A Comparison of obstructive sleep apnoea detection using three different ECG derived respiration algorithms

In this paper, three different algorithms (QRS amplitude, PCA and kernel PCA) were applied to the ECG signal to extract information of the respiratory activity. Features were then extracted from the respiratory activity and used to classify sleep apnoea episodes using an Extreme Learning Machine classifier. Data from the first 60 minutes of the 35 ECG signal recordings from the MIT PhysioNet Apnea-ECG database was used throughout the study. Performance was measured with leave-on-record-out cross validation. The fan-out number for the ELM classifier was varied between one and ten. The results showed that the performance of the PCA algorithm was equal to or outscored the other two algorithms at all fan-out numbers we explored. Its highest performance was an accuracy of 79.4%, a sensitivity of 48.8%, and a specificity of 87.7% at a fan-out of ten.