Big-data analytics for Arrhythmia Classification using data compression and kernel methods

Big data analytics is broadly used today in multiple research fields to discover and analyze hidden patterns and other useful information in large databases. Although Cardiac Arrhythmia Classification (CAC) has been studied in depth to date, new CAC methods need to be still designed. In this work, we propose a new big data analytics method for automatic CAC of intracardiac Electrograms (EGMs) stored in Implantable Cardioverter Defibrillators (ICDs). The proposed method combines the effectiveness of a measure based on data compression concepts (Jaccard dictionary similarity), which exploits the information among EGMs, and the classification power of kernel methods. It also requires minimal EGM preprocessing and allows us to deal with EGMs of different duration. A database of 6848 EGMs extracted from a national scientific big data service for ICDs, named SCOOP platform, were used in our experiments. Performance for two classifiers (k-Nearest Neighbors or k-NN, and Support Vector Machines or SVM) were compared in two CAC scenarios using four different input spaces. Results showed that k-NN worked better than SVM when previous episodes from the same patient were available in the classifier design, and vice-versa. For the best cases, k-NN and SVM yielded accuracies near to 95% and 85%, respectively. These results suggest that the proposed method can be used as a high-quality big data service for CAC, providing a support to cardiologists for improving the knowledge on patient diagnosis.