Intracardiac signal processing device

Abstract

An apparatus, method, computer program and data storage medium for processing intracardiac signals are provided. [Solution] The intracardiac signal processing device comprises a memory means 4 configured to receive electrocardiogram data and time-aligned electrogram data, a detection means 6 configured to analyze the electrocardiogram data and detect QRS wave time points from the electrocardiogram data, an analysis means 8 configured to perform a wavelet transform of the electrogram data, an extraction means 10 configured to derive coefficients from the wavelet transform that correspond to the QRS wave time points detected by the detection means 6 and store these in a storage means 14, and a configuration means 12 configured to extract a QRS fingerprint signal from the storage means 14, subtract the QRS fingerprint signal from the QRS wave time points of the wavelet transform, and generate noise-reduced electrogram data as an output by an inverse wavelet transform of the obtained signal.

Description

[Technical field]

[0001] The present invention relates to the field of intracardiac signal processing and noise reduction. [Background technology]

[0002] Measurements of electrograms in the atria proximal to the ventricles (e.g., reference catheters used in coronary venous ablation procedures) can be disturbed by ventricular activity, which produces waveforms in the intracardiac electrograms, also known as "far-field" (FF) waveforms. This distortion by the ventricles makes the analysis of the atrial electrogram signal more difficult, and this non-atrial information can be erroneously mixed in (e.g., in cycle length estimates).

[0003] Specifically, the present invention aims to remove far-field ventricular contributions, considered as noise, from electrogram tracks obtained from intracardiac probes, even if they overlap with near-field activity, while leaving the near-field activity intact.

[0004] When such signals are detected in clinical settings (i.e., in the operating room), the far-field effects are dealt with by crudely cancelling them, which is not very effective and also results in signal degradation.

[0005] The academic literature offers some guidance: - TMS ("Template matching and subtraction") involves calculating the average value of the QRS complexes during the recording period and subtracting this average value from all QRS complexes encountered (see, for example, Rieta, JJ, et al. "Atrial activity extraction based on blind source separation as an alternative to QRST cancellation for atrial fibrillation analysis." Computers in Cardiology 2000. Vol. 27 (Cat. 00CH37163), IEEE, 2000); - the Independent Component Analysis (ICA) method seeks to find a set of components that minimizes the mutual information present in the various segments of the ECG signal. By aggregating these independent components within the subspaces of ventricular and atrial activity, it may be possible to reconstruct the atrial activity at each observation point from the subspace of atrial activity (see, for example, F. Castells, et al. “Multidimensional ICA for the Separation of Atrial and Ventricular Activities from Single Lead ECGs in Paroxysmal Atrial Fibrillation Episodes”, Lecture Notes in Computer Science, vol. 3195, p. 1229-1236, 2004); - Adaptive Ventricular Cancellation (AVC) is a method that estimates the interference by applying a finite impulse response filter to a reference channel and then removes the interference from the reference channel (see, for example, Widrow B., et al. "Adaptive noise cancelling: principles and applications", Proc. IEEE 63, 1692-716, 1975 (Non-Patent Document 3)); - The wavelet decomposition combined ICA method applies the ICA method to the wavelet decomposition of electrograms and electrocardiograms (see, for example, Simanto Saha, et al. "A Ventricular Far-field Artefact Filtering Technique for Atrial Electrograms", 2019 Computing in Cardiology Conference, 2019 (Non-Patent Document 4)); and - Principal component analysis (PCA) is a method of converting variables that are related to each other (correlated in statistical terms) into new variables that are non-correlated to each other (see, for example, Christopher Schilling, “Analysis of Atrial Electrograms”, Vol. 17 Karlsruhe Transactions on Biomedical Engineering, 2012 (Non-Patent Document 5)). [Prior art documents] [Non-patent literature]

[0006] [Non-Patent Document 1] Rieta, JJ, et al. “Atrial activity extraction based on blind source separation as an alternative to QRST cancellation for atrial fibrillation analysis.” Computers in Cardiology 2000. Vol. 27 (Cat. 00CH37163), IEEE, 2000 [Non-Patent Document 2] F. Castells, et al. “Multidimensional ICA for the Separation of Atrial and Ventricular Activities from Single Lead ECGs in Paroxysmal Atrial Fibrillation Episodes”, Lecture Notes in Computer Science, vol. 3195, p. 1229-1236, 2004 [Non-Patent Document 3] Widrow B., et al. “Adaptive noise cancelling: principles and applications”, Proc. IEEE 63, 1692-716, 1975 [Non-Patent Document 4] Simanto Saha, et al. “A Ventricular Far-field Artefact Filtering Technique for Atrial Electrograms”, 2019 Computing in Cardiology Conference, 2019 [Non-Patent Document 5] Christopher Schilling, “Analysis of Atrial Electrograms”, Vol. 17 Karlsruhe Transactions on Biomedical Engineering, 2012 Summary of the Invention [Problem to be solved by the invention]

[0007] All of the above techniques have major drawbacks: TMS distorts the atrial electrogram, making any techniques based on it ineffective; ICA loses most of its effectiveness as soon as the signal becomes irregular and disorganized; AVC is not as effective as ICA, and the effectiveness of the cancellation is highly dependent on a reference signal; PCA is an improvement over TMS, but has similar drawbacks.

[0008] In other words, none of the techniques can effectively reduce noise in electrogram signals to enable use in the setting of atrial arrhythmia. [Means for solving the problem]

[0009] The present invention improves this situation. For this purpose, the invention provides a device for processing intracardiac signals, comprising storage means arranged to receive electrocardiogram data and timed electrogram data, detection means arranged to analyse said electrocardiogram data and to detect QRS wave time points from said electrocardiogram data, analysis means arranged to perform a wavelet transformation of said electrogram data, extraction means arranged to derive from said wavelet transformation coefficients respectively corresponding to the QRS wave time points detected by said detection means and to store these in storage means, and configuration means arranged to extract a QRS fingerprint signal from said storage means, subtract said QRS fingerprint signal from the QRS wave time points of said wavelet transformation and generate as output noise-reduced electrogram data by inverse wavelet transformation of the obtained signal.

[0010] This device is extremely advantageous as it not only cancels out the far-field activity but also reconstructs the near-field that overlaps with the far-field, greatly improving the signal-to-noise ratio.

[0011] In various embodiments, the present invention may include one or more of the following features: - the storage means is arranged to receive electrogram data corresponding to separate tracks, and the detection means, the analysis means, the extraction means and the configuration means are arranged to separately process the electrogram data corresponding to the separate tracks; - the extraction means is arranged to derive, for selected coefficients corresponding to a given time point of a given wavelet transform level, coefficients such that a wavelet signal corresponding to the given wavelet level and centred on the given time point overlaps with a window centred on the QRS complex time point corresponding to each coefficient extracted from the electrogram data from which the selected coefficients are derived; - the extraction means is arranged to weight the coefficients stored in the accumulation means in response to a time overlap between the wavelet signal centred on a time point corresponding to each coefficient and corresponding to a wavelet level of each coefficient, and the window centred on the QRS complex time point corresponding to each coefficient; - the construction means is arranged to form, for each wavelet level of the wavelet transform, a QRS fingerprint signal respectively on the basis of a function of the coefficients derived for each wavelet level by the extraction means; - the constructing means is arranged to apply a function selected from the group consisting of geometric median, PCA and ICA; - the configuration means is arranged to use, for each wavelet level, a predetermined number of coefficients in the storage means, counting from the most recent ones; and said extraction means being arranged to perform a wavelet transform of the SWT type;

[0012] The present invention further provides a method for processing intracardiac signals, comprising the steps of: a) receiving electrocardiogram data and timed electrogram data; b) analyzing the electrocardiogram data and detecting QRS complex time points from the electrocardiogram data; c) performing a wavelet transform of the electrogram data; d) deriving coefficients from the wavelet transform, each of which corresponds to a time point of the QRS wave detected in step b), and storing these coefficients in a storage means; e) extracting a QRS fingerprint signal from said storage means and subtracting said QRS fingerprint signal from said QRS complex time points of said wavelet transform of step c); f) performing an inverse wavelet transform of the signal of step e) and returning as output corresponding noise-reduced electrogram data; The present invention relates to a method for processing an intracardiac signal, comprising:

[0013] In various embodiments, the method may include one or more of the following features: - step d) comprises the substep of deriving coefficients from the wavelet transform of step c) such that, for any coefficient corresponding to any time point of any wavelet transform level, a wavelet signal corresponding to the any wavelet level and centered on the any time point overlaps with a window centered on the QRS complex time point corresponding to the any coefficient, extracted from the electrogram data from which the any coefficient was derived; - step d) comprises the substep of weighting each derived coefficient before storing in the storage means in dependence on the time overlap between the wavelet signal centered on a time point corresponding to each derived coefficient and corresponding to the wavelet transform level of each derived coefficient, and the window centered on the QRS complex time point corresponding to each derived coefficient; - step e) comprises a substep of forming, for each wavelet level of the wavelet transformation of step b), a respective QRS fingerprint signal based on a function of the coefficients corresponding to each wavelet level of step d); - said function is selected from the group consisting of geometric median, PCA and ICA; - the function uses a predetermined number of coefficients in the storage means, counting from the most recent one, for each QRS fingerprint signal; and Step c) performs an SWT type wavelet transform.

[0014] The invention further 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 connected to a memory having such a computer program recorded thereon.

[0015] Other characteristics and advantages of the invention will become more apparent upon reading the following description, derived from non-limiting illustrative examples, with reference to the drawings, in which: [Brief description of the drawings]

[0016] [Figure 1] 1 is a schematic diagram of an apparatus according to the present invention; [Diagram 2] FIG. 2 illustrates one embodiment of the functions performed by the device of FIG. 1. [Diagram 3] 1 shows an example of electrograms of a signal before noise reduction, a signal after noise reduction, a superposition of these signals, and a difference therebetween. [Figure 4] 13 is another example of electrograms for a signal before noise reduction, a signal after noise reduction, a superposition of these signals, and the difference therebetween. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0017] The drawings and the following description substantially include structures / components of a specific nature, which may be utilized not only for a better understanding of the invention, but also contribute to the definition of the invention, where appropriate.

[0018] 1 is a schematic diagram of a device 2 for processing intracardiac signals according to the invention. The device 2 comprises storage means 4, detection means 6, analysis means 8, extraction means 10, construction means 12 and storage means 14.

[0019] The storage means 4 may consist of any type of data storage capable of receiving digital data, such as a hard disk, a solid state drive, any form of flash memory, a random access memory, a magnetic disk, local or distributed storage on the cloud, etc. Data resulting from the computation of the device may be stored in any type of storage means similar to the storage means 4, or in the storage means 4. These data may be erased or may be retained after the device has performed its task.

[0020] The storage means 4 receives various types of data: - electrogram data representing signals measured on one or more tracks from a cardiac catheter (where each data comes from a separate track, allowing for a collation of data corresponding to each separate track; for example, it is common to receive five separate tracks); - ECG data; - QRS complex time point data (obtained as described below); - QRS fingerprint data (obtained as described below); and - Electrogram data after noise reduction.

[0021] The storage means 14 may be formed as part of the storage means 4 or may be separate from the storage means 4. The storage means 14 may be realized by means similar to those described above with reference to the storage means 4.

[0022] Device 2 takes advantage of the diversity and complementarity of signals obtained by electrophysiological techniques.

[0023] As will be seen later, the detection means 6 first determine the location of the far-field QRS complex time points on the electrocardiogram, which are in fact aligned with the ventricular activity noise of the intracardiac branches, and then, by means of a time-frequency analysis, this noise is removed from the other electrogram without causing a reduction in the near-field activity, even in the case of far-field and near-field overlap.

[0024] The analysis means 8 and extraction means 10 separate the far-field components originating from the ventricles from the locally recorded (ie near-field) bipolar electrograms by implementing a time-frequency approach.

[0025] Using simultaneously recorded surface electrocardiogram and chamber signals, the chamber signals were decomposed into multiple time-frequency components.

[0026] The extraction means 10 collects and stores the far-field ventricular activities for each track. These activities are detected based on their synchronicity with the QRS complex time points that can be detected from the electrocardiogram data by a threshold method in the wavelet domain. These are stored in the storage means 14. Alternatively, the QRS complex time points may be detected by implementing the Pan-Tompkins algorithm.

[0027] Initially, nothing is stored in the storage means 14. Then, suddenly, there is enough storage to be able to calculate each fingerprint of each track, allowing the elimination of far-field activity.

[0028] The construction means 12 then subtracts these fingerprints from the wavelet domain before applying an inverse wavelet transform to reconstruct the noise-reduced signal.

[0029] The basic idea is that the atrial and ventricular activity can be viewed as statistically independent activities originating from two separate sources, and the intracardiac branch then consists of a mixture of atrial and ventricular components that can be effectively processed in the wavelet domain.

[0030] The detection means 6, the analysis means 8, the extraction means 10 and the configuration means 12 have direct or indirect access to the storage means 4. They may be formed in the form of suitable computer code executed by one or more processors. By processor it is to be understood that any processing means suitable for the calculations described below is meant. Such a processor may be in any form capable of providing the necessary computing power for the implementation described below, formed in any known manner, such as a microprocessor for a personal computer, laptop, tablet or smartphone computer, a dedicated chip of the FPGA or SoC type, a computing resource on a grid or cloud, a cluster of graphic processors (GPU), in the form of a microcontroller, etc. One or more of these elements may also be formed in the form of a dedicated electronic circuit, such as an ASIC. A combination of a processor and an electronic circuit may also be considered.

[0031] FIG. 2 is a diagram showing an example of the implementation of the noise reduction function performed by the device 2. As shown in FIG.

[0032] The first step 200 involves receiving synchronized electrogram and electrocardiogram data, typically in two second slices each.

[0033] Next, in step 210, the detection means 6 executes a function QRS( ) that receives the electrocardiogram data as an argument, and returns the aforementioned QRS wave time point detected from the electrocardiogram data.

[0034] In parallel or subsequently, the analysis means 8 executes a step 220, in which a function SWT() receives electrogram data as an argument and returns a number of wavelet decompositions, each corresponding to an individual track from which the electrogram data is derived. In the example described here, the function SWT() implements a stationary wavelet transform type algorithm (SWT), which has the advantage of compensating for the lack of shift invariance of the discrete wavelet transform (DWT), thereby avoiding a factor of two in signal size at each new decomposition level. Alternatively, a discrete wavelet transform or a continuous wavelet transform (CWT) may be used.

[0035] Once the wavelet transform and the QRS complex time points are obtained, the extraction means 10 executes the function Buf() in step 230. The function Buf() receives the wavelet transform and the QRS complex time points and stores the extraction from the wavelet transform in the storage means 14.

[0036] Since each wavelet level of each track is stored separately in the storage means 14, it is possible to determine for each extract in the storage means 14 which track it corresponds to and, moreover, which wavelet level it corresponds to.

[0037] This is important because the applicants have found it useful to store the wavelet coefficients corresponding to the QRS complex time points of each track, so that the components corresponding to the far field can be removed from the signal.

[0038] In particular, a QRS window is defined around a QRS complex time point that defines a period during which the electrogram signal is considered to contain a far-field signal. In the example described herein, this window is a 120 millisecond window around each QRS complex time point, which corresponds to the typical duration of a QRS complex in medical knowledge. Alternatively, the window may be defined in another way, for example by a function QRS().

[0039] Associated with each wavelet level is a wavelet function whose time is related to the wavelet transform and wavelet level used, i.e., each wavelet coefficient of the wavelet transform corresponds to a time window whose width is the wavelet function of that wavelet level, centered on the time point corresponding to that coefficient.

[0040] Thus, for each QRS complex time point of a given wavelet level and track, the function Buf( ) selects the wavelet transform coefficients of the time window that overlaps with the QRS window.

[0041] Although optional, applicants have found it advantageous to weight the wavelet coefficients stored in storage means 14 according to the amount of overlap between the time window of a given coefficient and the QRS window.

[0042] In fact, when the time point of the QRS window is the most extreme, the overlap with the time window centered on that time point is small. However, since the signal corresponding to this coefficient is the subject of suppression, there is a risk of suppressing signals that do not correspond to the far field. Similarly, the time window of higher level wavelets obviously has the same problem in that it can be larger than the QRS window. By applying weighting according to the relative overlap, such edge effects can be suppressed.

[0043] As another variation, a trade-off may be made between constraining the wavelet levels so that the widest time window size is coextensive with the QRS window range.

[0044] Once the storage means 14 is sufficiently stored (e.g., there are five or more extracts of each track in the storage means 14 from step 230), the construction means 12 can execute the function Out() in step 240. In the function Out(), the construction means 12 determines the median of each track and wavelet level of the extracts of interest and subtracts it from the wavelet transform of the electrogram data of the track and the corresponding wavelet level. The median is understood to be any technique that allows to determine a value that represents a series of extracts, whether it is obtained by geometric median, PCA or ICA. The function Out() finally performs an inverse wavelet transform on the signal from which the median has been subtracted, returning the corresponding signal with reduced noise.

[0045] 3 and 4 may illustrate the gain achieved by the device 2, particularly the gain in the overlapping far and near fields. These figures, from the top, illustrate electrogram data (i.e., the signal without noise reduction), the signal after noise reduction, the superposition of these signals, and their difference, and the effect on the inverse wavelet transform of subtracting the QRS fingerprint signal is clear.

[0046] These figures show the efficiency of the device 2 in completely reshaping the near-field signals that were overlapping with the far-field, without corrupting the signals outside these parts.

[0047] The applicant also carried out tests to quantify the gains achieved by the device 2. For this purpose, the applicant compared three different approaches based on electrograms with known cycle times and investigated their suitability. For this purpose, the applicant took the corresponding electrogram signal and made three copies of it: a first copy without any changes, a second copy in which the signal at the time of the ventricular QRS complex was suppressed (i.e., the signal was replaced by an isoelectric line that preserved the continuity of the signal), and a third copy in which the noise was reduced using the device 2. The cycle times determined using the obtained signals were then compared with the actual known cycle times.

[0048] The study was conducted on a group of 52 patients who could be in sinus rhythm, atrial tachycardia, or atrial fibrillation. For each replicate, a root mean square error score was calculated according to the following formula:

[0049]

number

[0050] (where c is a subscript indicating the replicate of interest, and y c (i) is the cycle time determined based on the i-th electrogram signal of the replica of subscript c, and y(i) is the actual cycle time.

[0051] The first copy scores 13.36%, the second copy scores 8.45%, and the third copy scores 4.06%.

[0052] Thus, it is demonstrated that device 2 provides a gain in signal to noise ratio.

Claims

1. 1. An apparatus for processing an intracardiac signal, comprising: storage means (4) arranged to receive the electrocardiogram data and the synchronized electrogram data; detection means (6) arranged to analyze the electrocardiogram data and detect QRS wave time points from the electrocardiogram data; analysis means (8) arranged to perform a wavelet transformation of said electrogram data; extraction means (10) for deriving coefficients corresponding to the QRS wave time points detected by the detection means (6) from the wavelet transform and storing them in a storage means (14); a configuration means (12) for extracting a QRS fingerprint signal from the storage means (14), subtracting the QRS fingerprint signal from the QRS wave time point of the wavelet transform, and performing an inverse wavelet transform on the resulting signal to generate noise-reduced electrogram data as an output; 1. A device for processing an intracardiac signal, comprising:

2. 2. The apparatus of claim 1, wherein the storage means (4) is arranged to receive electrogram data corresponding to separate tracks, and the detection means (6), the analysis means (8), the extraction means (10) and the configuration means (12) are arranged to process the electrogram data corresponding to the separate tracks separately.

3. 3. The apparatus according to claim 1, wherein the extraction means (10) is arranged to derive selected coefficients corresponding to a given time point at a given wavelet transform level such that a wavelet signal corresponding to the given wavelet level and centred on the given time point overlaps with a window centred on the QRS complex time point corresponding to each coefficient extracted from the electrogram data from which the selected coefficients are derived.

4. 4. The apparatus of claim 3, wherein the extraction means (10) is arranged to weight the coefficients stored in the accumulation means (14) in accordance with the temporal overlap of the wavelet signal centered on a time point corresponding to each coefficient and corresponding to the wavelet level of each coefficient, and the window centered on the QRS complex time point corresponding to each coefficient.

5. 3. The device according to claim 1 or 2, wherein the construction means (12) is arranged to form a QRS fingerprint signal for each wavelet level of the wavelet transform based respectively on a function of the coefficients derived for each wavelet level by the extraction means (10).

6. 6. The device according to claim 5, wherein said construction means (12) is arranged to apply a function selected from the group comprising geometric median, PCA and ICA.

7. 6. The apparatus of claim 5, wherein said configuration means (12) is arranged to use, for each wavelet level, a predetermined number of coefficients in said storage means (14), counting from the most recent.

8. 3. Device according to claim 1 or 2, wherein said extraction means (8) are arranged to perform a SWT type wavelet transform.

9. 1. A method for processing intracardiac signals, comprising: a) receiving electrocardiogram data and synchronized electrogram data; b) analyzing the electrocardiogram data and detecting QRS wave time points from the electrocardiogram data; c) performing a wavelet transform of the electrogram data; d) deriving coefficients corresponding to the QRS wave time points detected in step b) from the wavelet transform and storing them in a storage means (14); e) extracting a QRS fingerprint signal from said storage means (14) and subtracting said QRS fingerprint signal from said QRS complex time points of said wavelet transform of step c); f) performing an inverse wavelet transform of the signal of step e) and returning as output corresponding noise-reduced electrogram data; A method for processing intracardiac signals, comprising:

10. 10. The method of claim 9, wherein step d) includes the substep of deriving coefficients from the wavelet transform of step c) for selected coefficients corresponding to a given time point at a given wavelet transform level such that a wavelet signal corresponding to the given wavelet level and centered at the given time point overlaps with a window centered at the QRS complex time point corresponding to the selected coefficient extracted from the electrogram data from which the selected coefficient was derived.

11. 11. The method of claim 10, wherein step d) includes the substep of weighting each derived coefficient before storing it in the storage means (14) in accordance with the temporal overlap of the wavelet signal centered on the time point corresponding to each derived coefficient and corresponding to the wavelet transform level of each derived coefficient with the window centered on the QRS complex time point corresponding to each derived coefficient.

12. 12. The method according to claim 9, wherein step e) includes a sub-step of forming each QRS fingerprint signal for each wavelet level of the wavelet transform of step b) based on a function of the coefficients corresponding to each wavelet level of step d).

13. 13. The method of claim 12, wherein the function is selected from the group comprising geometric median, PCA, and ICA.

14. A computer program comprising instructions for performing the method according to any one of claims 9 to 11.

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

Images