Oleg Stelia
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Oleg Stelia has a total of 19 co-authors. Here are the top co-authors by number of shared publications:
- Iurii Krak
Taras Shevchenko National University of Kyiv
6 shared publications - Leonid Potapenko
Faculty of Computer Science and Cybernetics Taras Shevchenko National University of Kyiv 03680 Ukraine
3 shared publications - Anatolii Pashko
Taras Shevchenko National University of Kyiv, Kyiv, Ukraine
2 shared publications - I. Sirenko2 shared publications
Igor Stelia
EPAM Systems, kyiv, Ukraine
2 shared publications- Mykola Yefremov
Department of Theoretical Cybernetics, Taras Shevchenko National University of Kyiv, Kyiv, Ukraine, Ukraine
2 shared publications
Latest Publications
- 1 August 2026
- Published by Springer Nature in Cybernetics and Systems Analysis
- Vol. 62 (4) , 690-700
Accurate R-Peak Detection in Electrocardiograms: Piecewise Polynomial Approximation, Noise Reduction, and First-Derivative Analysis
- 1 August 2026
- Published by Springer Nature in Cybernetics and Systems Analysis
- Vol. 62 (4) , 690-700
Accurate R-Peak Detection in Electrocardiograms: Piecewise Polynomial Approximation, Noise Reduction, and First-Derivative Analysis
- 22 January 2026
- Published by National Aerospace University - Kharkiv Aviation Institute in RADIOELECTRONIC AND COMPUTER SYSTEMS
- Vol. 2026 (1) , 22-34
An efficient threshold method for detecting R-peaks in electrocardiogram
Automated R-peak detection in electrocardiogram (ECG) signals is essential to heart rhythm analysis, with applications in heart rate monitoring, heart rate variability (HRV) assessment, arrhythmia diagnosis ets. Its accuracy, however, is highly sensitive to noise and artifacts present in the ECG recording. The study proposes a method and its software implementation for R-peak detection, based on signal smoothing followed by differentiation and the application of a thresholding approach. The method is designed for use in resource-constrained environments, such as portable and embedded monitoring systems. The objective of the study is to develop a computationally efficient and accurate method for the automatic detection of R-peaks in ECG signals, tailored to a personalized patient approach. The primary focus is on robustness to noise and QRS complex morphology, as well as the algorithm’s ability to operate under limited computational resources and in real-time conditions. The proposed method relies on staged ECG signal processing. To validate the approach and compare its effectiveness, existing studies in the field were analysed. The subject of the study involves processing ECG signals and addressing challenges in R-peak detection in ECGs recorded via Holter monitors by constructing a smoothed continuous signal based on discrete data obtained during digitization. The discrete nature of the initial signal complicates differentiation and the precise identification of characteristic points, specifically R-peaks, which play a crucial role in diagnosing cardiac conditions. The scientific novelty of this investigation lies in the use of a second-order piecewise polynomial approximation to represent the ECG signal. This approach enables noise reduction in the signal and represents the discrete signal as a continuous function together with its first derivative, thereby permitting the analytical computation of its derivative. Results involve the detection of R-peaks by analysing the derivative of the smoothed signal: regions with sharp changes characteristic of the QRS complex were identified, and an iterative smoothing scheme was developed, with the number of iterations determined by a proposed stopping criterion. The proposed method was implemented in software and tested on data from the open-access MIT-BIH Arrhythmia Database, which includes over 60 recordings and more than 100,000 annotated R-peaks. The results were compared with studies by other authors using the metric, based on standard precision and sensitivity metrics. Conclusions: The study proposes an effective and adaptive approach to ECG signal processing, ensuring reliable R-peak detection under conditions of significant noise, baseline drift, and physiological variability across patients. The obtained results demonstrated high performance metrics: up to 99.5% (average above 99.1%), consistently exceeding 99%, and in some recordings reaching 100%. Thus, the proposed approach is competitive, demonstrating high accuracy in detecting R-peaks in cases of arrhythmias, peak inversions, non-standard QRS complex morphology, and other challenging signal conditions. The study’s results can be applied in ECG analysis practice, particularly in the development of automated diagnostic systems or signal preprocessing before the application of classification methods.
- 22 January 2026
- Published by National Aerospace University - Kharkiv Aviation Institute in RADIOELECTRONIC AND COMPUTER SYSTEMS
- Vol. 2026 (1) , 22-34
An efficient threshold method for detecting R-peaks in electrocardiogram
Automated R-peak detection in electrocardiogram (ECG) signals is essential to heart rhythm analysis, with applications in heart rate monitoring, heart rate variability (HRV) assessment, arrhythmia diagnosis ets. Its accuracy, however, is highly sensitive to noise and artifacts present in the ECG recording. The study proposes a method and its software implementation for R-peak detection, based on signal smoothing followed by differentiation and the application of a thresholding approach. The method is designed for use in resource-constrained environments, such as portable and embedded monitoring systems. The objective of the study is to develop a computationally efficient and accurate method for the automatic detection of R-peaks in ECG signals, tailored to a personalized patient approach. The primary focus is on robustness to noise and QRS complex morphology, as well as the algorithm’s ability to operate under limited computational resources and in real-time conditions. The proposed method relies on staged ECG signal processing. To validate the approach and compare its effectiveness, existing studies in the field were analysed. The subject of the study involves processing ECG signals and addressing challenges in R-peak detection in ECGs recorded via Holter monitors by constructing a smoothed continuous signal based on discrete data obtained during digitization. The discrete nature of the initial signal complicates differentiation and the precise identification of characteristic points, specifically R-peaks, which play a crucial role in diagnosing cardiac conditions. The scientific novelty of this investigation lies in the use of a second-order piecewise polynomial approximation to represent the ECG signal. This approach enables noise reduction in the signal and represents the discrete signal as a continuous function together with its first derivative, thereby permitting the analytical computation of its derivative. Results involve the detection of R-peaks by analysing the derivative of the smoothed signal: regions with sharp changes characteristic of the QRS complex were identified, and an iterative smoothing scheme was developed, with the number of iterations determined by a proposed stopping criterion. The proposed method was implemented in software and tested on data from the open-access MIT-BIH Arrhythmia Database, which includes over 60 recordings and more than 100,000 annotated R-peaks. The results were compared with studies by other authors using the metric, based on standard precision and sensitivity metrics. Conclusions: The study proposes an effective and adaptive approach to ECG signal processing, ensuring reliable R-peak detection under conditions of significant noise, baseline drift, and physiological variability across patients. The obtained results demonstrated high performance metrics: up to 99.5% (average above 99.1%), consistently exceeding 99%, and in some recordings reaching 100%. Thus, the proposed approach is competitive, demonstrating high accuracy in detecting R-peaks in cases of arrhythmias, peak inversions, non-standard QRS complex morphology, and other challenging signal conditions. The study’s results can be applied in ECG analysis practice, particularly in the development of automated diagnostic systems or signal preprocessing before the application of classification methods.
- 19 October 2023
- Published by Academy of Cognitive and Natural Sciences
Hardware Usage Improvement for Small Data Problem Solving by Deep Learning Methods
This article dedicated to the question of improvement of deep calculations, particularly, the price of appliances, the price of electricity and others, which may, in turn, affect not only the price of research but also repeatability and consistency of the experiments. According to the research, we propose an idea, which should be useful for sophisticated calculations, such as small data machine learning using different constructs of neural networks, including RCNNs, DCGANs, which are known to be very time and memory consuming methods if applied on a particular data. Here, the improvement underlies the usage of our own approaches to process data, to compare hardware using well known techniques to benchmark hardware as central processing units (CPUs) as well as graphic processing units (GPUs). We use this straightforward approach in our experiments in order to assess the hypotheses about computational devices as well as algorithms, using proven and online available datasets. This in turn, helped us to infer the device properties based upon their behavior in experiments and, hence, enhance the productivity and quality of our experiments. For our experiments, we used a wide range of libraries such as scikit-learn, TensorFlow, Pytorch and Direct ML for the hands-on available desktop and laptop computing devices running CPUs and GPUs as well. According to the study, we figured the main differences in behavior of CPUs and CPUs based upon available hands-on in the particular situation – training of deep neural networks. Our findings gave us the possibility, with some confidence, to choose appropriate devices for the deep learning tasks based on their sophistication.
- 19 October 2023
- Published by Academy of Cognitive and Natural Sciences
Hardware Usage Improvement for Small Data Problem Solving by Deep Learning Methods
This article dedicated to the question of improvement of deep calculations, particularly, the price of appliances, the price of electricity and others, which may, in turn, affect not only the price of research but also repeatability and consistency of the experiments. According to the research, we propose an idea, which should be useful for sophisticated calculations, such as small data machine learning using different constructs of neural networks, including RCNNs, DCGANs, which are known to be very time and memory consuming methods if applied on a particular data. Here, the improvement underlies the usage of our own approaches to process data, to compare hardware using well known techniques to benchmark hardware as central processing units (CPUs) as well as graphic processing units (GPUs). We use this straightforward approach in our experiments in order to assess the hypotheses about computational devices as well as algorithms, using proven and online available datasets. This in turn, helped us to infer the device properties based upon their behavior in experiments and, hence, enhance the productivity and quality of our experiments. For our experiments, we used a wide range of libraries such as scikit-learn, TensorFlow, Pytorch and Direct ML for the hands-on available desktop and laptop computing devices running CPUs and GPUs as well. According to the study, we figured the main differences in behavior of CPUs and CPUs based upon available hands-on in the particular situation – training of deep neural networks. Our findings gave us the possibility, with some confidence, to choose appropriate devices for the deep learning tasks based on their sophistication.
- 23 July 2021
- Published by Springer Nature
- p. 507-518
Baseline Wander Correction of the Electrocardiogram Signals for Effective Preprocessing
- 23 July 2021
- Published by Springer Nature
- p. 507-518
Baseline Wander Correction of the Electrocardiogram Signals for Effective Preprocessing
- Open Access
- 12 July 2021
- Published by Cambridge University Press (CUP) in Cambridge Open Engage
Computer Simulation and Optimization to Solve Groundwater Flow Problems
Examples of using the developed mathematical methods and software for solving environmental problems caused by groundwater are given.
- 12 July 2021
- Published by Cambridge University Press (CUP) in Cambridge Open Engage
Open AccessComputer Simulation and Optimization to Solve Groundwater Flow Problems
Examples of using the developed mathematical methods and software for solving environmental problems caused by groundwater are given.
- Open Access
- 15 February 2021
- Published by Cambridge University Press (CUP) in Cambridge Open Engage
The geometric method for solving a heat equation
In the proposed publication, we present a geometric method for solving a boundary value problem for a nonstationary heat equation. The equivalence of the geometric solution to an explicit difference method is shown with the corresponding initial parameters. There are considered the first and second boundary value problems. The physical interpretation of the method is based on local balance.
- 15 February 2021
- Published by Cambridge University Press (CUP) in Cambridge Open Engage
Open AccessThe geometric method for solving a heat equation
In the proposed publication, we present a geometric method for solving a boundary value problem for a nonstationary heat equation. The equivalence of the geometric solution to an explicit difference method is shown with the corresponding initial parameters. There are considered the first and second boundary value problems. The physical interpretation of the method is based on local balance.
- Open Access
Local piecewise polynomial curve of variable order and its appliance for calculating of nodal derivatives
In this paper, we present a new Hermite type curve piecewise polynomial of C1 continuity on nonuniform grids. Depending on the location of the knots of the grid (uniform and nonuniform), the curve is quadratic or cubic. The approximation error is investigated. For nonuniform grids a method for calculating the first and second derivatives in knots is proposed. We also perform several numerical experiments which exemplify the properties of the proposed curve. Graphic examples are given.
Open AccessLocal piecewise polynomial curve of variable order and its appliance for calculating of nodal derivatives
In this paper, we present a new Hermite type curve piecewise polynomial of C1 continuity on nonuniform grids. Depending on the location of the knots of the grid (uniform and nonuniform), the curve is quadratic or cubic. The approximation error is investigated. For nonuniform grids a method for calculating the first and second derivatives in knots is proposed. We also perform several numerical experiments which exemplify the properties of the proposed curve. Graphic examples are given.
- 26 July 2020
- Published by Springer Nature
- p. 409-422
Isolation of Informative Features for the Analysis of QRS Complex in ECG Signals
- 26 July 2020
- Published by Springer Nature
- p. 409-422
Isolation of Informative Features for the Analysis of QRS Complex in ECG Signals
- 24 July 2019
- Published by Springer Nature
- p. 215-224
Controlled Spline of Third Degree: Approximation Properties and Practical Application
- 24 July 2019
- Published by Springer Nature
- p. 215-224
Controlled Spline of Third Degree: Approximation Properties and Practical Application
- Open Access
- 10 April 2018
- Published by Private Company Technology Center in Eastern-European Journal of Enterprise Technologies
- Vol. 2 (4 (92)) , 46-52
Application of piecewisecubic functions for constructing a Bezier type curve of C1 smoothness
We have proposed and implemented a new method for constructing a spline curve of third degree, which possesses the properties of both a cubic spline and the Bezier curve. Similar to building the Bezier curves, control points are assigned, whose position affects the shape of the curve. In the proposed approach, the sections of the straight line that connect control points are tangent to the curve that is constructed. The location of touch points can be different, allowing the construction of different curves for one set of control points. A special feature of the proposed method is assigning, in abscissa of control points, some unknown spline values that are found from conditions for the continuity of the first derivatives of the curve at these points. Finding the coefficients of polynomials that make up the curve comes down to solving a system of linear equations with a three-diagonal matrix. The built curve is a piecewise-cubic function, continuous along with its first derivative throughout the entire interval. Conditions were found for any set of control points in the form of inequalities, which parameters of the curve must meet, at which the curve does exists and it is unique. These conditions follow from the requirement for a diagonal advantage of the matrix of the system for determining coefficients of the curve. A series of computational experiments were performed, which showed that the curve effectively inherits the shape assigned by control points. Similar to the Bezier curves, the proposed curve could be used in computer graphics systems and computer systems for technical design, specifically for the creation of fonts, drawings of parts, elements of transportation vehicles' bodies, etc.
- 10 April 2018
- Published by Private Company Technology Center in Eastern-European Journal of Enterprise Technologies
- Vol. 2 (4 (92)) , 46-52
Open AccessApplication of piecewisecubic functions for constructing a Bezier type curve of C1 smoothness
We have proposed and implemented a new method for constructing a spline curve of third degree, which possesses the properties of both a cubic spline and the Bezier curve. Similar to building the Bezier curves, control points are assigned, whose position affects the shape of the curve. In the proposed approach, the sections of the straight line that connect control points are tangent to the curve that is constructed. The location of touch points can be different, allowing the construction of different curves for one set of control points. A special feature of the proposed method is assigning, in abscissa of control points, some unknown spline values that are found from conditions for the continuity of the first derivatives of the curve at these points. Finding the coefficients of polynomials that make up the curve comes down to solving a system of linear equations with a three-diagonal matrix. The built curve is a piecewise-cubic function, continuous along with its first derivative throughout the entire interval. Conditions were found for any set of control points in the form of inequalities, which parameters of the curve must meet, at which the curve does exists and it is unique. These conditions follow from the requirement for a diagonal advantage of the matrix of the system for determining coefficients of the curve. A series of computational experiments were performed, which showed that the curve effectively inherits the shape assigned by control points. Similar to the Bezier curves, the proposed curve could be used in computer graphics systems and computer systems for technical design, specifically for the creation of fonts, drawings of parts, elements of transportation vehicles' bodies, etc.