Modified Matching Pursuit algorithm for application in sleep screening

Sleep Apnea (SA) is considered to be a major, underdiagnosed public health problem with a prevalence of 2 – 4% for middle-aged women and men, respectively. Therefore a reliable, ambulant screening test is requested, which is easy to perform and does not necessarily demand profound knowledge of sleep medicine. In this paper a new Matching Pursuit based algorithm is presented that uses only information from one single pulse oxymetry sensor (SpO2, pulse wave amplitude, pulse frequency) for detecting different sleep disturbing events. Thus, sleep diagnosing parameters are calculated, which until now could only be determined by using multiple sensors. A signal decomposition algorithm based on a dictionary of time-frequency atoms (known as “Matching Pursuit method”) has been modified in order to analyze different patterns in the mentioned signals. The conventional procedure of Matching Pursuit, using dictionaries of signal templates, is optimized to be implemented in a standard embedded system. The algorithm was tested on 62 consecutive adult patients with suspected SA. All patients underwent standard overnight polygraphy (PG) with SOMNOcheck2 (WEINMANN, Germany), which is an established method for PG diagnosis. The correlation coefficient between manual scored AHI and automatic RDI, calculated from the new algorithm, using only pulse oximetry channels, was r = 0.967. Bland-Altmann analysis showed a mean difference of −0.6 between the two parameters. Using a cut-off value of RDI ≥ 15/h for SA classification, a sensitivity of 96.2% and specificity of 91.7% was reached. This novel computer algorithm provides a simple and highly accurate tool for quantification of SA.