Obstructive Sleep Apnea Classification Based on Spectrogram Patterns in the Electrocardiogram

In this paper we describe the findings of an exploratory study of the effect of obstructive sleep apnea (OSA) on the electrocardiogram (ECG) signal. Episodes of sleep apnea are characterized by periodic cycles of breathing cessation and restoration. Our analysis was guided by the hypothesis that these cycles synchronously alter the ECG. We discovered several characteristic indicators of apnea in the ECG signal. Our study focused on data sets provided for the Computers in Cardiology (CINC) 2000 apnea classification competition. After careful QRS detection, artifact removal, and preprocessing, we found that we could recognize sleep apnea by visually inspecting spectrograms of various features of the ECG such as the heart rate (HR), S-pulse amplitude, and pulse energy. As part of this study we entered both CINC competitions. We were able to correctly classify 28 out of 30 subjects in our initial competition entry and 30 out of 30 in our third entry. Once each signal was classified as a whole, we were able to correctly classify each minute in 13,626 out of 17,268 cases in our initial entry and 15,994 cases in our fourth entry.