Session S53.4

Screening For Obstructive Sleep Apnoea Based On The ECG: The CinC Challenge

B. Raymond, R. Cayton, R. Bates, M. Chappell

Birmingham Heartlands Hospital
Birmingham, UK

The definitive diagnostic tool for obstructive sleep apnoea (OSA) is the polysomnogram: an overnight, multichannel recording which is expensive and time-consuming to carry out and analyse. There is continuing interest in screening methods to reduce the burden on sleep laboratories performing polysomnography. The changes in heart rate and respiration during OSA are well known and it is possible to detect episodes of apnoea from the ECG alone.
The beat times of each ECG recording were extracted using QRS template matching with manual editing. The ECG-derived respiratory (EDR) signal and the RR interval tachogram were constructed. Cycles of tachy/bradycardia (consistent with a cardiovascular arousal from sleep, as would be expected at the end of an episode of apnoea) were identified but only those cycles which were accompanied by a brief episode of hyperventilation (assessed using the EDR signal) were retained. This was to assist in reducing the number of "false" arousals due to other causes (such as other sleep disorders or spontaneous arousals). Power spectral features from the EDR and RR interval signals were computed using wavelet and time-frequency distribution methods, considering principally the power at respiratory frequency and at frequencies below 0.1 Hz. Movement artifact in the ECG signal was also identified and used as a marker of arousal. These features were collated into minute-by-minute vectors and passed to a shared mixture classifier. This classifier models the data distribution using a mixtures of Gaussians, allowing application to non-normal data. Screening results were determined from the minute-by-minute classification of the data from each test subject.
Preliminary results at the time of submission offered correct classification in 22 of the 30 subjects in the Challenge test database.
This work was supported by the Mathematics in Medicine Initiative at the University of Warwick.