ECG Database for Evaluating the Efficiency of Recognizing Dangerous Arrhythmias

One of the difficult tasks of practical medicine is to identify rhythm disturbances in the work of the cardiovascular system in patients under supervision. Early detection of a dangerous violation could save the patient's life due to the use of resuscitation measures, primarily such as defibrillation. The article is devoted to a review of the developed database of short ECG fragments for evaluating the operation of algorithms for recognizing dangerous rhythm disturbances. The database consists of six classes of arrhythmias (90 ECG fragments of each class), representing varying degrees of danger to life. Each ECG fragment in the database is represented by the following four data files. This is a 2-second ECG signal, as well as three types of signal power spectral density: a full spectrum and two smoothed spectra (15 and 10 signs, respectively). There are also illustrations of different classes from this database. A detailed description of the base and primary sources of ECG records is presented. The developed database is used in works to classify various arrhythmias by spectral description by ECG fragments using the weighted k nearest neighbors (kNN) method, the nearest convex hull method (LP-NCH), linear discriminant analysis (LDA), support vector machine (SVM), and neural network methods.