Device for processing cardiac signals

Abstract

A device for processing cardiac signals includes a memory (4) for receiving an input data set comprising a plurality of P-wave segments associated with an electrocardiogram track and an acquisition time window, and a plurality of coronary sinus signals associated with the same acquisition time window and having one or more activation sequences, and an extractor (6) configured to determine, for a given input data set, for at least some P-wave segments of the given input data set, a wave polarity profile type, at least one extremum feature of a type selected from a group of types comprising a number of positive local extrema, a number of negative local extrema, a positive prominence maximum, and a negative prominence maximum, and at least one integral value of these P-wave segments, and to associate the resulting data with a group of track P-wave data as a function of the electrocardiogram track with which each P-wave segment for which the resulting data was calculated is associated. The extractor (6) is also configured to determine, for each group of track P-wave data, a set of track P-wave features comprising the polarity profile type, a dataset extremum feature value of each calculated extremum feature type, and an integral value determined based on the data of the corresponding group of track P-wave data. The extractor is further configured to determine activation times for at least some of the activation sequences of the coronary sinus signal of a given input dataset, to estimate therefrom a set of time values, and to return a set of dataset features comprising the set of time values ​​on the one hand and the set of track P-wave features on the other hand. The device also comprises a machine learning-based locator that uses a configured decision tree to receive the set of dataset features as input and to return a cardiac identifier as output.

Description

[Technical Field]

[0001] The present invention relates to the field of cardiac signal processing, and more particularly to the field of atrial fibrillation treatment. [Background technology]

[0002] Generally, treatment of atrial tachycardia (or "AT") consists in ablating the region of the heart at the source of the tachycardia. To do this, this region must first be detected.

[0003] Most research papers outlining the problem of AT localization are primarily based on the analysis of the "P wave," a specific portion of the electrocardiogram (hereafter referred to as "ECG") that corresponds to the depolarization of the atrial appendage.

[0004] In 1995, in their paper "Use of p wave configuration during atrial tachycardia to predict site of origin," Journal of the American College of Cardiology, 26(5):1315-1324, Tang et al. focused on analyzing the polarity of P waves in surface electrodes to determine which atrial appendage (right or left) was at the origin of the tachycardia. Leads aVL and V1 proved to be the most useful for distinguishing between right and left sites; a positive P wave in lead aVL predicted right site with 88% sensitivity and 79% specificity. The sensitivity and specificity of a positive P wave in lead V1 to predict left site were 93% and 88%, respectively.

[0005] Since then, several research efforts have extended the analysis of P-wave polarity to predict site location with greater accuracy. In 2006, in the paper "P-Wave Morphology in Focal Atrial Tachycardia: Development of an Algorithm to Predict the Anatomic Site of Origin," 48:1010-1017, Kistler et al. studied 130 focal Atrial Tachycardia (AT)s and constructed a decision tree to find their origin among 11 possible regions. These regions divide the nodes according to the polarity of the P-waves in different ECG leads. The algorithm successfully classified the origin correctly in 93% of 30 new TAs.

[0006] However, the criteria maintained in these papers prevent their use in a real-time context. In fact, they are based on an inductive process that allows P-waves to be detected in a very accurate manner, which is not possible in real time. Based on P-waves determined in real time, the results would be more inconclusive and such methods would be inappropriate. Summary of the Invention

[0007] The present invention improves this situation. To this end, the present invention provides a device for processing cardiac signals, comprising: a memory configured to receive input data sets each comprising a plurality of P-wave segments, each associated with an electrocardiogram track and an acquisition time window, and a plurality of coronary sinus signals associated with the same acquisition time window and having one or more activation orders; and an extractor configured to determine, for a given input data set, for at least some P-wave segments of the given input data set, a wave polarity profile type, at least one extremum feature of a type selected from a group of types comprising a number of positive local extrema, a number of negative local extrema, a positive prominence maximum, and a negative prominence maximum, and at least one integral value of these P-wave segments, and to associate the resulting data with a group of track P-wave data as a function of the electrocardiogram track with which each P-wave segment for which the resulting data was calculated is associated. The extractor is further configured to determine, for each group of tracked P-wave data, a set of tracked P-wave features comprising a polar profile type, a dataset extremum feature value for each calculated extremum feature type, and an integral value determined based on the data of the corresponding group of tracked P-wave data. The extractor is further configured to determine activation times for at least some of the activation sequences of the coronary sinus signal of a given input dataset, to estimate a set of time values ​​therefrom, and to return a set of dataset features comprising the set of time values ​​on the one hand and the set of tracked P-wave features on the other hand. The device also comprises a machine learning-based locator using a decision tree configured to receive the set of dataset features as input and to return a cardiac region identifier as output.

[0008] This device is particularly advantageous because it allows for the real-time determination of possible areas of abnormality at the source of AT persistence. This means that the physician can focus on the reduced portion of the heart and determine the actual area at the source of AT. Thus, the determination of the likely source area of ​​AT allows the physician to save valuable time, although the latter has no established medical significance or diagnostic value.

[0009] According to various embodiments, the invention may have one or more of the following features. The locator is configured to implement a random forest classifier; the extractor is configured to determine dataset extreme feature values ​​indicating a lack of decision for the extremal feature types that have not been calculated, and to return a set of dataset features comprising the dataset extreme feature values ​​for each extremal feature type; The extractor is configured to determine a dataset extreme feature value for each type of the set of types; the extractor is configured to determine, for a given set of track P-wave features, a polarity profile type while retaining a dominant wave polarity profile type within a corresponding group of track P-wave data; the extractor is configured to determine, for a given set of track P-wave features, a dataset extremum feature value for each calculated extremum feature type and an integral value based on an average of these values ​​within a corresponding group of track P-wave data; the extractor is configured to calculate a set of time values ​​based on differences between activation times of activation sequences of the coronary sinus signal; and The locator is configured to receive as input a set of dataset features comprising nine sets of track P-wave features and a set of time values ​​comprising four values.

[0010] The invention also relates to a method for processing a cardiac signal, comprising the following operations. a) receiving an input data set each comprising a plurality of P-wave segments each associated with an electrocardiogram track and an acquisition time window, and a plurality of coronary sinus signals associated with the same acquisition time window and having one or more activation sequences; b) for a given input data set, for at least some of the P-wave segments of the given input data set: b1) determining a wave polarity profile type; b2) determining at least one extremum feature of a type selected from a group of types comprising a number of positive local extrema, a number of negative local extrema, a positive prominence maximum, and a negative prominence maximum; b3) determining the integral of at least one of these wave segments P; b4) associating the data resulting from operations b1) to b3) with a set of track P-wave data as a function of the electrocardiogram track with which each P-wave segment from which this resulting data was calculated is associated; b5) determining, for each group of track P-wave data, a set of track P-wave features comprising a polar profile type, a dataset extremum feature value for each calculated extremum feature type, and an integral value determined based on the data of the corresponding group of track P-wave data; b6) determining activation times for at least some of the activation sequences of the coronary sinus signal of the given input data set and estimating a set of time values ​​therefrom; b7) returning a set of dataset features comprising the set of time values ​​of operation b6) on the one hand and the set of track P-wave features of operation b5) on the other hand; and c) providing the set of dataset features obtained in step b7) as input for a machine learning based locator using a decision tree and returning a heart region identifier as output.

[0011] According to various embodiments, the method may have one or more of the following features. Operation b2) comprises determining a dataset extremum feature value for each type of the set of types; Operation b5) comprises determining, for a given set of track P-wave features, a polarity profile type while retaining a dominant wave polarity profile type within a corresponding group of track P-wave data; Operation b5) comprises determining, for a given set of track P-wave features, a dataset extremum feature value for each calculated extremum feature type and an integral value based on the average of these values ​​within the corresponding group of track P-wave data; Operation b6) comprises calculating a set of time values ​​based on differences between activation times of activation sequences of coronary sinus signals.

[0012] The invention also relates to a computer program comprising instructions for carrying out the method according to the invention, a data storage medium having such a computer program recorded thereon, and a computer system comprising a processor coupled to a memory having such a computer program recorded thereon.

[0013] Other characteristics and advantages of the invention will become more apparent from reading the following description with reference to examples given for illustrative and non-limiting purposes and with reference to the drawings, in which: [Brief explanation of the drawings]

[0014] [Figure 1] 1 shows a schematic diagram of a device according to the present invention. [Figure 2] An example of a function implemented by the device of FIG. 1 is shown below. [Figure 3] An example of a P-wave polarity profile searched for in the function of FIG. 2 is shown. [Figure 4] 1 shows an example of a P-wave segment and the type of features measured. [Figure 5] 1 shows an example of a coronary sinus segment and time values ​​estimated therefrom. [Figure 6] The division of the atrial appendage into 21 regions is shown. DETAILED DESCRIPTION OF THE INVENTION

[0015] The drawings and the following description inherently contain elements of a certain nature, and therefore they may not only serve to better understand the invention, but may also, where appropriate, contribute to its definition.

[0016] In an attempt to classify the origins of TA, the applicant has conducted extensive research based on the division of the atrial appendage into 21 regions as shown in Figure 6. Throughout these studies, and in search of a solution that can be used in real time, it has been found appropriate to group these regions into four groups: the left atrial appendage (regions 7, 8, 9A, 9B, 10, 11A, 11B, and 12), the right atrial appendage (regions 18, 19, 20B, 20H, and 22), the septum (regions 5, 6, 14, 15, 17, and 21), and the lateral parts of the left atrial appendage (regions 1, 2, 3, 4, 13, and 16).

[0017] As will be seen below, the device of the present invention allows for the determination of groups containing the origin of AT for a set of ECG signals and an intracardiac reference catheter, typically placed in the coronary sinus vein, hereinafter referred to as "CS." This determination is particularly interesting, as it provides the physician with a starting point for searching and determining the specific origin region of AT and, therefore, the diseased part of the heart. In fact, if these groups define four larger regions than the original 21 regions, this is already sufficient to provide a significant advantage in treatment.

[0018] Furthermore, the combination of the ECG signal with a signal derived from the coronary sinus is particularly innovative, as it has not been previously disclosed and allows for the use of machine learning, which allows the device 2 to operate in real time.

[0019] 1 shows a schematic diagram of a cardiac signal processing device 2 according to the present invention. As shown in this figure, the device 2 comprises a memory 4, an extractor 6 and a locator 8. The extractor 6 determines a set of dataset features 7 that serve as input to the locator 8, which returns source regions 9, as described below.

[0020] The memory 4 receives input data sets as input. Each input data set comprises a number of P-wave segments, each associated with an electrocardiogram track and an acquisition time window, and a number of segments estimated from the coronary sinus signal associated with the same acquisition time window. Specifically, this means splitting the ECG into nine tracks (estimated from 12 sensor tracks, three of which are ignored because they correspond to a linear combination of some of the other tracks) and five CS signal tracks, typically using a time window 10 seconds before the current measurement time. In the specific case of an ECG, only segments corresponding to P-waves are used. Because each heart rate is unique, this means that the input data set receives a variable number of P-wave segments, each estimated from an ECG track. The present invention does not cover a specific method for obtaining P-wave segments, but assumes that they form the input. In some embodiments, the device 2 can be configured to directly analyze ECG tracks and determine P-wave segments therein. By P-wave segments, it is understood that a subset of each ECG track is split within a time window to define the latter segments.

[0021] As shown below, all of the P-wave segments of each data set are processed to estimate features therefrom, which serve as input to Locator 8. Thus, for each ECG track, six features are determined, while four features are estimated from the CS signal. This therefore results in a total of 58 features (six features for nine tracks, plus four features) forming the input vector for Locator 8. Alternatively, Device 2 may determine that some P-wave segments or CS signals are difficult to utilize and discard some of them rather than estimating their respective features. Further alternatively, machine learning may be performed on a smaller number of features, for example, by determining only three to five features per ECG track. Finally, Device 2 may choose to retain only some ECG tracks (and therefore the corresponding P-wave segments), thus reducing the number of features.

[0022] The memory 4 may consist of any data storage type capable of receiving digital data, such as a hard drive, a solid state drive, any form of flash memory, random access memory, a magnetic disk, or distributed storage, either local or in the cloud.

[0023] In the example described herein, memory 4 receives all data related to device 2, i.e., the programs and software instantiating extractor 6 and locator 8, their parameters and hyperparameters, tree weights, datasets received as input (if appropriate), features determined by extractor 6, data stored in buffer memory, and source region data at output. Data computed by the device may be stored in any type of memory similar to memory 4 or in memory 4. These data may be erased or retained after the device has performed its task.

[0024] The extractor 6 and the locator 8 have direct or indirect access to the memory 4. They may be made in the form of suitable computer code executed on one or more processors. A processor is understood to be any processor suitable for the calculations described below. Such a processor may be made in any known manner in the form of a microprocessor for a personal computer, laptop, tablet, or smartphone, a dedicated chip of FPGA or SoC type, a computing resource on a grid or in the cloud, a cluster of graphical processors (GPUs), a microcontroller, or any other form capable of providing the computing power necessary to complete the processes described below. One or more of these elements may also be made in the form of specialized electronic circuits such as ASICs. A combination of a processor and an electronic circuit may also be considered. A processor dedicated to machine learning may also be used.

[0025] FIG. 2 shows an example of an implementation of the functions performed by the extractor 6 to extract a set of dataset features 7 and to infer a source region 9 therefrom.

[0026] In operation 200, the extractor 6 implements a function Init() that receives a current time point and selects in the memory 4 a P-wave segment corresponding to a time window defined by the current time point. Thus, as explained above, the time window has, as its extremes, the current time point on one hand and a time point located 10 seconds upstream of the current time point on the other hand. The applicant has noted that a 10-second duration allows for obtaining a sufficient number of P-wave segments, and that a current time point obtained every 10 seconds provides a reliable source region. Alternatively, this duration may be different, e.g., a time window duration of 2, 5, or 15 seconds, as long as at least one heartbeat is detected to obtain a P-wave.

[0027] Thus, the function Init() extracts the P-wave segments and CS signals corresponding to the unique time windows. Four functions are then run in parallel to determine the data that allows for generating a set of dataset features 7.

[0028] In operation 210, the function PSegPT() is executed to determine, for each P-wave segment, its corresponding P-wave polarity profile. Indeed, the applicant has determined that the P-wave polarity profile is a significant indicator of the source region of the TA. To this end, the function PSegPT() implements a dynamic time warping calculation ("Dynamic Time Warping" or "DTW") to determine the similarity between each P-wave segment and each of four P-wave polarity profiles that the applicant has identified as discriminatory for determining the source region of the TA. Figure 3 shows an example of each of these four profiles.

[0029] Recall that dynamic time warping is an algorithm that allows for measuring the similarity between two time-varying sequences of values. The sequences of values ​​are transformed by a nonlinear transformation of the time variable to determine a measure of their similarity independently of any nonlinear transformation of time. Thus, by comparing the DTW measurements between a given P-wave segment and four P-wave polarity profiles, it is possible to retain the P-wave polarity profile for the given P-wave segment with the lowest DTW between the two. This ultimately leads to this P-wave polarity profile being considered the "most similar" profile to the given P-wave segment. Alternatively, if no P-wave polarity profile is deemed most relevant, an "Undetermined" type may be retained to improve machine learning. Alternatively, the DTW may be replaced by Euclidean distance or another distance in mathematical terms that is minimized between the P-wave and the four profiles. Other measures of correlation with the profiles may also be used.

[0030] In operation 220, the function PSegEV() is executed. This function is intended to extract, for each P-wave segment, values ​​that characterize its extrema (in English, "peaks"). Thus, as shown in FIG. 4, the relevant extrema that can be retained are the number of positive extrema, the number of negative extrema, the maximum positive prominence, and the maximum negative prominence. These four features make it possible to define the image of each P-wave segment, i.e., its number of positive and negative peaks (at what point the P-wave segment is chaotic), as well as the magnitude of these peaks (in geography, prominence is the difference in altitude between a given peak and the highest saddle or pass that allows reaching a higher peak; this concept is naturally extended in signal graphical analysis). Depending on the situation, only one or two types of features may be calculated, i.e., for example, only the number of positive peaks, as explained above, or complementary to the maximum positive prominence. Also, some P-wave segments may be discarded if their shape is not suitable for these calculations.

[0031] In operation 230, the function PSegInt() is executed. This function is intended to calculate, for each P-wave segment, the integral of the corresponding signal. In fact, this integral also characterizes the measured power of the P-wave. Here again, some P-wave segments may be discarded if this proves to be relevant. Alternatively, two integral values ​​can be calculated, namely the integral of the positive part and the integral of the negative part, which together make it possible to describe the signal even better. In this case, the set of data set features contains rather 67 entries (58 values, plus 9 further integral values).

[0032] Finally, in operation 240, the function SCSeg() is executed. This function is intended to determine the activation time points within each coronary sinus track and return a time value indicating the offset between activation onset times. Indeed, the order of peaks in the CS signal can indicate whether the AT origin region is in the left or right atrial appendage. Applicant has discovered that these features combine particularly well with features estimated from P-wave segments in the context of machine learning. FIG. 5 shows an example of CS signals collected according to the location of the AT on the left atrial appendage, indicated by triangles with indexes 1, 2, or 3, as well as the corresponding activation order (arrows indicate each collection time). In the example described here, all time values ​​are considered relative to the activation time point of the first coronary sinus signal. Alternatively, these values ​​may be included between two consecutive signal activation times (e.g., the difference between the first and second signals, the second and third signals, the third and fourth signals, and the fourth and fifth signals). What is important is that the measured time values ​​represent the order of the peaks in the CS signal.

[0033] Although operations 210, 220, and 230 are described separately, they are performed by three separate functions, which may be performed sequentially or in parallel within a single function. Similarly, operation 240 may be performed sequentially rather than in parallel with operations 210-230. However, parallelizing these operations has the advantage of being faster and therefore facilitating real-time execution.

[0034] However, problems arise when performing calculations by machine learning. As mentioned above, since it is the measurement time point (current time) that determines the time window, the number of P-wave segments is not constant during the execution of device 2. Furthermore, there is even a risk of over-representation of P-wave segments associated with certain ECG tracks compared to others. Finally, if the number of P-wave segments is not constant, it is not possible to define a fixed-dimensional input vector for machine learning.

[0035] Thus, the function of FIG. 2 continues with operation 250, in which the function FeatGp() is executed by the extractor 6. First, this function groups together all of the data calculated by operations 210-230 according to the ECG track, from which the P-wave segment at the origin of these values ​​is estimated. This means that all of the P-wave polarity profiles, extrema, and integral values ​​associated with the same ECG track are grouped together into a group of track P-wave data. Then, in each group of track P-wave data, a unique value is retained for each feature type, i.e., P-wave polarity profile, extrema of each type calculated, and integral value. As mentioned above, preferably, there are four extrema, corresponding to the number of peaks (positive and negative) and the maximum prominence (positive and negative), respectively. In the case of P-wave polarity profiles, a majority vote can be performed. That is, for a given ECG track, it is the maximum number of P-wave polarity profiles of the P-wave segments that determines the retained profile. In the case of extrema and integral values, average values ​​can be used. Other methods may be used: weighted average, median, regression, and minimum threshold of majority vote.

[0036] Thus, six features are obtained for each ECG track, which form a set of track P-wave features. By grouping these sets with the set of time values ​​estimated from the CS signal, a dataset feature set is obtained comprising the aforementioned 58 features. If the CS signal contains several activation sequences, the set of values ​​may be averaged in a similar manner to the extrema of the P-wave segments in operation 250.

[0037] This set may then be sent as an input vector to a locator 8 which executes a function RF() in operation 260. The function RF() implements a random forest classifier trained on a set of training data where the output source regions are known. Thus, thanks to the training of the machine learning engine, the source regions 9 can be determined in real time.

[0038] Alternatively, the locator 8 may implement machine learning using decision trees different from random forests, for example based on gradient boosting (in English "gradient boosting") or extreme gradient boosting (in English "XGBoost").

[0039] The applicant's tests have been able to prove that the device 2 makes it possible to obtain very satisfactory results even with a fairly limited training base, in which the ECG and CS signals are associated with AT origin regions according to the following distribution: left atrial appendage (47), right atrial appendage (7), septum (158), lateral (24). The results obtained with this training base are therefore summarized in the table below.

[0040] [Table 1]

[0041] The Cohen Kappa score measures a very favorable 0.55, with an overall prediction rate of 74%. Recall that the Cohen Kappa score measures the agreement between the annotations of the test set and the predictions made on it, taking into account the imbalance between the representation of classes in the latter.

[0042] Device 2 thus allows a "high level" determination of the source region of AT to be performed in real time, which allows significantly accelerating the ablation procedure in the context of AT treatment, while reducing the risk of errors associated with searching "useless" regions of the heart, i.e., regions unlikely to be the source of AT.

Claims

1. 1. A device for processing cardiac signals, comprising: a memory (4) configured to receive an input data set each comprising a plurality of P-wave segments each associated with an electrocardiogram track and an acquisition time window, and a plurality of coronary sinus signals associated with the same acquisition time window and having one or more activation sequences; For a given input data set, for at least some P-wave segments of said given input data set: Wave polar profile type; at least one extremum feature of a type selected from a group of types comprising the number of positive local extrema, the number of negative local extrema, the positive prominence maximum, and the negative prominence maximum; the integral of at least one of these P-wave segments; and and associating the result data with the group of track P-wave data as a function of the electrocardiogram track with which each P-wave segment for which the result data is calculated is associated. An extractor (6) configured as follows: The extractor (6) is further configured to determine, for each group of track P-wave data, a set of track P-wave features comprising a polar profile type, a dataset extremum feature value for each calculated extremum feature type, and an integral value determined based on the data for the corresponding group of track P-wave data; the extractor (6) is further configured to determine activation times in at least some of the activation sequences of the coronary sinus signal of the given input dataset, to deduce therefrom a set of time values, and to return, on the one hand, the set of time values ​​and, on the other hand, a set of dataset features comprising the set of track P-wave features; a machine learning based locator (8) using a decision tree configured to receive as input the set of dataset features and to return as output a heart area identifier; A device comprising:

2. The device of claim 1 , wherein the locator (8) is configured to implement a random forest classifier.

3. 3. The device according to claim 1 or 2, wherein the extractor (6) is configured to determine dataset extremum feature values ​​indicative of a lack of decision for extremum feature types that have not been calculated, and to return a set of dataset features comprising a dataset extremum feature value for each extremum feature type.

4. 3. The device according to claim 1 or 2, wherein the extractor (6) is configured to determine a dataset extremum feature value for each type of the set of types.

5. 5. The device of claim 1, wherein the extractor is configured to determine, for a given set of track P-wave features, the polarity profile type while retaining the dominant wave polarity profile type within a corresponding group of track P-wave data.

6. 6. The device of claim 1, wherein the extractor is configured to determine, for a given set of track P-wave features, the dataset extremum feature value of each calculated extremum feature type and the integral value based on the average of these values ​​within a corresponding group of the track P-wave data.

7. 7. The device according to claim 1, wherein the extractor (6) is configured to calculate the set of time values ​​based on the differences between the activation times of the activation sequences of the coronary sinus signals.

8. 8. The device of claim 1, wherein the locator is configured to receive as input a set of dataset features comprising nine sets of track P-wave features and a set of time values ​​comprising four values.

9. The following works: a) receiving an input data set each comprising a plurality of P-wave segments each associated with an electrocardiogram track and an acquisition time window, and a plurality of coronary sinus signals associated with the same acquisition time window and having one or more activation sequences; b) for a given input data set, for at least some of the P-wave segments of the given input data set: b1) determining a wave polar profile type (210); b2) determining (220) at least one extremum feature of a type selected from a group of types comprising the number of positive local extrema, the number of negative local extrema, the positive prominence maximum, and the negative prominence maximum; and b3) determining the integral of at least one of these wave segments P (230); b4) associating (250) the data resulting from operations b1) to b3) with a group of track P-wave data as a function of the electrocardiogram track with which each P-wave segment for which the resulting data was calculated is associated; b5) determining for each group of track P-wave data a set of track P-wave features comprising a polar profile type, a dataset extremum feature value for each calculated extremum feature type, and an integral value determined based on the data for the corresponding group of track P-wave data (250); b6) determining (240) activation times for at least some of the activation sequences of the coronary sinus signal of the given input data set and estimating a set of time values ​​therefrom; b7) returning (250) a set of dataset features comprising, on the one hand, the set of time values ​​of act b6) and, on the other hand, the set of track P-wave features of act b5); and c) providing (260) the set of dataset features obtained in step b7) as input to a machine learning based locator (8) using decision trees and returning a heart area identifier as output; 1. A method for processing a cardiac signal, comprising:

10. The method of claim 9 , wherein act b2) comprises determining a dataset extremum feature value for each type in the set of types.

11. 11. The method of claim 9 or 10, wherein operation b5) comprises, for a given set of track P-wave features, determining the polarity profile type while retaining the dominant wave polarity profile type within a corresponding group of track P-wave data.

12. 12. The method of claim 9, wherein operation b5) comprises determining, for a given set of track P-wave features, the dataset extremum feature value for each calculated extremum feature type and the integral value based on the average of these values ​​within a corresponding group of track P-wave data.

13. 13. The method of claim 9, wherein operation b6) comprises calculating the set of time values ​​based on the differences between the activation times of the activation sequences of the coronary sinus signals.

14. A computer program comprising instructions for carrying out the method of any one of claims 9 to 13 when implemented by a computer.

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

Images