Підхід до оцінювання електрокардіосигналів на основі багатофакторного регресійного аналізу функції часової варіабельності

Abstract

The study presents the development of a multifactor regression model designed for automated identification of cardiac conditions through the analysis of statistical parameters of electrocardiographic data temporal variability. The developed model is based on five key statistical indicators: arithmetic mean, median, mode, root mean square deviation, and kurtosis. These parameters were identified through a sequential regression ion procedure a total set of thirteen primary characteristics at a threshold level of statistical significance p < 0.05. The weight parameters of each indicator were calculated using quadratic deviation optimization. The arithmetic mean demonstrates the dominant influence on the diagnostic outcome (coefficient β = 153.952), while the remaining indicators provide a refining effect with negative weight coefficients. The developed model demonstrates high efficiency in distinguishing three categories of cardiac activity: physiological norm, arrhythmic disorders such as premature cardiac contractions, and structural myocardial conduction disturbances (partial left bundle branch block). The modeling quality is confirmed by the coefficient of determination R² = 0.97636 (adjusted R² = 0.97577). The physiological state is characterized by minimal variance of time intervals (0.00003-0.00004 s²), while premature contractions demonstrate a thousand-fold increase in this indicator (0.011…0.012 s²). Validation on an experimental dataset of 204 cardiac activity records confirmed the high reliability of the system (Fisher's F-statistic = 1635.7 at p < 0.001) and compliance with fundamental requirements of regression modeling. The analysis of prediction errors revealed their correspondence to Gaussian distribution and variance homogeneity, confirming the correctness of the developed system. The proposed approach combines the advantages of classical statistical methods with the proposed use of the time variability function for comprehensive analysis of morphological and rhythmic characteristics of cardiac signals. The practical significance lies in creating mathematical tools for automated cardiovascular disease diagnostic systems and clinical decision support systems.