Device for processing spatiotemporally distributed signals of an electrocardiogram

Abstract

A device for processing electrocardiogram spatiotemporal variance signals comprises a memory (4) for receiving, on the one hand, time markers and, on the other hand, electrocardiogram spatiotemporal variance signals associated with electrocardiogram tracks; a computer (8) for receiving as input an electrocardiogram track identifier and a time marker, for analyzing the electrocardiogram spatiotemporal variance signal associated with the electrocardiogram track identifier, for analyzing a signal excerpt comprised between the time marker and the first previous time marker for which the electrocardiogram spatiotemporal variance signal value indicates no variance, for deriving from the signal excerpt, on the one hand, a signal flatness value and, on the other hand, a duration value derived from the duration of the signal excerpt, and for feeding back a track priority value calculated from the flatness value and the duration value; and a monitor (6) for receiving the electrocardiogram spatiotemporal variance signals associated with each electrocardiogram track, for calling the computer (8) with the corresponding time marker and electrocardiogram track identifier if the value of the electrocardiogram spatiotemporal variance signal indicates relevant variance.

Description

[Technical Field]

[0001] The present invention relates to the field of analyzing signals obtained from an electrocardiogram. [Background technology]

[0002] The field of treating atrial fibrillation has made considerable progress over the past decade. To treat atrial fibrillation, practitioners operate by inserting a catheter equipped with multiple electrodes. These electrodes travel through the heart and measure the electrical signals propagating within it. The resulting signals are called electrocardiograms. These electrocardiograms are processed to help practitioners detect the areas of the heart that are causing the atrial fibrillation. Once these areas are identified, practitioners can inactivate them by burning them, thereby restoring normal heart function and suppressing atrial fibrillation.

[0003] Most existing solutions are based on CFAE ("Complex Fractionated Atrial Electrograms") analysis, the principle of which is to find the locations of the atria where the electrocardiogram loses continuity, i.e. splits.

[0004] Applicant has developed electrogram processing techniques that have sparked interest in the scientific community since the publication of highly supportive clinical studies, which have led to the publication of several papers, including the publication by Seitz et al., "AF Ablation Guided by Spatiotemporal Electrogram Dispersion Without Pulmonary Vein Isolation: A Wholly Patient-Tailored Approach," Journal of the American College of Cardiology, Vol. 69, No. 3, January 24, 2017, pp. 303-321.

[0005] This process is based on the detection of a quantity of the electrocardiogram signal, which Applicant calls variance. More specifically, Applicant has discovered that measuring the spatiotemporal variance of the electrocardiogram, based both on the evolution of the electrocardiogram signal at each electrode over time, as well as taking into account the electrocardiogram signals at adjacent electrodes, is particularly effective for determining the area of ​​the heart from which atrial fibrillation originates. This measurement of magnitude has been the subject of patents issued in many countries worldwide, including in Europe as published under EP 3,236,843.

[0006] Applicant extends that work by incorporating variance into machine learning tools to determine in near real time the presence or absence of variance in the electrocardiogram during a procedure, thereby alerting practitioners, speeding up ablation procedures, and improving the quality of the results obtained while reducing surgical risk, increasing the reliability of the procedure and limiting the number of patients who must undergo another procedure for the same cause.

[0007] This work has been incorporated into software called VX1, which has received FDA (Food and Drug Administration) clearance and CE (European Conformity) marking. The VX1 software does not provide a diagnosis, but rather assists the practitioner in making a diagnosis based on dispersion measurements presented in the software, and the practitioner can choose to mark areas of the heart to undergo catheter ablation.

[0008] In its current version, the VX1 software uses a database of annotated ECGs to train a machine learning engine that feeds back a value indicating the likelihood that the area analyzed for ECG signals is subject to atrial fibrillation. Therefore, while this is not strictly a measure of variance, nor does it constitute a medical diagnosis, it is an indication of whether the ECG is more or less likely to demonstrate the occurrence of atrial fibrillation. The value fed back by the VX1 software is therefore a value between 0 and 1 (0 indicating a near-zero probability that the area being measured is involved in fibrillation, and 1 indicating this probability is almost certain). The software is configured to use a value of 0.5 as the threshold above which the possible presence of variance is brought to the practitioner's attention.

[0009] Applicant noticed during the development of the VX1 software that in some cases many areas of the heart would feed back values ​​greater than 0.5 which would complicate the practitioner's work as they needed to proceed with as few ablations as possible, and when areas causing too much variance were identified, the practitioner tended to lose confidence in the software's measurement quality as the software fed back very noisy information. Summary of the Invention

[0010] The present invention improves this situation and therefore provides a device for processing electrocardiogram spatiotemporal variance signals, the device comprising: a memory configured to receive, on the one hand, time markers and, on the other hand, electrocardiogram spatiotemporal variance signals associated with electrocardiogram tracks; a calculator configured to receive an electrocardiogram track identifier and the time markers as input, and to analyze the electrocardiogram spatiotemporal variance signal associated with this electrocardiogram track identifier, analyzing a signal excerpt comprised between the time marker and the first previous time marker for which the electrocardiogram spatiotemporal variance signal value indicates no variance, deriving from this signal excerpt, on the one hand, a signal flatness value and, on the other hand, a duration value derived from the duration of the signal excerpt, and feeding back a track priority value calculated from the flatness and duration values; and a monitor configured to receive the electrocardiogram spatiotemporal variance signals associated with each electrocardiogram track and, if the value of the electrocardiogram spatiotemporal variance signal indicates relevant variance, to call the calculator with the corresponding time marker and electrocardiogram track identifier.

[0011] This device is particularly advantageous because it can provide additional information regarding the variance value that allows the practitioner to indicate the priority of the involvement in fibrillation of the cardiac region responsible for the fed-back variance value. Thus, the track priority value is information that allows the clinician to classify cardiac regions for ablation.

[0012] Also, the track priority value tends to increase when the practitioner stops in a given area and this is the location of variance. Thus, the track priority value allows the practitioner to perform the procedure more efficiently, i.e., when variance is detected, it is known to be desirable to stop in the area concerned and wait to see if the track priority value increases. If not, it can be assured that it is desirable to continue searching.

[0013] According to various embodiments, the present invention comprises the following features: the electrocardiogram spatiotemporal variance signal is a sequence of values ​​derived from the electrocardiogram signal, each of which indicates a confidence level in the fact that variance has occurred for a possible electrocardiogram track and time marker, and the monitor and / or computer is configured to determine whether the electrocardiogram spatiotemporal variance signal indicates relevant variance by comparing the value of the electrocardiogram spatiotemporal variance signal to a threshold value; the calculator is configured to calculate the flatness value from at least one of a standard deviation of the signal extraction, a total variation of the signal extraction, an entropy of the signal extraction, or a value derived from one or more first or higher derivatives of the signal extraction; the computer is configured to calculate the duration value by comparing the duration of the signal extraction with a minimum and / or maximum value and feeding back a value of 0 if the duration of the signal extraction is less than the minimum value, feeding back a value of 1 if the duration of the signal extraction is greater than the maximum value, and feeding back a value between 0 and 1 otherwise; the computer is configured to determine a value between 0 and 1 by applying to the duration of the signal extraction a projection function of the interval between the minimum and maximum values ​​onto the interval between 0 and 1, the projection function being selected from the group comprising an affine function, an exponential function, a polynomial, and a threshold function; receiving, by a computer, a spatiotemporal distribution signal of an electrocardiogram associated with time markers on the one hand and electrocardiogram tracks on the other hand; b) determining whether the values ​​of the spatiotemporal dispersion signal of the electrocardiogram indicate dispersion; c) if operation b) is false, repeating with the variance value having the subsequent time marker; d) if operation b) is possible, analyzing the spatiotemporal dispersion signal of the electrocardiogram associated with the corresponding electrocardiogram track identifier by analyzing the signal extract included between the time marker and the first previous time marker for which the spatiotemporal dispersion signal value of the electrocardiogram indicates no dispersion, and by deriving from this signal extract on the one hand a signal flatness value and on the other hand a duration value derived from the duration of the signal extract; and e) feeding back a track priority value calculated from the flatness value and duration value of operation d); may have one or more of the following:

[0014] According to various embodiments, the method comprises the following features: the electrocardiogram spatiotemporal variance signal is a sequence of values ​​derived from the electrocardiogram signal, each of which indicates a confidence level in the fact that variance has occurred for a possible electrocardiogram track and time marker, and operation b) comprises comparing a value of the electrocardiogram spatiotemporal variance signal with a threshold value; operation d) includes calculating the flatness value from at least one of the standard deviation of the signal extraction, the total variation of the signal extraction, the entropy of the signal extraction, or a value derived from one or more first or higher derivatives of the signal extraction; operation d) includes calculating the duration value by comparing the duration of the signal extraction with a minimum value and / or a maximum value and feeding back a value of 0 if the duration of the signal extraction is less than the minimum value, feeding back a value of 1 if the duration of the signal extraction is greater than the maximum value, and feeding back a value between 0 and 1 otherwise; Operation d) comprises determining a value between 0 and 1 by applying a projection function of the interval between the minimum and maximum values ​​to the interval between 0 and 1 for the duration of the signal extraction, the projection function being selected from the group comprising an affine function, an exponential function, a polynomial, and a threshold function; and operation e) includes calculating the track priority value by performing a weighted average, a harmonic average, a weighted average, or an arithmetic average of the flatness value and the duration value; may have one or more of the following:

[0015] The invention also relates to a computer program comprising instructions for carrying out the method according to the invention, a data storage medium on which such a computer program is recorded, and a computer system comprising a processor coupled to a memory on which such a computer program is recorded. [Brief explanation of the drawings]

[0016] Further characteristics and advantages of the invention will become apparent from the following description with reference to examples given by way of illustration and not of limitation with reference to the drawings, in which:

[0017] [Figure 1] 1 shows a schematic diagram of a device according to the present invention. [Figure 2] 2 illustrates an example of an implementation of an operating loop of the device of FIG. 1.

[0018] The drawings and the following description contain elements that are essentially certain in nature and, as such, may serve not only to enhance the understanding of the invention but also, where appropriate, to contribute to its definition. DETAILED DESCRIPTION OF THE INVENTION

[0019] Figure 1 shows a schematic example of a device 2 according to the invention. As specified in the introduction, the signals used by the device are based on the electrocardiogram measured by a pair of electrodes on a catheter in the patient's heart.

[0020] However, in the particular case of the present invention, it is not these signals that are processed, but rather the dispersion measure derived from the signals. As indicated in the introduction, the article by Seitz et al., "AF Ablation Guided by Spatiotemporal Electrogram Dispersion Without Pulmonary Vein Isolation: A Wholly Patient-Tailored Approach" (Journal of the American College of Cardiology, Vol. 69, No. 3, January 24, 2017, pp. 303-321) and the patent published under number EP 3236843, allow us to better understand what dispersion is, both in terms of its similarities and differences with CFAE.

[0021] Considering the fields considered and the fact that dispersion and CFAE represent completely different phenomena insofar as CFAE ignores spatial aspects, they are not of great interest in the context of the present invention. In fact, as will be seen below, the present invention generally aims to evaluate the stability of the signal formed by dispersion values. This same analysis makes little or no sense in the case of CFAE.

[0022] The device 2 includes a memory 4 , a monitor 6 , and a computer 8 .

[0023] The memory 4 is configured to receive all data, whether input or output, of a global or local nature from the device 2. The memory 4 may consist of any data storage type capable of receiving digital data, i.e., hard drive, solid state drive, any form of flash memory, random access memory, magnetic disk, locally distributed or cloud-based storage, etc.

[0024] In the example described herein, memory 4 receives all of the data related to device 2, i.e., the programs and software instantiating monitor 6 and calculator 8, their parameters and possible hyperparameters, possible neural network weights, neural network outputs and intermediate data, (where appropriate) electrocardiogram spatiotemporal variance signal data received as input, signal flatness and signal duration values, data stored in buffer memory, and output track priority value data. Data calculated by the device may be stored in any type of memory similar to memory 4 or in the latter. These data may be erased or retained after the device has performed its task.

[0025] As will be seen below, the signal flatness value and duration value are two values ​​used to evaluate whether a variance signal is associated with an ablation priority cardiac area. These values ​​are summed to generate a track priority value that indicates whether the determined variance is associated with a variance that is likely to be associated with an area that is the source of atrial fibrillation. In addition to other patent applications filed by the applicant, the track priority value is not a diagnosis, but rather constitutes an index that allows a physician to make a decision, much like blood pressure does in other contexts.

[0026] The monitor 6 and the calculator 8 access the memory 4 directly or indirectly. They may be implemented as suitable computer code executed by one or more processors. By processor, any processor adapted for the calculations described below should be understood. Such a processor may be implemented in any known manner in the form of a personal computer, laptop, tablet or smartphone, a dedicated chip of FPGA or SoC type, a computing resource on the network or in the cloud, a cluster of graphical processing units (GPUs), a microprocessor of a microcontroller, or any other form capable of providing the computing power required for the implementation described below. One or more of these elements may also be implemented in the form of specialized electronic circuits, such as ASICs. A combination of a processor and an electronic circuit may also be envisaged. A processor specialized for machine learning may also be envisaged.

[0027] The function of the monitor 6 is to analyze the data stream of the spatiotemporal variance signal of the electrocardiogram received as input and to detect therein facts in which the variance value indicates the need for treatment. As indicated in the introduction, in the context of the VX1 software, a variance value of 0.5 or greater is significant. Naturally, the detection of this value will depend on the value taken by the variance signal received as input. For example, it could be generated inversely to the VX1 software (e.g., a value derived from 1-VX1), in which case values ​​below 0.5 would be significant instead. This determination could also be performed differently, based on a value derived from the derivative of the variance signal, or in some other related way.

[0028] Due to the continuous nature of the processing by device 2, which will become apparent below, once a variance value has been detected by monitor 6, the detection of subsequent values ​​(but of course relating to the same track) may be different or simplified. Thus, in the above case, rather than comparing the current variance value to a threshold, monitor 6 may, for example, measure the derivative of the input variance value signal and consider the detection to be positive if the derivative is positive. In general, monitor 6 may rely on several tests to evaluate the detection at the relevant variance value.

[0029] The role of the monitor 6 is therefore of an "interrupt" nature. Indeed, if there is no relevant variance value, no variance is detected and therefore it is not necessary to calculate a track priority value. On the other hand, as soon as a relevant variance value is detected, the monitor 6 calls the calculator 8 to calculate the track priority value. The operation of the device 2 can therefore appear as a loop of the detection of each track by the monitor 6, and the calculator 8 is executed each time a relevant variance value is detected. Naturally, other embodiments can be envisaged.

[0030] The role of the calculator 8 is to calculate the track priority value of tracks whose variance values ​​are deemed relevant by the monitor 6. As a reminder, the variance signal is a signal that associates a time marker with a variance value. This variance value itself is derived from the analysis of several ECG signal values. In the context of the VX1 software, the variance value is updated approximately every 300 milliseconds, based on an extraction of an ECG signal lasting approximately 1.5 seconds. Alternatively, this update could occur every 100 milliseconds, 500 milliseconds, etc.

[0031] As will be seen below, the determination of the track priority value is based on the extraction of variance values ​​that can have a continuous duration of 1.5 seconds, i.e., about 5 variance values, up to several tens of seconds, i.e., about 100 variance values.

[0032] More precisely, the computer 8 analyzes each signal extraction of variance values ​​ending with the variance value that has just been deemed relevant by the monitor 6. This extraction includes exclusively variance values ​​that are consecutive, associated with the same track, and that contain time markers that are considered relevant by the monitor 6. This extraction can be carried out in a number of ways, namely the monitor 6 is capable of generating extracts during its operation by adding a current variance value detected as relevant to the current extract if the immediately preceding variance value is also detected as relevant, or creating a new extract otherwise; the calculator 8, when it receives a variance value associated with a given time marker, can analyze a buffer of past variance values ​​and stop at the oldest value deemed relevant by the monitor 6; or the calculator 8 is able to recover a buffer of past variance values ​​from the time markers associated with the variance values ​​received as input and to cleanly determine which extracts of this buffer it considers relevant; It is thought that it can be obtained by

[0033] The applicant's research has shown that the more accurate the variance value signal, the more useful the temporal continuity of the associated variance values. Indeed, if the variance is considered "noisy," it may be tempting to ignore irrelevant variance values ​​in order to obtain more data that allows for the calculation of track priority values. The applicant's research has shown that the combination of signal flatness values ​​and duration values ​​does not artificially expand the size of the extraction and allows for better results to be obtained.

[0034] Calculator 8 operates by performing two measurements on the extractions defined above: a signal flatness measurement and a duration measurement. In both cases, the objective is to determine whether the variance values ​​have some stability over time. Indeed, applicant's research has revealed that signals with stable variance values ​​have been associated with priority zones for ablation to suppress atrial fibrillation.

[0035] In the examples described herein, the signal flatness value is derived from the standard deviation of the extracted data. Applicant's research has revealed that standard deviation is the measure that yields the best results. Nevertheless, Applicant has determined that other types of measures may be employed, such as variance, total variation of the extraction, entropy of the extraction, values ​​derived from one or more derivatives of one or more orders of the extraction, range, interquartile range, or another similar measure.

[0036] In parallel, calculator 8 also determines a duration value that allows indicating how long the current extraction is relative to the time interval that is considered to indicate that the cardiac region associated with this variance has priority for ablation. In the example described herein, calculator 8 projects the duration of the extraction onto a standard interval between a minimum duration and a maximum duration. These two values, empirically estimated by applicant, indicate, respectively, the minimum duration that an extraction must have to designate a priority cardiac region, and the maximum duration that considers that the practitioner cannot afford to dwell too long in each region if he wants to perform the action within a reasonable time frame and minimize surgical risks.

[0037] In the example described herein, the calculator 8 determines the duration of the sampling and projects that value over an interval (minimum duration; maximum duration) to determine a value between 0 and 1. The projection can be of any type: linear, polynomial, exponential, threshold, etc. This involves indicating for a given sampling duration whether this duration is characteristic of a preferred cardiac segment.

[0038] The minimum duration has obvious utility as a floor value. The maximum duration has important operational utility: when the first variance value of an extraction longer than the maximum duration is detected, the output priority value will necessarily be lower due to the lower duration value. As the extraction gets larger, the track priority value will increase along with the duration value. When the operator notices that the track priority value is no longer moving, because the duration value cannot increase any further, he can determine that it is time to move the catheter.

[0039] Alternatively, calculator 8 can determine the duration values ​​differently, independent of the range [minimum duration; maximum duration]. Further alternatively, the minimum and maximum durations can be variable during treatment or customized for each patient.

[0040] The flatness and duration values ​​of the signal may be determined in parallel with each other, or alternatively, one may be calculated before the other.

[0041] The value between 0 and 1 is chosen because of the way that calculator 8 determines the track priority value from the signal flatness and duration values. Indeed, calculator 8 operates in the above example by computing a weighted harmonic mean. Alternatively, the mean may be a harmonic, weighted or arithmetic mean of the flatness and duration values.

[0042] If the signal flatness and duration values ​​are not within the same range of values, a relative adjustment may be made or a different formula may be maintained to calculate the track priority value.

[0043] 2 shows an example of an operation loop of device 2. In operation 200, monitor 6 is called with the current variance value of a given track. When monitor 6 determines that the variance value is relevant, it triggers operation 210 in which an extraction is determined from the time marker of the variance value of operation 200, the associated track identifier, and the variance value signal already received for this track identifier. Once the extraction is determined, calculator 8 can determine flatness and duration values ​​in operation 220 and a track priority value, which can be fed back in operation 230.

[0044] If applicable, device 2 would determine a track priority value for each track for which a variance value signal is received as input. This would enrich the information transmitted to the practitioner, enabling the practitioner to make a diagnosis as to whether or not to ablate the area associated with a given track.

Claims

1. 1. A device for processing spatiotemporally distributed signals of an electrocardiogram, comprising: On the other hand, time markers and a memory (4) configured to receive, on the other hand, a spatiotemporal distribution signal of an electrocardiogram associated with an electrocardiogram track; a computer (8) configured to receive as input an electrogram track identifier and a time marker, to analyze the spatiotemporal dispersion signal of the electrocardiogram associated with said electrogram track identifier, to analyze a signal excerpt comprised between said time marker and the first previous time marker for which the electrocardiogram spatiotemporal dispersion signal value indicates no dispersion, to derive from said signal excerpt on the one hand a flatness value of said signal and on the other hand a duration value derived from the duration of said signal excerpt, and to feed back a track priority value calculated from said flatness value and said duration value; a monitor (6) configured to receive the electrocardiogram spatiotemporal variance signal associated with each electrocardiogram track, and to call the calculator (8) with the corresponding time marker and electrocardiogram track identifier when the value of the electrocardiogram spatiotemporal variance signal indicates a relevant variance; A device comprising:

2. the electrocardiogram spatiotemporal variance signal is a sequence of values ​​derived from the electrogram signal, each of which indicates a confidence level in the fact that variance has occurred for the considered electrogram track and time marker; 2. The device of claim 1, wherein the monitor (6) and / or the computer (8) are configured to determine whether an electrocardiogram spatiotemporal dispersion signal exhibits relevant dispersion by comparing a value of the electrocardiogram spatiotemporal dispersion signal with a threshold value.

3. 3. The device of claim 1 or 2, wherein the calculator (8) is configured to calculate the flatness value from at least one of the standard deviation of the signal extraction, the total variation of the signal extraction, the entropy of the signal extraction, or a value derived from one or more first or higher derivatives of the signal extraction.

4. 4. The device according to claim 1, wherein the calculator (8) is configured to calculate the duration value by comparing the duration of the signal extraction with a minimum and / or a maximum value and by feeding back the value 0 if the duration of the signal extraction is smaller than the minimum value, feeding back the value 1 if the duration of the signal extraction is larger than the maximum value, and feeding back a value between 0 and 1 otherwise.

5. the calculator (8) is configured to determine the value between 0 and 1 by applying a projection function of the interval between the minimum and maximum values ​​onto the interval between 0 and 1 for the duration of the signal extraction; The device of claim 4 , wherein the projection function is selected from the group comprising an affine function, an exponential function, a polynomial, and a threshold function.

6. 6. The device of claim 1, wherein the calculator (8) is configured to calculate the track priority value by performing a weighted average, a harmonic average, a weighted or an arithmetic average of the flatness value and the duration value.

7. 1. A method for determining track priority values ​​of spatiotemporal dispersion signals of an electrocardiogram, comprising the following operations: a) receiving a spatiotemporal dispersion signal of an electrocardiogram associated with time markers on the one hand and electrocardiogram tracks on the other hand; b) determining whether the values ​​of the spatiotemporal dispersion signal of the electrocardiogram exhibit dispersion; c) if operation b) is false, repeating with the variance value having the subsequent time marker; d) if operation b) is possible, analyzing the spatiotemporal dispersion signal of the electrocardiogram associated with the corresponding electrocardiogram track identifier by analyzing a signal extract included between the time marker and the first previous time marker for which the spatiotemporal dispersion signal value of the electrocardiogram indicates no dispersion, and by deriving from this signal extract on the one hand a flatness value of the signal and on the other hand a duration value derived from the duration of the signal extract; and e) feeding back a track priority value calculated from the flatness value and the duration value of operation d); A method comprising:

8. the electrocardiogram spatiotemporal variance signal is a sequence of values ​​derived from the electrogram signal, each of which indicates a confidence level in the fact that variance has occurred for the considered electrogram track and time marker; The method of claim 7 , wherein operation b) comprises comparing the value of the electrocardiogram spatiotemporal variance signal with a threshold value.

9. 9. The method of claim 7 or 8, wherein operation d) comprises calculating the flatness value from at least one of the standard deviation of the signal extraction, the total variation of the signal extraction, the entropy of the signal extraction, or a value derived from one or more first or higher derivatives of the signal extraction.

10. 10. The method according to claim 7, wherein operation d) comprises calculating the duration value by comparing the duration of the signal extraction with a minimum and / or a maximum value and feeding back the value 0 if the duration of the signal extraction is smaller than the minimum value, feeding back the value 1 if the duration of the signal extraction is larger than the maximum value, and feeding back a value between 0 and 1 otherwise.

11. operation d) comprises determining the value between 0 and 1 by applying a projection function of the interval between the minimum value and the maximum value onto the interval between 0 and 1 for the duration of the signal extraction; The method of claim 10 , wherein the projection function is selected from the group comprising an affine function, an exponential function, a polynomial, and a threshold function.

12. 12. The method of claim 7, wherein operation e) comprises calculating the track priority value by performing a weighted average, a harmonic average, a weighted or an arithmetic average of the flatness value and the duration value.

13. A computer program comprising instructions for carrying out the method of any one of claims 7 to 12 when executed by a computer.

14. A data storage medium having recorded thereon the computer program according to claim 13.

Images