Detection of sleep apnea in single channel ECGs from the PhysioNet data base

Abstract

The authors have analyzed single channel ECGs from the Marburg Sleep Laboratory downloaded from Physionet. 35 Data sets with sleep apnea annotation (Learning Set) and 35 testing cases without annotation were available. The authors' analysis and annotations are based on spectral components of heart rate variability. Frequency analysis was performed using Fourier and Wavelet Transformation with appropriate application of the Hilbert Transform. Classification is based on four frequency bands. The authors defined: ULF band (0-0.013 Hz) VLF band (0.013-0.0375 Hz) LF band (0.0375-0.06 Hz) and the HF band (0.17-0.28 Hz). For classification linear discriminant functions using spectral components and other variables derived from the recorder have been used. Classification of cases by means of three variables resulted for the Learning Set in a sensitivity for apnea of 95.0% at a specificity of 100.0% and for the minutes annotation a sensitivity was 90.8% and specificity 92.7% Allocation per minute based on a discriminant function using 30 variables.