Detection of sleep apnea on a per-second basis using respiratory signals

There has been a growing interest in out-of-center sleep testing with portable devices for accurate diagnosis of sleep apnea syndrome. This paper presents a new algorithm that extracts features based on filtering and statistical dispersion of the nasal airflow respiration signal and detects apnea events on a per-second basis. The data records were randomly selected from the Sleep Heart Health Study (SHHS-2) database to represent 100 control subjects with Apnea-Hypopnea Index (AHI) less than 5, and 100 apnea subjects with AHI values from 30 to 75. The algorithm was optimized according to the product of sensitivity and positive predictive value of apnea events among a training dataset of 50 apnea subjects with a constraint on the false positive rate among a training dataset of 50 control subjects. From testing of the algorithm on separate datasets, the false positive rate among 50 control subjects was found to be 1.3 events per hour, which corresponds to 100% specificity of classifying apnea subjects. The sensitivity and positive predictive value among 50 apnea subjects were found to be 83.6% and 72.3%, respectively. Among the identified false positive events in the apnea subjects, 64.3% of the events were found to be hypopnea events. Thus, incorporation of hypopnea detection would enhance the performance of the apnea detection algorithm.