RU2756157C1 - Method for analysis of ballistocardiographic signal for detecting singular heart beats in real time - Google Patents

Abstract

FIELD: medicine.

SUBSTANCE: invention relates to cardiology, in particular, to a method for detecting singular heart beats (cardiac cycles) from the ballistocardiogram signal in real time. The proposed method for detecting consists in the fact that the digital signal of the measured ballistocardiogram is filtered using a bandpass filter with a bandpass in the range of location of patterns of cardiac cycles, all extrema of the processed signal are determined, the standard deviation σ(i) is calculated at each i^th extremum of the ballistocardiogram signal in a ±50 ms window:

wherein N is the amount of samples in the ±50 ms window,

is the average value of the signal in a ±50 ms window, and the standard deviations are analysed at each extremum in order to determine the estimated location of cardiac cycle, wherein the cardiac cycle is considered detected if there is no extremum in the ±300 ms window with a value of standard deviation above the current σ(i), the estimated locations of cardiac cycles are corrected based on a model template calculated based on the previously detected heart beats.

EFFECT: invention provides an increase in the accuracy of processing the ballistocardiogram signal as compared to the existing state of the technology for the purpose of detecting cardiac cycles in real time.

6 cl, 2 dwg

Description

The technical field to which the invention relates

The invention relates to cardiology. In particular, the invention relates to a method and a device for detecting single heart beats (cardiocycles) from a balistocardiogram signal. The invention can be used both for medical purposes and non-medical purposes, in particular, for assessing the physiological state of a person in order to inform or monitor the user.

Background of the invention

Currently, algorithms for analyzing ballistocardiographic signals (hereinafter BCG signals) in order to determine the frequency of heart contractions use spectral methods, or methods of determination in time to detect multiple manifestations of certain patterns, for example, by assessing the autocorrelation function of the signal. In all such approaches, signal segments with a duration of a few seconds should be considered so as to cover multiple heartbeats. As a result, information is available on the average rate of heart contractions over a certain period of time, but accurate information in the interval between successive heart contractions is not available.

Algorithms for evaluating sequential heart contractions based on BCG signals are known, but such approaches for detecting single heart beats from a balistocardiography signal have a number of disadvantages.

Known solutions use either one characteristic signal indicator and do not allow detecting cardiocycles with sufficient accuracy ([1]), or a large number of signal characteristic indicators, the calculation of which can take significant time on embedded systems ([2], [3]).

Other solutions use cluster, correlation or wavelet-like computations over the entire signal, which is also a problem in terms of computation speed for sufficiently long sessions or application in real-time systems even on modern embedded systems ([4], [5]).

The use of machine learning also does not achieve the required accuracy of detecting single cardiocycles ([6]), while, like a number of other methods using the accumulated database of templates ([7]), it requires additionally storing the database locally or having constant access to it.

The closest analogue to the claimed invention in technical essence is a method for analyzing BCG signals ([8]), according to which the duration of the time interval between successive heart contractions is calculated, as in the case of an ECG. To do this, the frequency filtering of the signal is performed in the range of the heart rate pattern, the characteristic indicator is calculated according to the signal (characteristic indicators: high-frequency component λ (t), long-term prediction μ (t) and the probability distribution function σ (t)), determine the place of heart beat by finding the maximum value of the polynomial αλ (t) + βμ (t) + χσ (t), where α, β and χ are scalar quantities, for t∈ [tn + tmin, tn + tmax]. This maximum value is preferably determined by lowpass filtering the addition result following the maximum search. The maximum value thus determined becomes the next estimated heart rate tn + 1, and the iterative process continues from this estimated value until the end of the signal is reached. The result is the first segmentation of the ballistocardiographic signal into intervals of heartbeats, which can be used to provide an initial value for granularity methods that find the exact times of heartbeat events or the duration of heartbeat intervals.

The disadvantages of this method are:

1) inability to work on microcontrollers or embedded systems in real time,

2) low speed of detection of cardiocycles, and, consequently, the inability to use the proposed method of analysis on slow systems, for example, microcontrollers or embedded systems.

3) no correction after detection with previous shocks, which leads to lower detection accuracy:

o heart beats in subjects with an unstable pattern (who have similar H and J or J and L peaks in amplitude),

o heartbeats during deep breathing,

o heartbeats with a partially noisy pattern or in the presence of artifacts lasting 1-2 seconds,

o heart beats in subjects who have strongly expressed diastolic waves (LMN complex) in the ballistocardiogram signal.

Thus, there is a need for an improved method for analyzing a ballistocardiogram signal for detecting single heart beats in real time with high detection accuracy that can be applied with minimal latency in embedded systems.

The essence of the invention

A technical problem of the present invention is the lack of simple and accurate methods for analyzing ballistocardiogram signals from arbitrary subjects for detecting single heart beats in real time, which can be used in embedded systems.

The technical result of the present invention, achieved when using it, is to improve the accuracy of the method for analyzing the signal of a ballistocardiogram in comparison with existing processing algorithms and to simplify its use in embedded systems.

The technical result is provided due to the fact that the method includes the initial detection of heart beats and their subsequent correction by the method of correlation with the template calculated on the basis of previously detected heart beats in the same patient.

This method requires only 300 ms after a heartbeat to detect it and can be applied with minimal latency in embedded systems. does not require heavy computation.

The stages of the proposed, according to the present invention, the algorithm:

1. Filtration of the ballistocardiogram signal with a bandpass filter with a passband from 1-4 Hz to 15-35 Hz, which is the range of finding the heart rate pattern;

2. Finding the extrema of the ballistocardiogram signal (upper and lower peaks);

3. Calculation of the standard deviation at each extremum of the signal in the ± 50ms window;

4. Analysis of the obtained standard deviations at each extremum: if there is no extremum with a large standard deviation in the ± 300 ms window, then we consider it to be a cardiocycle;

5. Correction of cardiocycles based on a model signal with a 450ms search window and a 350ms correlation window.

Compared to the closest analogue in the proposed method, only one parameter is calculated by the signal, and it is very simple to calculate, in contrast to the calculation in the analogue of three indicators, which are much more difficult to calculate (for example, obtaining a spectrogram in the high-frequency component λ (t)) ... In this case, for the calculation of indicators in the proposed method, signal extremes are used, in contrast to the analogue, in which the calculation of indicators is carried out over the entire signal.

To improve the detection accuracy, in addition to the optimal parameters of the algorithm, in the proposed method, the current cardiac cycle is corrected based on the previous ones.

Thus, the improved algorithm retains only the idea of calculating the absolute deviation. The search function was redesigned (only the absolute deviation at the peaks is considered), which made it possible to make a decision about the cardiocycle if there is no peak with a large absolute deviation around the ± 300 ms peak.

Brief Description of Drawings

The accompanying drawings, which are incorporated and form part of the present specification, illustrate embodiments of the invention and, together with the foregoing general description of the invention and the following detailed description of the embodiments, serve to explain the principles of the present invention.

FIG. 1 shows the cardiocycle in the signal of the balistocardiogram. The letters denote peaks reflecting the strikes of the blood wave against the atria, ventricles and arterial bifurcations, corresponding to the presystolic, systolic or diastolic wave.

FIG. 2 shows a block diagram of an algorithm for detecting cardiocycles.

Detailed description of the invention

In the present description and in the claims, the terms "includes", "including" and "includes", "having", "equipped", "containing" and their other grammatical forms are not intended to be construed in an exclusive sense, but, on the contrary, are used in a non-exclusive sense (that is, in the sense of "including"). Only expressions of the type “consisting of” should be considered as an exhaustive list.

By "embedded systems" in the present description is meant a specialized microprocessor control, control and monitoring system, the design concept of which is that such a system will work, being embedded directly into the device that it controls. It is assumed that embedded systems, unlike modern computers, are limited in processor and / or memory power and therefore do not allow the use of methods with heavy computations (wavelets, a large number of transformations, etc.) in real time.

In general, the present invention is directed to the development of a method for analyzing a balistocardiogram signal for detecting single heart beats in real time with high accuracy, sensitivity and minimal average heart rate error. The developed algorithm can be applied in embedded systems, since it does not require heavy computations.

The balistocardiogram displays the movement of the human body caused by heartbeats, represented by a periodic pattern, breathing, and other body movements.

The cardiocycle in the BCG signal has a similar peak complex as the QRS complex in the ECG signal, called the IJK complex (FIG. 1). In contrast to the ECG, the task of recognizing the cardiac cycle in the BCG signal is a more complex process. The cardiocycle in the BCG signal has less pronounced forms, in contrast to the QRS complex of the ECG signal, and is also much more susceptible to distortions, depending on the age, body weight and physiological characteristics of a person. In addition, significant distortions of the BCG signal can be caused by other acoustic / vibrational movements not related to heart rate, movements of the heart, body, organs and the environment.

One of the main ways to determine the position of the cardiocycle on the time axis from the BCG signal is to match the BCG signal with a predetermined model template (for example, the W-shape of the IJK complex). However, due to the large number of different patterns of cardiocycles in the BCG signal in different people, this method has significant drawbacks and cannot be applied universally for all people. Obviously, it will perform better on some data and worse on others.

The present invention seeks to overcome the problems described above, including. The method according to the invention is briefly described with reference to the flowchart shown in FIG. 2.

At step 101, the ballistocardiogram data is collected and pre-processed.

Based on the physiological characteristics of cardiocycles and empirical experiments, it is known that the BCG signal contains the following signals in different frequency ranges: the range of respiratory waves (0.5 - 4 Hz), the range containing patterns of cardiocycles (3 - 35 Hz), high-frequency noises (25 Hz and higher).

Pre-processing of the signal consists in filtering the signal with a bandpass filter with a bandwidth from 1-4 Hz to 15-35 Hz, which allows you to get rid of the low-frequency influences of respiratory waves on the BCG signal and at the same time get rid of high-frequency oscillations that are not related to the cardiocycle itself. Any bandpass filter with a specified bandwidth, which depends on the original data, can be used as a bandpass filter. The filter can be either analog or digital. When using an analog filter, the signal is digitized after step 101, before step 102.

At step 102, carcycycles are detected using an adaptive sliding maximum signal.

To do this, determine the presumptive location of the heart beat on the time axis of the BCG signal.

To calculate the estimated place of the cardiocycle in the BCG signal, the sliding standard deviation of the signal is first calculated in a window of ± 50 ms, after which the position of the cardiocycle is determined by the calculated points. To speed up the calculations, not all points of the signal are used, but only the extrema, since the cardiocycle in the BGK signal is an IJK complex consisting of extrema. The choice of a window of ± 50 ms is optimal for calculating the sliding standard deviation of the signal, since on average it is half the duration between the J-K peaks of the cardiocycle.

Further, if around the extremum in the ± 300 ms window there are no extrema with a standard deviation value higher than the current one, then, presumably, this point is the peak of the cardiocycle. The choice of the window range for searching for extrema with a standard deviation value higher than the current one is due to a long complete cardiocycle, which includes systole and diastole (HIJKLMN complex).

Thus, step 102 can be broken down into sub-steps that look like this:

1. Finding all extrema of the signal, including extrema that characterize the peaks of the IJK complex, by comparing the current sample with the neighboring ones. If the current count x (i) is greater than the previous x (i-1) and simultaneously less than the next x (i + 1), then it is considered to be a local extremum point. In other words, if x (i)> x (i-1) and x (i) <x (i + 1), x (i) is a local extremum of the signal.

в каждом i-ом экстремуме сигнала баллистокардиограммы в окне ±50 мс следующим образом:2. Calculation of the standard deviation

at each i-th extremum of the ballistocardiogram signal in the ± 50 ms window as follows:

- текущее значение сигнала в окне ±50 мс,

- среднее значение сигнала в окне ±50 мс.where N is the number of samples in the ± 50 ms window,

- the current value of the signal in the window ± 50 ms,

- average value of the signal in the window ± 50 ms.

3. Analysis of the obtained standard deviations at each extremum: if there is no extremum in the ± 300 ms window with a standard deviation value higher than the current one, then it is considered that this is the peak of the cardiocycle.

In step 103, the hypothetical locations of the cardiocycles are corrected based on the model template calculated based on the previously detected heart beats. Correction of the cardiocycle based on a template calculated on the same person allows detecting signals from any people, even if they change their body position at the moment of signal pickup.

Correction of the moment of a heartbeat is made only if at least 5 heartbeats have been detected before. In a preferred embodiment, the calculation of the cardiocycle model requires an array of 10 detected heart beats. In the event that fewer than 5 heart beats are detected, the algorithm takes the current extremum calculated in step 102 as the true peak of the cardiocycle and adds it to the cardiocycle array for the template. As soon as 5 or more cardiocycles are accumulated in the array for the template, the correction is applied.

Despite the fact that each cardiocycle in the signal differs from each other, the amplitude of the noise is much less than the pattern of the cardiocycle, and, therefore, it is possible to select such a minimum value of the correlation between the model template and the current cardiocycle, which will make it possible with a sufficient degree of probability to say that in each cardiocycle the same peaks of the IJK complex are determined. Thus, the developed method makes it possible to calculate the same peaks in the cardiocycles themselves: always calculate I peaks or always calculate K peaks of cardiocycles. Because Since the patterns of cardiocycles are different, situations often occur when first peaks are detected, the next cardiocycle has a K peak, the next one again has I peak, etc. Due to this, the heart rate begins to "float", then increasing, then decreasing. This is important for the subsequent analysis of heart rate indicators (for example, LF, HF, SDNN, etc.).

The following indicator is called the intercorrelation function between two signals (template and current):

,

- среднее значение выборок двух сигналов., where

,

is the average value of the samples of two signals.

To correct the moment of a heart beat, the time interval around the putative cardiac cycle is found, in which the correlation function between the putative heart beat and the template has a maximum value.

.Those. for each expected heart beat moment calculated in step 102, consider intervals of 350 ms in the range of ± 450 ms (the interval in which the search is best) and find the interval with the maximum value of the correlation function with the model template

...

- исходный сигнал баллистокардиограммы

- модельный шаблон,

- исследуемое окно.

- скорректированное время детектированного кардиоцикла,

значение функции корреляции между шаблонным и текущим сигналом кардиоцикла в исследуемом окне,

- интервал, в котором лучше всего проявляет себя поиск., where

- the initial signal of the ballistocardiogram

- model template,

- investigated window.

- corrected time of the detected cardiocycle,

the value of the correlation function between the template and the current signal of the cardiocycle in the studied window,

- the interval in which the search manifests itself best.

In step 104, the model template is corrected, in particular, the adaptation of the model template of the cardiocycle is performed using the already detected cardiocycles.

, необходимо скорректировать непосредственно сам шаблон кардиоцикла.After having determined in the previous step 103 the exact time of the cardiocycle

, it is necessary to correct directly the cardiocycle template itself.

мс с центром в точке

, посредством вычисления между ними медианы по времени. Для коррекции модельного шаблона возможно использовать большее количество последних детектированных кардиоциклов. Однако, чем больше количество детектированных кардиоциклов используется в модели кардиоцикла, тем медленнее система будет реагировать на изменения и в то же время есть вероятность возникновения ежемоментных ошибок. Можно использовать вычисление модельного шаблона и по времени, например, за последние 10/30/60 секунд.To correct the moment of a heart beat, a correlation with a model template is used, which is calculated on the basis of the last 10 detected cardiocycles (or less, but not less than 5 last detected cardiocycles) in the window

ms centered at point

, by calculating the median in time between them. To correct the model template, it is possible to use a larger number of the last detected cardiocycles. However, the more the number of detected cardiocycles is used in the cardiocycle model, the slower the system will respond to changes and, at the same time, there is a likelihood of momentary errors. You can use the calculation of the model pattern and by time, for example, for the last 10/30/60 seconds.

будет использоваться в дальнейшем на повторяющемся на всей длительности сигнала БКГ предыдущем шаге алгоритмаThis model cardiocycle template

will be used in the future on the previous step of the algorithm, which is repeated for the entire duration of the BCG signal

Thereafter, the BCG signal processing procedure (steps 102-104) is iteratively repeated from left to right along the time axis, starting from the initial value in the initial part of the BCG signal, and until the end of the BCG signal.

The method for analyzing the ballistocardiogram signal proposed according to the present invention detects single heart beats based on CEBS (https://physionet.org/content/cebsdb/; total of 17 hours of recordings, 69757 heart beats) with a sensitivity of 99.61%, a specificity of 99.57% and a total detection error rate of 0.8%.

This accuracy is achieved through the application of correction for the detection of a heartbeat by the correlation method based on the previously calculated heartbeats of a particular subject, as well as the optimally selected parameters of the algorithm.

Compared to previous works based on CEBS (https://physionet.org/content/cebsdb), the proposed algorithm is the most accurate of all available options ([9-16]).

Below is a general diagram of a computing device that provides data processing necessary for the implementation of the claimed method.

In general, the device contains such components as: one or more processors, at least one memory, data storage means, input / output interfaces, I / O means, networking means, ballistocardiography sensor.

The processor of a device performs the basic computational operations required for the operation of the device or the functionality of one or more of its components. The processor executes the necessary machine-readable instructions contained in the main memory.

Memory, as a rule, is made in the form of RAM and contains the necessary program logic that provides the required functionality.

The data storage medium can be performed in the form of HDD, SSD disks, array raid, network storage, flash memory, optical information storage devices (CD, DVD, MD, Blue-Ray disks), etc. The tool allows you to perform long-term storage of various types of information, for example, the aforementioned files with user data sets, databases containing records of time intervals measured for each user, user identifiers, etc.

The interfaces are standard means for connecting and working with the server side, for example, USB, RS232, RJ45, LPT, COM, HDMI, PS / 2, Lightning, FireWire, etc.

The choice of interfaces depends on the specific version of the device (N00), which can be a personal computer, mainframe, server cluster, thin client, smartphone, laptop, etc.

As means of I / O data in any embodiment of a system that implements the described method, a keyboard must be used. The hardware design of the keyboard can be any known: it can be either a built-in keyboard used on a laptop or netbook, or a stand-alone device connected to a desktop computer, server or other computer device. In this case, the connection can be either wired, in which the connecting cable of the keyboard is connected to the PS / 2 or USB port located on the system unit of the desktop computer, or wireless, in which the keyboard exchanges data via a wireless communication channel, for example, a radio channel, with base station, which, in turn, is directly connected to the system unit, for example, to one of the USB ports. In addition to the keyboard, I / O data can also include: joystick, display (touch screen), projector, touchpad, mouse, trackball, light pen, speakers, microphone, etc.

Networking means are selected from a device that provides network reception and transmission of data, for example, Ethernet card, WLAN / Wi-Fi module, Bluetooth module, BLE module, NFC module, IrDa, RFID module, GSM modem, etc. The tools provide the organization of data exchange via a wired or wireless data transmission channel, for example, WAN, PAN, LAN, Intranet, Internet, WLAN, WMAN or GSM.

A ballistocardiography sensor can provide a ballistocardiographic signal in analog or digital form to the processor of the device. Alternatively, the ballistocardiography sensor may be equipped with an analog-to-digital converter so that the ballistocardiographic signal is digitally fed into the processor. The device can receive a ballistocardiographic signal using any appropriate means, such as a wired or wireless connection, from a ballistocardiography sensor.

The components of the device are linked via a common data bus.

Although the invention has been described with reference to the disclosed embodiments, it should be apparent to those skilled in the art that the specific embodiments described in detail are provided for the purpose of illustrating the present invention only and should not be construed as in any way limiting the scope of the invention. It should be understood that various modifications are possible without departing from the spirit of the present invention.

Bibliography

1. Choe S., Cho W., Simplified real-time heartbeat detection in ballistocardiography using a dispersion-maximum method // Biomedical Research, 2017).

2. WO2010067294A1 METHOD AND APPARATUS FOR THE ANALYSIS OF BALLISTOCARDIOGRAM SIGNALS.

3. Bruser S., Winter S., Leonhardt S., Robust inter-beat interval estimation in cardiac vibration signals // Physiological Measurement, 2013. V.32 (2), p: 123-128.

4. WO2016072940A1 MULTI-CHANNEL BALLISTOCARDIOGRAPHY WITH CEPSTRUM SMOOTHING AND QUALITY-BASED DYNAMIC CHANNEL SELECTION.

5. Brucer C. et al. Improvement of Force-sensor-based Heart Rate Estimation Using Multi-channel Data Fusion // IEEE Journal on Biomedical Health Informatics, 2014.

6. Thakkar H., Sahoo P Towards Automatic and Fast Annotation of Seismocardiogram Signals Using Machine Learning // IEEE Sensors Journal, 2020. Vol. 20 (5).

7. US2013158415A1 BALLISTOCARDIOGRAM ANALYSIS APPARATUS AND METHOD, AND SYSTEM FOR UTILIZING BALLISTOCARDIOGRAM FOR VEHICLE USING THE SAME.

8. WO2010067294A1 METHOD AND APPARATUS FOR THE ANALYSIS OF BALLISTOCARDIOGRAM SIGNALS.

9. Choudhary T., Sharma L. N., Bhuyan M. K., Automatic Detection of Aortic Valve Opening Using Seismocardiography in Healthy Individuals // IEEE Journal of Biomedical and Health Informatics, vol. 23 (30), 2019, doi: 10.1109 / JBHI.2018.2829608.

10. Siecinski S., Tkacz E., Kostka P. Comparison of HRV indices obtained from ECG and SCG signals from CEBS database // BioMedical Engineering Online 18, vol. 69, 2019. Doi: 10.1186 / s12938-019-0687-5.

11. Thakkar H., Sahoo P Towards Automatic and Fast Annotation of Seismocardiogram Signals Using Machine Learning // IEEE Sensors Journal, 2020. Vol. 20 (5). Doi: 10.1109 / JSEN.2019.2951068.

12. Y. Li, X. Tang, and Z. Xu, “An approach of heartbeat segmentation in seismocardiogram by matched-filtering,” in 2015 7th International Conference on Intelligent Human-Machine Systems and Cybernetics, vol. 2, Aug 2015, pp. 47-51.

13. Choudhary T., Sharma L. N., Bhuyan M. K., Automatic Detection of Aortic Valve Opening Using Seismocardiography in Healthy Individual.

14. Siecinski S., Tkacz E., Kost ka P. Comparison of HRV indices obtained from ECG and SCG signals from CEBS database.

15. Thakkar H., Sahoo P Towards Automatic and Fast Annotation of Seismocardiogram Signals Using Machine Learning.

E. and

F.E., "Robust Detection of AO and IM Points in the Seismocardiogram Using CWT," in IEEE Latin America Transactions, vol. 14, no. 11, pp. 4468-4473, Nov. 2016, doi: 10.1109/TLA.2016.7795816.16. Rivero I.,

E. and

FE, "Robust Detection of AO and IM Points in the Seismocardiogram Using CWT," in IEEE Latin America Transactions, vol. 14, no. 11, pp. 4468-4473, Nov. 2016, doi: 10.1109 / TLA.2016.7795816.

Claims (18)

1. A method for analyzing a ballistocardiographic signal for detecting single heart beats in real time, comprising the steps at which: - carry out preliminary processing of the received ballistocardiogram signal by filtering it using a bandpass filter with a passband in the range of finding the heart rate pattern, - cardiocycles are detected by the method of adaptive sliding maximum of a digital signal, while all extrema of the processed signal are determined, the standard deviation σ (i) is calculated at each i-th extremum of the ballistocardiogram signal in a window of ± 50 ms:

where N is the number of samples in the ± 50 ms window,

- the average value of the signal in the window ± 50 ms,

- the current value of the signal in the window ± 50 ms, and analyzing the obtained standard deviations at each extremum to determine the location of the cardiocycle, and the cardiocycle is considered detected if there is no extremum in the ± 300 ms window with a standard deviation value higher than the current σ (i), - carrying out the correction of the positions of the cardiocycles on the basis of a model template calculated on the basis of previously detected heart beats. 2. The method of claim 1, wherein the model template is calculated based on at least five of the most recently detected cardiocycles. 3. The method of claim 2, wherein the model pattern is calculated based on the ten most recently detected cardiocycles. 4. The method of claim 1, wherein correcting cardiocycle locations comprises the following steps: - determine the correlation function between the signal of the model template and the currently found cardiac cycle

, where

,

Is the average value of the samples, - find the time interval around the supposed place of the cardiocycle, in which the correlation function has a maximum value:

,

, where

- the initial signal of the ballistocardiogram,

- model template,

- the investigated window,

- corrected time of the detected cardiocycle,

the value of the correlation function between the template and current signals of the cardiocycle in the studied window,

- the interval in which the search manifests itself best. 5. The method of claim 1, further comprising adjusting the simulated cardiac cycle after obtaining an accurate cardiac cycle time. 6. The method of claim 1, wherein the bandwidth of the bandpass filter is in the range of 1-4 Hz to 15-35 Hz.

Images