WO2020242334A1

WO2020242334A1 - System and method of automated electrocardiogram analysis and interpretation

Info

Publication number: WO2020242334A1
Application number: PCT/RU2019/000367
Authority: WO
Inventors: Tatiana Vladimirovna PODLADCHIKOVA; Nitalia Yurievna GLAZKOVA; Daria Korillovna STEPANOVA
Original assignee: Autonomous Non-Profit Organization For Higher Education "Skolkovo Institute Of Science And Technology"
Priority date: 2019-05-24
Filing date: 2019-05-24
Publication date: 2020-12-03
Also published as: WO2020242334A8

Abstract

The disclosure relates to a system and method for the automated determination of the characteristics of an electrocardiographic (ECG) in conditions of a high noise signal. The disclosure comprises the steps of receiving an ECG signal using an electrocardiogram recording device and providing a smoothed ECG signal by filtering the noise of the received ECG signal, so that the smoothed ECG signal is as close as possible to the received ECG signal without distorting the experimental patterns of the ECG curve, and further comprises the steps of calculating the first derivatives of the smoothed ECG signal and determining the characteristics of the ECG based on the first derivatives of the smoothed ECG signal. In addition, the R wave is automatically separated from V pathology based on changes in the relative position of the local maxima and minima of the first and second derivatives.

Description

SYSTEM AND METHOD OF AUTOMATED ELECTROCARDIOGRAM ANALYSIS

AND INTERPRETATION

Field of the invention

The invention relates to a system and method for automated extraction of the morphology of the electrocardiographic signal, representing the ECG signal as a sequence of cardiocycles, determining characteristic waves for specialized devices for monitoring cardiac activity, namely for stationary cardiac monitors, Holier ECG monitors, wearable ECG recording devices, mobile devices.

Background art

Diagnostics of the human cardiovascular system is one of the most important tasks of cardiology. The main causes of death for people of working age are associated with cardiovascular diseases. This explains the need to develop and improve monitoring instruments for objective assessment and prediction of the state of the cardiovascular system. Currently, an electrocardiogram (ECG) is the most common method for diagnosing the functioning of the cardiovascular system of a person. The ECG signal carries a large amount of information, and detailed automatic analysis of the ECG signal allows timely generation of alarms, as well as making predictive conclusions.

Automatic analysis of electrocardiogram is a rather complicated theoretical problem. This is primarily due to the physiological origin of the signal, which causes its non-determinism, diversity, variability, unpredictability, and susceptibility to numerous types of interference. One of the main reasons for the ineffective diagnosis of the heart condition is intense interference of various types - contour drift, movement artifact, muscle tremor, and network disturbance, which distort information about the heart condition.

The problem of extracting a useful signal against the background of a whole complex of interference and distortion is one of the main problems in conducting modem electrocardiological studies. Interference filtering can introduce distortions into the ECG and lead to an interpretation error. The presence of artifacts in the electrocardiological signal significantly complicates its analysis and the identification of diagnostic signs. When solving this problem, the difficulty lies in the choice of filtering methods to eliminate a certain type of artifacts. Therefore, not only the invisible to the user internal filtering, in which the useful signal is distorted to some extent, is important; but also the removal of noise from the signal received from the recorder since it will be used to diagnose it.

At the moment there are several solutions aimed at highlighting the characteristics of the electrocardiographic curve and the subsequent electrocardiogram interpretation. US 2013/0190638A1, Motion and nose artifact detection for ECG data, discloses a method aimed at overcoming the main disadvantages of popular methods of filtering ECG signals associated with the need to build models, determine thresholds and optimal parameters. The method is based on the separation of pure parts of the ECG signal from motion and noise artifacts, prevents distortion of the location of the R wave peaks and, using R-R intervals selected after separation of the artifacts, analyzes for the presence of atrial fibrillation.

Empirical mode decomposition (EMD) is used to detect motion and noise artifacts in real time, in which the ECG signal is decomposed into the sum of modal functions (IMF). Each subsequent IMF has a lower frequency than the previous one. Thus, artifacts can be isolated, since they are mainly focused on a high frequency. At the next stage, statistical indicators are used, including the Shannon entropy, to characterize randomness, the mean value of the mode functions, and the root-mean-square value of the R-R difference sequence. Based on the comparison of these three indicators with predetermined thresholds, a decision is made whether the ECG signal contains artifacts or not.

One of the significant drawbacks of EMD, which reduces the quality of decomposition, is the frequent occurrence of mixing mode functions, which complicates the frequency separation and reduces its quality (Wu Z., and N.E. Huang (2009), Podladchikova T., R.A.M. Van der Linden, and A.M. Veronig (2017), Ensemble empirical mode decomposition: a noise-assisted data analysis method, Advances in Adaptive Data Analysis, Vol. 01, No. 01, pp. 1-41, doi.org/10.1 142/S1793536909000047). The components of a multicomponent signal with a certain effect of destabilizing factors (noise, impulse noise, etc.) and neighboring components close in frequency can“flow” into separate mode functions of neighboring IMFs at individual time intervals. In addition, to ensure reliable operation of these algorithms, it is necessary to adjust the thresholds for detecting artifacts in the ECG signal for each patient, which reduces the resistance of the method to interference and ECG variability.

US 2017/0112401 Al, Automatic method to delineate or categorize an electrocardiogram, discloses a method of delineation and classification of the ECG signal based on a convolutional neural network. This method has expanded the capabilities of neural networks to interpret ECG, representing not one, but more anomalies. The method allows determining the time of the beginning and end of each wave: these are the sections between the characteristic points: base of the P wave, QRS complex, and T wave. The neural network displays a list of the anomalies found in the form of vector points for anomalies that exceed a predefined threshold.

The disadvantages of convolutional neural networks are a large number of variable network parameters. The varied parameters include the number of layers, the dimension of the convolution kernel for each of the layers, the number of cores for each of the layers, the pitch of the kernel shift during processing the layer, the activation function, the parameter for highlighting anomalies, etc. All these parameters significantly affect the result, but they are selected empirically. Of great importance is also a set of training data. In addition, difficulties arise with the accuracy of determining temporal characteristic segments. The loss of accuracy systematically leads to a partial mixing of the QRS complex with the T wave, erasing the ST segment on the timeline. The limitations of this method include the fact that it does not include an estimate of the wave amplitudes (ECG voltage) characterizing the excitability of certain sections of the myocardium and is not intended to determine the basic R-R interval, the variations in the duration of which determine the rhythm of the cardiac activity.

To clarify the position of the QRS complex, a signal derivative can be used (see EP 2 676 604 Al, 2013, Real-time QRS duration measurement in electrocardiogram). The method involves extracting a portion of the ECG signal around the QRS peak, based on the total duration of the cardiac cycle pattern and the normal duration of the QRS. After low-frequency interpolation (filtering), the extracted signal is normalized so that its maximum (corresponding to the R wave) is equal to one. The normalized signal is filtered by exponentiation and then differentiated. Two equations of straight lines are formed using the points of maximum and minimum of the derivative signal and the R wave. The points of intersection of these lines with the baseline approximately determine the boundaries of the QRS complex, i.e. points Q and S. The disadvantage is the high sensitivity of the algorithm to the baseline shift and a decrease in the accuracy of determining the characteristic points Q and S due to the drift of the baseline. Applying the method of estimating the duration of QRS will lead to a failure if the amplitude of the R wave is lower than the amplitude of the T wave, or if pathology V appears in the ECG signal, a sign of monomorphic ventricular tachycardia. This method does not provide any identification of the P wave, the main electrographic sign of manifestation of atrial fibrillation, and T wave, which reflects the phase of recovery of the muscle tissue of the heart ventricles between myocardial contractions.

Substance of the invention

A method is proposed for representing the ECG signal as a sequence of cardiocycles, automatic selection of an R-R sequence of intervals to identify possible cardiac rhythm disturbances, and heterogeneity of contractions in strength and frequency. In particular, methods are presented for diagnosing and analyzing heart diseases and conditions such as atrial fibrillation by determining the presence or absence of the P wave, the main electrographic feature of manifestation of atrial fibrillation, and isolating monomorphic gastric tachycardia (pathology V) when premature ventricular contractions are observed.

Methods for detecting a heart rhythm disorder include noise removal from an ECG signal obtained from a recording device. After receiving the ECG signal using an electrocardiogram recorder, it is being smoothed. Smoothing differs in that it does not distort the dynamics of an ECG signal, since it does not require preliminary assumptions about its dynamics, which usually, in varying degree, can lead to noticeable temporal and amplitude distortions of characteristic points of the ECG signal. It is possible to preserve all the information about the original signal and at the same time filter out noise, abrupt (steplike) changes that do not contain a useful signal.

The high noise immunity of the filtering method makes it possible to use the first and second derivatives of the smoothed (filtered) signal without distortion by noise to delineate the ECG signal and extract R, T and P waves. The effect of noise amplification in calculating derivatives when using this filtering method is neutralized.

The characteristic R, T, and P waves of an ECG signal are determined by applying simple logic rules with respect to the first and second derivatives, which are estimated using the first and second differences of the smoothed ECG signal. The first derivatives eliminate errors due to the displacement of the baseline of the ECG signal and approach the stationary process. The use of the first derivatives to find the R, T and P waves also eliminates the main disadvantages of the method of isolating local extrema in analyzing the shape of the electrocardiogram, namely the need to set the sensitivity threshold and skip the wave on inclined signal sections.

Next, a histogram of the distribution of all local maxima and minima of the first derivatives is formed, which allows us to automatically determine the ranges of changes in extremes corresponding to the R, T and P waves, and highlight the maxima and minima of the first derivatives for these waves. Automatic analysis of the histograms of the distribution of local minima and maxima of the first derivatives allows us to quickly automatically detect signs of a large number of pathologies in the ECG signal and the patient’s critical state, significantly minimizing the processing time of the ECG signal for automatic decision making.

To ensure the effectiveness and guarantee the right decision in the automatic mode, the relative position of the local extrema of the second derivatives of the smoothed ECG signal is additionally analyzed.

The criteria for changing the relative position of local maxima and minima of the first and second derivatives are used to separate the R wave automatically from pathology V when there is a premature ventricular contraction. The number of erroneous decisions varies from 3% to 30% depending on the complexity of the ECG signal.

The accuracy of ECG signal detection based on the proposed approach varies from 95% to 99.9%, depending on the complexity of the original ECG signal. The limitations of this approach are situations when the same ECG waveform is interpreted by a specialist differently depending on additional features of the ECG signal, as well as other leads. There is also a number of unidentified signals that are interpreted individually. An important advantage of the proposed method is the ability of a flexible and quick response to changes occurring in the ECG curve during the observation process, which is a necessary requirement for processing an ECG signal in automatic mode. These methods form the basis of the described method for analyzing the ECG signal, which has reliability, flexibility and ease of implementation.

In one embodiment of an automated electrocardiogram characterization system, an electrocardiogram recorder is a stationary cardiac monitor, a Holier ECG monitor, a wearable ECG recorder, or a mobile device with an ECG recording function.

Brief description of drawings

Fig. 1 - Block diagram of the algorithm for the automatic analysis of the ECG signal;

Fig. 2 - Fragment of recording the ECG signal 111 obtained by means of an electrocardiograph from the MIT-BIH Arrhythmia database and the corresponding smoothed ECG signal;

Fig. 3 - Fragment of recording 111 of the ECG signal obtained by means of an electrocardiograph from the MIT-BIH Arrhythmia database and the corresponding first derivatives of the smoothed ECG signal;

Fig. 4 - Histograms of the distribution of local maxima and minima of the first derivatives and smoothed histogram values for recording 111 (a, b) and recording 203 (c, d) from the MIT- BIH Arrhythmia database;

Fig. 5 - The first (a) and second derivatives (b), as well as the sum of the second derivatives of the smoothed ECG signal (c) for a fragment of the ECG signal with a marked R wave.

Fig. 6 - The first (a) and second (b) derivatives, as well as the sum of the second derivatives of the smoothed ECG signal (c) for a fragment of the ECG signal with marked pathology V.

Fig. 7 - The first derivatives for a fragment of the ECG signal with marked R, T and P waves.

Implementation of the invention

The algorithm for an automatic analysis of the ECG signal consists of a number of steps shown in the block diagram (Fig. 1). Further implementation of the invention is disclosed in detail.

1. ECG signal smoothing

The main obstacle to an automated ECG analysis is a lot of disturbances, as well as high level of a noise signal typical for wearable electronics. The basis of the proposed approach to automated ECG analysis is finding local amplitude extrema based on extracting local extrema of the first and second ECG signal derivatives. On the one hand, after taking the derivatives, the original nonstationary signal approaches the stationary one, which increases the reliability of the selection of its local extrema. However, the first and, even more so, the second derivatives amplify the interference of the ECG signal received, as from stationary electrocardiographs, ECG Holters, and from mobile devices. Therefore, it is necessary to carefully filter out the original signal in order to neutralize the effect of noise amplification when taking derivatives and at the same time not to introduce dynamic distortions into the useful signal. Podladchikova T., R.A.M. Van der Linden, and A.M. Veronig (2017), Sunspot number second differences as precursor of the following 11-year sunspot cycle, The Astrophysical Journal, 850, 81, doi.org/10.3847/1538- 4357/aa93ef proposed an effective noise filtering method, based on the balance of smoothness and proximity of the curve of the smoothed values to the original curve. This method does not require preliminary assumptions and assumptions about the dynamics of the process and allows us to reproduce the initial experimental patterns regardless of assumptions about the process model, thereby minimizing the risk of distorting important process features, erasing important small details and conclusions, which is of great practical importance for building automated ECG processing curve in conditions of high noise level of ECG signal received from wearable devices, mobile devices, ECG-Holters, stationary cardiographs.

In order to make the selected useful signal as close as possible to the received ECG signal, but at the same time to be insensitive to random deviations and fluctuations of the measured signal, it is proposed to look for smoothed values that are optimal according to a criterion that includes the requirements of the desired curve smoothness and its proximity to the experimental ECG curve. The level of confidence in the received ECG signal is achieved by the requirement that the smoothed curve be close to the original data based on minimizing the sum of squares of deviations. The smoothness level of the smoothed curve is achieved on the basis of minimizing the second derivatives of the original ECG signal.

Thus, the smoothed X_t estimate of the ECG signal z_\ is found from the minimization of the functional J.

Here, b is the smoothing coefficient, which determines the proximity of the smoothed ECG signal to the original ECG curve. As b increases, the smoothed signal approaches the original ECG curve, but at the same time the degree of smoothing decreases. As b decreases, the degree of smoothing increases, but the deviation from the original ECG curve increases. Term

J' ¹ (Z, - X_t)² characterizes the deviation indicator, which is responsible for the proximity of the smoothed estimates X to the original ECG signal _¾. Term ² (X ₊₂ -2X_J+l + X/)² shows the variability indicator X_t being responsible for the smoothness of the smoothed curve.

The evaluation of the smoothed ECG signal is reduced to solving the following set of equations:

= bA- Z (2)

This approach to noise filtering and optimization of the smoothing algorithm does not require preliminary assumptions and presuppositions, often based on the developers' subjective ideas about the ECG signal process under investigation, which can lead to false conclusions.

Fig. 2 shows a fragment of the ECG signal obtained by means of an electrocardiograph from the MIT-BIH Arrhythmia database, record 111 (dashed curve). The solid line shows the smoothed ECG signal obtained from equation (2).

Effective filtering of noise without distorting the true dynamics of an ECG signal allows us to estimate with high accuracy the first and second finite ECG curve differences, which by their nature enhance the noise component in the data.

Fig. 3 shows the first derivatives of the ECG signal (solid curve), obtained on the basis of a smoothed ECG curve, which characterize the rate of change of the ECG signal. The corresponding original ECG signal is shown as a dotted curve. As can be seen from Fig. 3, the original ECG signal (dotted curve) is a non-stationary process, while the first derivatives of the smoothed ECG signal approach a quasi-stationary process. While the ECG signal between the RR intervals on the dotted curve is characterized by strong variations and shifts, the dynamics of the first derivatives of the smoothed ECG signal (solid curve) vary around a constant level. The exclusion of the amplitude trend of a smoothed ECG signal reduces by 5 times or more the scatter of local maxima of the solid curve of the corresponding R, T and P waves in comparison with the dotted curve.

As a rule, the ECG signal received from wearable mobile devices is characterized by a much greater degree of noise and scatter of data. Nevertheless, even under conditions of high noise, reliable estimates of the first derivatives, obtained on the basis of effective noise suppression of the original ECG signal, characterize the rate of change of the ECG signal and represent a quasistationary process, the analysis of which greatly simplifies the analysis of the ECG curve, the identification of stable patterns and the identification of pathologies in automatic mode.

2. Histogram of local extrema distribution

As can be seen from Fig. 3, the highest values of the local minima and maxima of the derivatives of the smoothed ECG signal are observed in the vicinity of the R wave. The R wave is characterized by a rapid growth and a rapid fall in the first derivatives of the ECG signal. The rapid growth of the ECG signal to the top of the R wave is characterized by a large positive local maximum of the first derivatives. The fall, which follows the apex of the R wave, is characterized by a large negative local minimum of the first derivatives. To automatically select large local maxima and minima of the first derivatives, it is proposed to analyze a histogram of the distribution of all local maxima and minima of the first derivatives of the ECG curve.

Figs. 4a and 4c are histograms of the distribution of local minima in the first derivatives of the smoothed ECG signal for recording 111 and 203 (MIT-BIH Arrhythmia Database). The X- axis reflects the absolute values of the local minima of the first derivatives, and the Y-axis reflects the frequency of observation of local minima of various sizes.

In Fig. 4a, the elevation indicated by the arrow characterizes the neighborhood of the peaks of the R wave, which are distinguished by the largest negative local minima of the first derivatives. To automatically determine the range of local minima, the values of the histogram of the distribution corresponding to the R waves are smoothed further based on equation (2), and the results are shown in Fig. 4b (record 111) and Fig. 4d (record 203). In Fig. 4b, the local minimum of the smoothed histogram located between its last local maximum, to which the arrow is directed, and the previous one (second from the end), automatically indicates the threshold that exceeds all absolute values of the local minima of the first derivatives corresponding to the R waves. Similarly, the range of variations local maxima of the first derivatives corresponding to the R wave is automatically determined. The areas of growth and decrease in the rate of change of the ECG signal, which is slower than the R wave, are also characteristic of other waves, for example, the T wave. The local maximum (the second from the end) reflects the values of the local minima of the first derivatives, corresponding mainly to the vicinity of the T wave. In order to separate uniquely large local maxima and minima of the first derivatives corresponding to the neighborhood of the R wave, a normalization coefficient is introduced based on the value of the estimated threshold.

As can be seen from Fig. 4b (record 111), smoothed histogram values have at least three pronounced local maxima. The ECG signal (record 203) is characterized by a large number of pathologies and is difficult even for visual diagnostics by doctors. At the same time, Fig. 4d (record 203) is characterized by only two local maxima, which automatically confirms the presence of a large number of pathologies in the ECG signal. Automatic analysis of the histograms of the distribution of local minima and maxima of the first derivatives opens up the possibility of fast automatic detection of signs of a large number of pathologies in the ECG signal and the patient’s critical state, minimizing the processing time of the ECG signal for automatic decision making.

3. Determination of time and amplitude parameters of the R wave peaks

Fig. 5a shows a fragment of the ECG signal for recording 1 11 (solid curve) and the first derivatives of the smoothed ECG signal (dashed curve). The local maximum and minimum of the first derivatives, determined automatically based on the analysis of the histogram of their distribution, are marked by a marker and the numbers 1 and 2, respectively. The top of the R wave will always be between the local maximum (1) and the minimum (2) of the first derivatives of the smoothed ECG signal, while the maximum (1) is always to the left of the minimum (2) of the first derivatives, which opens the possibility of its fast-automatic detection.

To ensure a guaranteed result in the automatic mode, the second derivatives of the smoothed ECG signal are additionally analyzed (Fig. 5b, dashed curve). These additions are required mainly in case of pathology or to eliminate the effects of noise residues. A large negative local minimum of the second derivatives (marker 2) reflects a sharp drop in the rate of change of the ECG signal in the vicinity of the R wave.

A large positive local maximum of the second derivatives (marker 1) demonstrates an increase in the rate of change of the ECG signal following the R wave. To guarantee the allocation of the R wave, it is required that the local minimum (2) is always to the left of the local maximum of the second derivatives (1).

For additional guarantees of operation in the automatic mode, the sum of the second derivatives of the smoothed ECG signal is also analyzed (Fig. 5c, dashed curve), which represents the integral characteristic of the increment of the rate of change of the ECG signal at a long stage. To guarantee the allocation of the R wave, it is also required that the local maximum (1), which shows a significant increment of speed before the characteristic point R, is always to the left of the local minimum (2) the sum of the second derivatives.

4. Detection of premature ventricular contraction (V)

Since in some cases a premature ventricular contraction (V) is observed in the work of the heart muscle instead of the R wave, it is necessary to separate the R wave from the pathology V automatically. Fig. 6 shows a fragment of an ECG signal for recording 203 (solid curve), first derivatives (a), second derivatives (b), and the sum of the second derivatives (c) of the smoothed ECG signal (dashed curve). The local maximum and minimum of the first derivatives, determined automatically based on the analysis of the histogram of their distribution, are marked by a marker and the numbers 1 and 2, respectively.

Thus, in the case of pathology V, a sharp drop in the ECG signal precedes its growth, then the local maximum (1) is located to the right of the local minimum (2) of the first derivatives, which is the opposite of the situation with the R peak. (2) and maximum (1) of the first derivatives. Similarly, the situation is opposite to the situation with R peak, is observed for the local minimum (2) of the second derivatives (Fig. 6b), which is always to the right of the local maximum (1), and for the local minimum (2) of the sum of the second derivatives (Fig. 6c), which is always to the left of the corresponding local maximum (1).

To isolate automatically the R wave and pathology V, it is necessary to identify the above signs of the behavior of local minima and maxima of the first and second derivatives, as well as the sum of the second derivatives of the smoothed ECG signal. In case of discrepancy of the indicated signs, the algorithm is additionally trained on the basis of the analysis of the neighboring local minima and maxima. The number of such cases varies from 3 to 30% depending on the complexity of the ECG signal.

5. Determination of time and amplitude parameters of T and P waves

The proposed approach also underlies the automatic extraction of T and P waves. Fig. 7 shows the ECG signal fragment for recording 111 from the MIT-BIH database (solid curve), as well as the first derivatives of the smoothed ECG signal (dashed curve).

As was shown above, the local maxima (1) and minima (2) of the first derivatives corresponding to the neighborhood of the R wave are the most pronounced among the peaks of all the waves. The next, smaller in absolute value, are local maxima (G) and minima (2') of the first derivatives in the vicinity of the top of the T wave. The third position in absolute value is occupied by local maxima (1 ") and minima (2”) of the first derivatives, surrounding the top of the P wave. The peeks of the T and P waves, similar to the R and T waves, are automatically separated between the corresponding local minimum and maximum of the first derivatives of the smoothed ECG signal.

The accuracy of ECG signal detection based on the proposed approach varies from 95% to 99.9%, depending on the complexity of the original ECG signal. The application of this approach is limited in a situation where the same ECG waveform is interpreted by a specialist differently depending on additional features of the ECG signal, as well as other leads. There is also a number of unidentified signals that are interpreted individually. However, the proposed algorithm automatically indicates difficult parts, which allows medical personnel to minimize the time spent on manual signal processing for subsequent diagnostics and clinical decision making.

An important advantage of the proposed method is the ability of a flexible and quick response to possible changes in the shape of the cardiac cycle occurring during the observation process, which is a necessary requirement for processing an ECG signal in automatic mode. These methods form the basis of the described method for analyzing the ECG signal, which has reliability, flexibility and ease of implementation.

The flexibility and ease of use of the proposed method allow minimizing the processing time of the ECG signal for decision-making in automatic mode and reliable automated diagnosis of cardiovascular diseases.

It is possible to combine a wearable device with the electrocardiogram recording option and software for direct analysis of the ECG morphology of the curve.

The given task of processing the ECG signal is solved by using both the methods of the phenomenological approach and the methods of statistics. Statistical processing of experimental data, as a rule, is aimed at building a mathematical model of the object under study. However, taking into account all secondary parties and relationships in the model can distort the description of the process, since it requires many provisions and assumptions, as well as distorting important features of the process, erasing important small details, worsening results and false conclusions in the context of insufficient information about the object of study. In conditions when the assumptions when using the model are fraught with the risk of distorting the conclusions about the process, it is necessary to develop methods of approximation of experimental data that do not require assumptions about the processes under study and reproduce the experimental pattern regardless of the assumptions about the model. This approach to noise filtering and optimization of the smoothing algorithm does not require preliminary assumptions and presuppositions, often based on the‘developers' subjective ideas about the ECG signal process under investigation, which can lead to false conclusions. The proposed solutions are confirmed experimentally and compared in practice with the known methods.

Effective filtering of noise without distorting the true dynamics of the ECG signal allows us to estimate with high accuracy the first and second derivatives of the process, which by their nature enhance the noise component in the data. Analysis of the derived ECG signal greatly simplifies the analysis of the ECG signal, the selection of stable patterns and the identification of pathologies in the automatic mode. An important feature of the proposed method is the ability of a flexible and fast response to changes in the shape of the cardiac cycle occurring during the observation process, which is a necessary requirement for processing an ECG signal in automatic mode. These methods underlie the described software for analyzing the ECG signal, which has reliability, flexibility, and ease of implementation.

Efficient revealing the ECG signal regularities makes it possible to analyze the choice of the optimal predicative model in the conditions of the high variability of the signal under study in order to build a short-term and long-term automated medical forecast.

Claims

1. A method of determining the characteristics of the electrocardiogram (ECG), the method comprising:

recording an ECG signal using an electrocardiogram recorder;

providing, by processing means, a smoothed ECG signal by filtering the noise of the received ECG signal;

characterized in that the method comprises:

performing the ECG signal smoothing so that the smoothed ECG signal is as close as possible to the received ECG signal;

calculating, by the processing means, the first derivatives of the smoothed ECG signal; determining, by the processing means, the ECG characteristics based on the first derivatives of the smoothed ECG signal.

2. The method of determining the ECG characteristics according to claim 1, further comprising:

determining, based on processing the distribution of local extrema of the first derivatives of the smoothed ECG signal, by the processing means, a threshold for distinguishing the range of extrema corresponding to different types of waves; and

determining, by the processing means, the characteristics of the peaks of waves of different types based on selected ranges of extrema changes and the relative position of the local maxima and minima of the first derivatives of the smoothed ECG signal.

3. The method of determining the ECG characteristics according to claim 2, further comprising:

calculating, by the processing means, the second derivatives of the smoothed ECG signal; and

determining, by the processing means, the characteristics of the peaks of the waves of different types further based on the local maxima and minima of the second derivatives of the smoothed ECG signal.

4. The method of determining the ECG characteristics according to claim 2 or 3, further comprising:

identifying, by the processing means, a possible pathology based on the determined characteristics of the peaks of different types of waves.

5. The method of determining the ECG characteristics according to claim 2 or 3, wherein determining the top of the R wave between a local maximum and a local minimum of the first derivatives of the smoothed ECG signal taking into account the relative position of the local maximum and the local minimum.

6. The method of determining the ECG characteristics according to claim 5, wherein R wave is distinguished from V pathology on the basis of changes in the relative position of the local maximum and local minimum of the first and second derivatives of the smoothed ECG signal.

7. The method of determining the ECG characteristics according to claim 2 or 3, wherein determining the peaks of the T waves and P waves between local maxima and local minima of the first derivatives of a smoothed ECG signal, while distinguishing R waves, T waves and P waves by the absolute value of the difference between the corresponding local maxima and local minima.

8. An electrocardiogram (ECG) characterization system, comprising:

an electrocardiogram recording device, configured to receive an ECG signal;

processing means configured to:

smooth the ECG signal by filtering the noise of the received ECG signal;

characterize in that the processing means is additionally configured to:

smooth the ECG signal so that the smoothed ECG signal is as close as possible to the received ECG signal;

calculate the first derivatives of a smoothed ECG signal; and

determine the characteristics of the ECG based on the first derivatives of the smoothed ECG signal.

9. The ECG characterization system according to claim 8, wherein the processing means is further configured to:

determine, on the basis of processing the distribution of local extrema of the first derivatives of a smoothed ECG signal, the threshold for selecting the range of changes of extrema corresponding to different types of waves; and

determine the characteristics of the peaks of different types of waves based on selected ranges of extrema changes and the relative position of the local maxima and minima of the first derivatives of the smoothed ECG signal.

10. The ECG characterization system according to claim 9, wherein the processing means is further configured to:

determine the second derivatives of the smoothed ECG signal; and

determine the characteristics of the peaks of different types of waves further based on local maxima and minima of the second derivatives of the smoothed ECG signal.

11. The ECG characterization system according to any one of claims 8 to 10, wherein an electrocardiogram recorder is one of a stationary cardiac monitor, a Holter ECG monitor, a wearable ECG recorder, and a mobile device with an ECG recording function.