ИСТИНА

Background and Goal: Among serious cardiovascular diseases, a special place is occupied by the left bundle branch block (hereinafter – LBBB). According to statistics, this disease affects 1.2% of adults under the age of 50 (by the age of 80, the proportion of diseases increases to 17%). LBBB - is a residual common syndrome, which is characterized by a violation of the conduction of an electric pulse in the left bundle branch of the heart's conducting system, which, in turn, can lead to a violation of the normal heart rate and the disease of heart failure progression [1]. In this regard, the development of mathematical methods and software with the use of machine learning and data mining technologies for automating the diagnosis of LBBB in patients' cardiograms, including in remote mode, is of particular relevance. The aim of this work is to develop a such model that provides automatic and high-precision diagnosis of LBBB in ECG studies. Methods: As the main method of training the model, the SVM (Support Vector Machine) method was used, which is based on the concepts of optimal class separation and hyperplane construction for data separation and is intended for their classification and regression. The key fact is that we are dealing with cardiograms that have undergone preliminary (standard) processing: high-pass, low-pass, network interference filters are applied, noisy segments are removed, which significantly increases the efficiency of solving the problem. The database for training this model was filled with about 10,000 ECG records provided by employees of ATES MEDICA Soft LLC, with confirmed and (positive and negative) medical diagnoses. Results: A work was carried out to select the parameters that most significantly affect the quality of the model, and a Confusion Matrix was constructed to evaluate the classifier's performance. Through testing, the optimal set of parameters was selected, providing the best results and the lowest percentage of medical error of the 2nd kind. Conclusions: Numerical experiments conducted on a test sample from the test database of ECG records, showed that the developed model can already be applied in practice. References [1] John R. Hampton. The ECG Made Easy (6th Edition). In: Churchill Livingstone, pp. 30–51, 2003.