Improvement of a P-wave detector by a bivariate classification stage

For more than three decades, beat-to-beat detection of electrical atrial activity has been subject of interest in the biomedical field. Several detection schemes and algorithms have been proposed; however, this problem has not been satisfactory solved and remains the main error source for accurate automatic detection and classification of supraventricular arrhythmias, mainly in ambulatory applications and in coronary care units. Some detection methods have been proposed, but none of them have been quantitatively tested or compared, due in part to the lack of a database with annotated P-waves. In this work, a classification stage is proposed as a new decision strategy to improve a P-wave detector previously developed in our laboratory. A quantitative evaluation of the original detector, the proposed classification stage and two other classical P-wave detectors is performed by means of the calculation of their receiver operating characteristics (ROC) curves. Results show the increased performance provided by the new classification stage with respect to the originally developed algorithm and two classical P-wave detection methods.