AUTOMATED DETECTION OF ANOMALIES IN ELECTROCARDIOGRAMS USING EMPIRICAL MODE DECOMPOSITION

Study objective: Develop an algorithm for detection and classification of heart arrhythmia in electrocardiograms. Methodology/approach: Different ways of using the collected data are discussed, starting from the simplest one, which is the beat counter, to the more complex ones, where the complete signals present in an electrocardiogram is used. Different Machine Learning techniques were also used: K-Nearest Neighbors, Logistic Regression, Support Vector Machines and Extra Trees. The beat counter approach considers the time difference between each cardiac cycle and can be collected by a simple smart watch or an oximeter. For the complete classification of the anomalies, two other signal processing techniques were considered: the Fourier Transform and the Empirical Mode Decomposition. Originality/Relevance: It is the first paper to use the Empirical Mode Decomposition combined with Machine Learning techniques for the classification and detection of cardiac anomalies. Main results: The beat counter is not efficient enough to distinguish between all classes of existing anomalies, even among those studied in this work, but it presents good results for binary distinction between normal and abnormal heart beat. For the complete classification of the anomalies, the Empirical Mode Decomposition presented the best results. It is even better than the time-frequency analysis technique used in other papers on electrocardiogram classification. Theoretical/methodological contributions: This paper presents a new application for Empirical Mode Decomposition and how it can be combined with classification techniques.