Patent Application
20170172443
The present invention relates to the field of digital processing of a bioelectrical signal, which is characteristic of the cardiac rhythm of a living being, and which is designated in the present text by the term cardiac signal. This is for example, but not exclusively, an electrocardiographic (ECG) signal. In this technical field, the invention relates to the filtering of an RR series obtained by sampling a cardiac signal, with implementation of an automatic quality control of the RR series.
From a physiological perspective, the heart of a living being, isolated from outside influence, contracts automatically and very regularly as does a metronome, under the action of the sinus node which generates an independent nerve impulse and, thereby, causes a spontaneous cardiac muscle contraction. The heart is not, however, isolated; rather, it is connected to the autonomic nervous system (ANS) via parasympathetic and sympathetic systems. The autonomic nervous system influences the activity of the heart: the sympathetic system accelerates the heart rate, while the parasympathetic system slows it down. Thus, despite a certain degree of autonomy, the heart undergoes influences from the autonomic nervous system, which allows, in particular, the body of a living being to adapt the heart rate depending on its needs, however within reasonable limits. It is understood, therefore, that the analysis of the evolution of the heart rate over time, and in particular changes in the heart rate (changes in the heart beat), provides important information on the activity of the cardiac system, particularly on the activity of the autonomic nervous system. Now, knowledge of ANS activity can be of great help in the development of a diagnosis of many clinical situations. On this subject, reference may be made, for example, to the following publication: Lacroix D, Logier R., Kacet S., Hazard J-R, Dagano J. (1992): “Effects of consecutive administration of central and peripheral anticholinergic agents on respiratory sinus arrhythmia in normal subjects, J. of the Autonomic Nervous System”, Vol 39, pages 211-218.
To study these fluctuations in heart rate, various filtering techniques and spectral analysis of a signal representing the evolution over time of the instant heart rate (or frequency) have already been developed since 1970, such signal which is obtained after sampling an analog bioelectrical signal characteristic of the heartbeat of a living being, and termed afterwards “analog cardiac signal.”
To acquire this cardiac signal, different techniques of invasive or non-invasive acquisition are known. One known invasive technique is, for example, to use a blood pressure sensor connected to a catheter inserted into an artery. Among the known non-invasive methods are included, for example, the use of an infrared pulse sensor, or the acquisition of an electrocardiographic (ECG) signal using an electrocardiograph. This latter method of acquiring an ECG signal is in practice the most commonly used to date, because, in addition to its noninvasive nature, it advantageously provides a more accurate signal than that which is obtained, for example, by means of an infrared pulse sensor.
The ECG signal is known as consisting of a succession of electrical depolarizations whose appearance is shown in
In practice, the R wave usually being the finest and most extensive part of the QRS, it is generally used to locate the heart beat with very good accuracy, in practice of the order of one thousandth of a second. Thus, the time interval between two successive R waves accurately characterizes the time separating two successive heartbeats; this is the period of the ECG signal, and the inverse of this period gives the instantaneous heart rate.
To automatically construct the signal, called afterwards the RR series, representing the evolution in time of the instantaneous heart rate, the ECG signal, which is an analog signal (analog/digital conversion of the ECG signal), is sampled, and the sampled digital ECG signal is processed by automatically detecting R waves in this digital signal. An RR series is thus, in a conventional manner, comprised of a plurality of successive RRi samples (or points), each RRi sample being a function of the time interval between two successive R waves of the ECG signal.
However, it should be noted, on the one hand, that the other waves of depolarization (P, Q, S or T) of the ECG signal can also be used for characterizing the heart rate, even if the measurement accuracy is not as good as when using the R waves. On the other hand, depending on the acquisition technique chosen, the cardiac signal may have a different shape from that of the above mentioned ECG signal. The cardiac signal is not necessarily analog, but may be a digital signal. Accordingly, in the present text, the term RR series is not limited to the aforementioned specific definition based on the R waves of an ECG signal, but is defined in a more general way in the context of the present invention as a series of several digital samples called RR, obtained from a cardiac signal that is characteristic of the heart rate, each RRi sample being a function of the time interval between two successive heartbeats. Each RR sample may be proportional, and in particular equal, to the time interval between two successive heartbeats, or inversely proportional to the time interval between two successive heartbeats.
In practice, disturbances in the cardiac signal, especially in an ECG signal, induce, in the RR series issued from this cardiac signal, abrupt changes of short duration, commonly called artifacts.
Disturbances, causing artifacts in the RR series, may be physiological and intrinsically linked to a temporary malfunction of the cardiac system; it may be, for example, an extrasystole. These disturbances may also be external and not related to the functioning of the cardiac system; it may be, for example, due to a patient's movement, briefly altering the measurement signal.
Artifacts in an RR series may result in a single incorrect sample or in a plurality of successive incorrect samples. In practice, an artifact in the RR series can be likened to a Dirac pulse, and is reflected, in the frequency domain, by a rectangular continuous broadband spectrum. Therefore, assuming that a series RR could be transposed in the frequency domain (by Fourier transform or other), without first taking special precautions, the presence of artifacts in the RR series would result in the frequency domain by obtaining a very disturbed frequency spectrum of the RR series, of rectangular broadband shape, masking the spectrum of the real signal.
For this reason, to obtain accurate frequency information, it is essential to eliminate the artifacts before performing the frequency transposition.
It was thus proposed, in international patent application WO 02/069178, as well as in the article from Logier R, De Jonckheere J, Dassonneville A., “An efficient algorithm for R-R intervals series filtering”. Conf Proc IEEE Eng Med Biol Soc. 2004; 6:3937-40, digital filtering algorithms, which generally allow filtering in real time a series RR obtained from a cardiac signal, by automatically detecting in the RR series the presence of one or more successive incorrect RRi samples, and by automatically replacing in the RR series the incorrect RRi samples that were detected by corrected RRc samples. Detecting incorrect RRi samples may be performed in various ways and the corrected (RRc) samples may also be calculated in various ways, and for example, and for example, but not exclusively, by linear interpolation.
A problem of these filtering algorithms, designated subsequently algorithms or filtering method “with reconstruction of incorrect samples of an RR series”, lies in the fact that the reconstruction of the RR series by replacing incorrect RRi samples which were detected by corrected RRc samples, can result in a final RR series partly rebuilt which is itself partially or completely distorted, especially when the cardiac signal that was taken is of poor quality. The lack of quality of this cardiac signal may be the result of many factors, such as, for example, and in a non-limiting, non-exhaustive manner, poor positioning of the electrodes or sensors of the heart signal, insufficient signal amplification in the signal processing chain, etc. . . .
But the reconstruction of a distorted RR series has not so far been detected by the filtering algorithms in a series RR. It follows that the information provided by these filtering algorithms can be completely wrong or insignificant without anyone noticing.
The present invention aims at providing a filtering solution of an RR series obtained from a cardiac signal, which implements an automatic reconstruction of incorrect samples of the RR series, but which can automatically control the quality of the RR series partly reconstructed.
The first purpose of the invention is thus a filtering method of an initial RR series consisting of a plurality of samples (RRi) which are respectively a function of time intervals (δti) which separate two successive heartbeats, filtering method in which one automatically detects, in the original RR series, if one or more successive samples (RRi) are incorrect, and one automatically corrects, in the RR series RR, the (RRi) sample(s) detected as being incorrect by replacing them with one or more reconstructed samples (RRc), in order to obtain a partly reconstructed series RR, optionally, and in which new samples of the RR series are optionally collected so as to obtain a RR series, if necessary partly reconstructed and re-sampled. Characteristically, in accordance with the invention, the quality of the RR series is automatically controlled by counting in a predefined sliding window, the number (NbPertub) of (RRc) samples of the RR series which were reconstructed and/or, optionally, the number (NbPertub) of (RRrc) samples of the RR series which were reconstructed and re-sampled.
In this text, and particularly in the claims, the term “cardiac signal” means any physical signal characteristic of the instantaneous heart rate (or frequency) of a living being. For the implementation of the invention, various invasive or non-invasive techniques can be used to acquire the cardiac signal. One known invasive technique consists, for example, in using a blood pressure sensor connected to a catheter inserted into an artery. Among the known non-invasive methods (and which is preferable) are, for example, one which consists in using an infrared pulse sensor, using an ultrasonic sensor for detection of the cardiac cycles, the type of sensor implemented in a cardiotocograph, or the acquisition of an electrocardiographic (ECG) signal. The acquisition of an electrocardiographic (ECG) signal is in practice the most commonly used method, because besides its noninvasive nature, it provides a more accurate cardiac signal than that obtained, for example, by means of an infrared pulse sensor.
In this text, and particularly in the claims, the term “RR series” generally means a series of various successive samples RRi obtained from a cardiac signal characteristic of the cardiac rhythm of a living being, each RRi sample being generally based on a time interval (δti) between two successive heartbeats. Generally, each sample (RRi) is proportional, in particular equal, to the time interval (δti) between two successive heartbeats. Each (RRi) sample may also be proportional, and more particularly equal to the inverse (1/δti) of the time interval between two successive heartbeats.
In the preferred exemplary embodiment described below with reference to the accompanying figures, the RR series is more particularly constructed from the R waves of an ECG signal. This is not, however, limiting the invention. In the case of an ECG type cardiac signal, one can build the series called “RR” using the other depolarization waves (P, Q, S or T) of the ECG signal to construct the RR series, the accuracy not being however as good as when using the R waves of the ECG signal. Also, when the cardiac signal is not an ECG signal, the samples of the RR series are not calculated by determining the time interval (δti) separating two successive R waves of the ECG signal, but are, more generally, determined by detecting in the cardiac signal the time interval between two successive heartbeats.
More particularly, but optionally according to the invention, the method of the invention may include additional and optional technical characteristics below, considered individually or in combination:
where N is the number of RRi samples in said window.
(RRc) of the RR series that were reconstructed and/or, optionally, the number (NbPertub) of samples (RRrc) of the RR series that were reconstructed and resampled, is greater than a preset value (THRESHOLD 1).
The invention also relates to a device for filtering a RR series consisting of a plurality of samples (RRi) which are respectively based on the time intervals (δti) separating two successive heartbeats, said device being designed to automatically filter the RR series and to control the quality of this RR series by implementing the aforementioned method.
Another purpose of the invention is a data acquisition and processing system for a cardiac signal, said system comprising electronic means for acquiring a cardiac signal, and electronic processing means designed for constructing an RR series, from the cardiac signal acquired by the electronic acquisition means, said RR series consisting of a plurality of samples (RRi) which are respectively a function of the time intervals (δti) separating two successive heartbeats of the cardiac signal. Typically according to the invention, said electronic processing means are designed to automatically filter the RR series and to control the quality of this RR series by implementing the aforementioned method.
The invention also provides a computer program comprising means for coding a computer program adapted to be executed by electronic processing means, and, when executed by electronic processing means, for implementing the filtering method of the aforementioned RR series.
Other features and advantages of the invention will appear more clearly upon reading the detailed description below of a preferred embodiment of the method of the invention, said detailed description being given by way of non-limiting and non-exhaustive example, with reference to the accompanying drawings in which:
This system comprises:
The processing means 3 of the ECG signal comprises an analog/digital converter 30, and an electronic processing unit 31. The input of converter 30 is connected to the output of the ECG monitor 2, and the output of the converter 30 is connected to an input port of the electronic processing unit 31. In one particular non-limiting embodiment of the invention, the processing unit 31 is constituted by a microcomputer, the converter 30 being connected to a serial port RS232 of this microcomputer. The invention is not limited to the implementation of a microcomputer as the electronic processing unit 31 can be implemented differently, for example as an FPGA type programmable electronic circuit, or as an integrated ASIC type circuit.
In operation, the electrodes 1 are applied to the body of the living being, and the ECG monitor 2 outputs in the usual way an analog electrical signal, called ECG signal, that has the shape of the signal shown in
Referring to
This analog ECG signal is digitized by the converter 4 with a predetermined sampling frequency (fc), equal for example to 256 Hz.
The sampled signal output from the converter 30 (signal shown in
A preferred variant of this filtering software will now be detailed.
In a particular variant embodiment of the invention, the main successive steps of the filtering algorithm are the following:
In practice, the system can be programmed to be used in real time or delayed time.
When the system is used in delayed time, step 1 is performed first in real time so as to build all RRi samples over all the period of analysis desired; all of these successive RRi samples are stored in memory, for example in a memory acquisition file of the processing unit 31. Secondly, the steps 2-6 are performed in a loop, offline, on the RRi samples stored in the acquisition file.
When the system operates in real time, step 1 of construction of the RRi samples on the one hand, and the other processing steps 2-6 on the other hand, are performed by two separate software modules operating in parallel, the first construction module (step 1) supplying the second processing and calculation module (steps 2-6) for example through a buffer file or register or equivalent.
Steps 1-5 will now be detailed.
The acquisition and construction of the RRi samples are performed by a first software sub-module which is input with the successive digital data constituting the digitized ECG signal (signal of
The first acquisition sub-module of RRi samples is designed to automatically detect each successive Ri peak in the digital signal delivered by the converter 30, and to automatically construct an RR series (
In the usual manner, the R wave usually being the finest and most extensive part of the QRS, it is preferably used to detect heart beat with very good accuracy, the time interval δti corresponding in practice to the time between two successive heartbeats. However, in another variant, one might consider using other waves (such as Q wave or S wave) of a heart beat of the ECG signal to detect and construct the RR series. In another variant, one could also consider using other cardiac signals such as the plethysmograph waveform or the invasive blood pressure.
Step 2: Filtering the RR Series with Optional Automatic Detection of Incorrect RRi Samples and Replacement by RRc Reconstructed Samples
This filtering step consists generally in automatically detecting in the RR series the presence of one or more incorrect successive RRi samples, and automatically replacing in the RR series the incorrect RRi samples that were detected by reconstructed RRc samples. The number of reconstructed RRc samples is, most of the time, different from the number of incorrect samples that were detected.
This filtering step with automatic reconstruction of incorrect RRi samples is known per se, and examples of implementation of this filtering step are described for example in international patent application WO 02/069178, as well as in the article Logier R, De Jonckheere J, Dassonneville A. , <<An efficient algorithm for R-R intervals series filtering>>. Conf Proc IEEE Eng Med Biol Soc. 2004; 6:3937-40.
It should however be noted that in the context of the invention, the detection of incorrect RRi samples is not limited to the detection methods described in the two aforementioned publications, and reconstructed RRc samples can also be calculated in various ways, such as, for example but not exclusively, by linear interpolation, as described in the two abovementioned publications.
Each reconstructed RRc sample of the RR series is identified, for example by an associated flag type identification variable. Thus, after this step, the RR series consists of RRi samples some of which are, optionally, identified by their identification variable as reconstructed RRc samples.
The filtered RR series (
During this resampling, each reconstructed RRc sample is replaced, as appropriate, by one or more reconstructed and resampled RRrc samples.
Each reconstructed and resampled RRrc sample of the RR series is identified, for example by an associated flag type identification variable. Thus, after this step, the RR series consists of RRi samples some of which are, optionally, identified by their identification variable as reconstructed and resampled RRrc samples.
Step 4: Selection of RRi Samples (of the RR Series, Optionally Partly Reconstructed and Resampled) Included in a Main Time Window of n Seconds (n>1/f)
This step consists in isolating a number N of successive RRi samples (N=n.f). As an indication, for example, a main window of 64 seconds (n=64) is chosen, which corresponds to 512 successive RRi samples (N=512) at a resampling frequency f of 8 hz.
The following steps are applied to the samples included in this main window.
This step is performed using a software sub-module that automatically calculate a NivQual quality index significant of the quality of the RR series.
In the particular embodiment described in detail below, this NivQual quality index has four quality levels from 0 to 3; the higher the index, the more reliable the RR series from step 1 is.
More particularly, the NivQual quality index is based on three variables (FCi; NORME; NbPertub) which are calculated in Step 5:
The heart rate is defined by FCi=60000/RRi, where RRi is the instantaneous value of the RRi sample in millisecond.
Calculating the mathematical norm value of the RR series resampled at the frequency f in the window of n seconds consists initially in calculating the average value M of RRi in the window.
where RRi represents the value of each RR interval and N the number of samples in the window.
This average value is then subtracted at each RRi interval of the window.
RR
i=(RRi−M),
The RRi values obtained are used for the calculation of the norm value (NORMS), or:
When taking into account the number (NbPertub) of the reconstructed and resampled RRrc samples contained in the time window of n seconds, it is considered that if the filter (Step 2) replaced too large a share of incorrect RRi samples by reconstructed RRc samples in the window of n seconds, the RR signal is, in fact, impossible to interpret.
Thus, in a first variant, the number (NbPertub) of reconstructed and resampled RRrc samples contained in the time window of n seconds is automatically counted, and this number (NbPertub) is used in Step 5 to calculate the NivQual quality index.
In a second variant, the number (NbPertub) of reconstructed RRc samples corresponding to reconstructed and resampled RRrc samples contained in the time window of n seconds is automatically counted, and this number (NbPertub) is used in Step 5 to calculate the NivQual quality index.
The aforementioned second variant may be implemented with or without resampling the RR series. In this case, the calculation of NbPertub number can be performed by automatically counting, in the RR series obtained from the filtering Step 2, the number of RRc samples of the RR series that were reconstructed, in a sliding window comprising a predefined number (N) of samples and equivalent to a time window. In this case, the aforementioned Step 6 consists in offsetting the calculation window of a predefined number p of samples (preferably p≦N), and repeating the calculation from Step 2. This offset corresponds to the sliding of the sample selection window.
The first and second variants above may also be combined.
An example of algorithm for calculating the NivQual quality index from the three aforementioned variables (FCi; NORME; NbPertub) is given below:
The values of the FCMax, FCmin, NormMax, NormMin parameters are predefined constants, which depend, for example, on the age of the human being or depend, for example, on the animal species in the context of a veterinary application. The values of the FCMax, FCmin thresholds are those commonly used by all heart monitoring devices. The values of NormMax, NormMin thresholds of the norm value are, for example, experimentally determined on 200 individuals in each category.
By way of non-limiting example:
The values of the THRESHOLD1, THRESHOLD2, THRESHOLD3 parameters are predefined constants, which depend on the number N (N=n.f) of RRi samples in the window of n seconds.
For example, the value of THRESHOLD1 can be set to one quarter of the number N (N=n.f) of RRi samples in the n seconds window, or THRESHOLD1=N/4. The value of THRESHOLD2 may be set to an eighth of the number N (N=n.f) of RRi samples in the n seconds window, or THRESHOLD2=N/8. The value of THRESHOLD3 may be set to one sixteenth of the number N (N=n.f) of RRi samples in the n seconds window, or THRESHOLD3=N/16.
The NivQual quality index calculated at each Step 5 may, for example, be displayed, especially in real time, so as to inform a practitioner of the quality level of the measured RR signal.
In the case of a NivQual quality index equal to 0, the RR series from Step 1 is considered as being of very poor quality and in fact unusable. This lack of quality of the RR series may result from many factors, such as, for example, and in a non-limiting and non-exhaustive manner, improper positioning of the electrodes 1 or the sensors for measuring the heart signal, insufficient signal amplification in the signal processing chain, etc.
When calculating a NivQual quality index equal to 0, processing unit 31 can be programmed to automatically trigger several actions, including and not limited to, triggering of a visual and/or audible alarm, and/or resetting acquisition Step 1 of RRi samples, including, in particular, a manual or automatic gain change of the source signal (ECG).
In the context of the invention, for the implementation of Step 5, the NivQual quality index calculation algorithm can be simplified by taking into account only the number NbPertub mentioned above, and by not taking into account the two other FCi and NORME parameters, or by taking into account the number NbPertub mentioned above and only one of the two other parameters, Fci or NORME.
When the NivQual quality index does not take into account the NORME parameter, re-sampling Step 3 is not necessary and may be omitted.
| Number | Date | Country | Kind |
|---|---|---|---|
| 1451487 | Feb 2014 | FR | national |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/FR2015/050417 | 2/20/2015 | WO | 00 |
