy and inter-scorer agreement