Local activation time analysis system

Abstract

The invention is entitled "Local Activation Time Analysis System." In one embodiment, a method for finding local activation times of intracardiac electrogram (IEGM) signals includes receiving a first IEGM signal and corresponding local activation time annotations of the first IEGM signal manually annotated by a respective annotator from an electrophysiology laboratory subsystem; training an artificial neural network to find local activation times of IEGM signals in response to the first IEGM signal and the corresponding local activation time annotations; receiving a second IEGM signal; and applying the trained artificial neural network to the received second IEGM signal to provide an indication of local activation times of the received second IEGM signal.

Description

[0001] Related Application Information

[0002] This application claims the benefit of U.S. Provisional Patent Application 63 / 077,780, filed September 14, 2020, the disclosure of which is incorporated herein by reference. Technical Field

[0003] This invention relates to medical systems, and specifically, but not exclusively, to the processing of cardiac signals. Background Technology

[0004] Electrode catheters have been widely used in medical practice for many years. They are used to stimulate and map electrical activity in the heart, as well as to ablate sites of abnormal electrical activity. In use, the electrode catheter is inserted into a major vein or artery, such as the femoral vein, and then guided to the cardiac chamber of interest. A typical ablation procedure involves inserting a catheter with one or more electrodes at its distal end into the cardiac chamber. A reference electrode can be provided, typically taped to the patient's skin, or a second catheter positioned in or near the heart can be used to provide the reference electrode. RF (radio frequency) current is applied between the catheter electrodes of the ablation catheter and an unrelated electrode (which may be one of the catheter electrodes), and the current flows through the medium between these electrodes (i.e., blood and tissue). The current distribution can depend on the amount of contact between the electrode surface and the tissue compared to blood, which has a higher conductivity than tissue. Heating of the tissue occurs due to its resistance. The tissue is sufficiently heated to destroy cells in the cardiac tissue, resulting in the formation of a non-conductive ablation focus within the cardiac tissue. In some applications, irreversible electroporation can be performed to ablate the tissue.

[0005] Electrophysiological (EP) cardiac mapping is a diagnostic medical procedure used to identify the location of cardiac dysfunction within the heart. Time-varying electrocardiogram (ECG) signals are received by electrodes that contact points along the surface of the patient's heart. The signals are processed, and different measures of cardiac function are calculated based on the processed (ECG) signals, which are then spatially mapped onto an image of the heart. The mapping map is then output for analysis by medical professionals.

[0006] Analysis of cardiac signals sometimes involves timing synchronization with ECG signals. For example, U.S. Patent Application Publication No. 2013 / 0123652 describes a method for analyzing signals, including sensing a time-varying intracardiac potential signal and detecting a fit between the time-varying intracardiac potential signal and a predefined oscillating waveform. The method also includes estimating the annotation time of the signal in response to the fit.

[0007] Electrophysiological (EP) cardiac mapping, or cardiac electroanatomical mapping, is used to identify areas of dysfunction within the heart tissue. An in vivo probe (typically a catheter with multiple mapping electrodes positioned near the distal end of the catheter body) is inserted into the heart chambers. Time-varying electrocardiogram (ECG) signals are recorded at multiple contact points between the mapping electrodes and the heart tissue. These ECG electrodes are then moved to different contact locations within the heart tissue, and the process is repeated. Measures of cardiac function are then calculated based on the local ECG signals, which are spatially mapped onto the surface of the heart chambers. This mapping helps healthcare professionals identify areas of cardiac dysfunction.

[0008] Power sources in the heart, such as the sinoatrial (SA) node and the atrioventricular (AV) node, initiate electrical activity waves that propagate throughout the heart, triggering the contraction of muscular tissue in the atria and ventricles into a characteristic sinus rhythm. Characteristic ECG waveforms are detected at these mapping electrodes as the active wavefront arrives at them during each cardiac cycle. These waveforms are time-shifted due to the different arrival times of the same wavefront at different electrodes that contact tissue at different spatial locations along the surface of the heart chambers.

[0009] The arrival times of ECG waveforms detected at these multiple mapping electrodes can be used to map the propagation time and / or velocity of active waves on the heart. Mapping of active waves is performed relative to a single time reference indicating the cardiac cycle (referred to herein as the reference annotation time).

[0010] Reference annotation time can be calculated by processing ECG signals obtained from surface (BS) electrodes or intracardiac (IC) reference electrodes on an additional catheter and placed in contact with the surface of the heart chambers. Typically, physicians specify whether the reference annotation time is calculated from the BS channel or the IC channel based on the suspected pathology. Summary of the Invention

[0011] According to embodiments of this disclosure, a method for finding local activation times of an intracardiac electrogram (IEGM) signal is provided, comprising: receiving a first IEGM signal and a corresponding local activation time annotation of the first IEGM signal manually annotated by a corresponding annotator from an electrophysiology laboratory subsystem; training an artificial neural network in response to the first IEGM signal and the corresponding local activation time annotation to find the local activation time of the IEGM signal; receiving a second IEGM signal; and applying the trained artificial neural network to the received second IEGM signal to provide an indication of the local activation time of the received second IEGM signal.

[0012] Further according to an embodiment of this disclosure, the method includes calculating the weight of an annotation performed by the corresponding annotator among the annotators in response to the local activation time annotation experience level of the corresponding annotator among the annotators, wherein the training includes training the artificial neural network in response to the first IEGM signal and the corresponding local activation time annotation to find the local activation time of the IEGM signal, the corresponding local activation time annotation being weighted according to a corresponding weight among the calculated weights of the corresponding annotator among the annotators who annotated the corresponding local activation time annotation in the local activation time annotation.

[0013] Furthermore, according to an embodiment of this disclosure, the method includes searching a database of published scientific literature in response to a corresponding number of search matches indicating the local activation time annotation experience level of the corresponding annotator among the annotators, and wherein the calculation includes calculating the weight of the annotation performed by the corresponding annotator among the annotators in response to the corresponding number of search matches of the corresponding annotator among the annotators.

[0014] Furthermore, according to the embodiments of this disclosure, the search is limited to the publication of scientific literature describing local activation time annotations.

[0015] Furthermore, according to the embodiments of this disclosure, the corresponding number of search matches is the corresponding number of scientific publications that match the corresponding annotators among the annotators.

[0016] Further according to an embodiment of this disclosure, the training includes: inputting the first IEGM signal into the artificial neural network; and iteratively adjusting the parameters of the artificial neural network in response to the output of the artificial neural network and the local activation time annotation of the first IEGM signal.

[0017] Furthermore, according to an embodiment of this disclosure, the method includes minimizing a loss function that is a function of the output of the artificial neural network and the local activation time annotation of the first IEGM signal weighted according to the corresponding weights in the calculated weights, wherein the iterative adjustment is performed in response to minimizing the loss function.

[0018] Furthermore, according to embodiments of this disclosure, the loss function includes a binary cross-entropy loss function.

[0019] Furthermore, according to embodiments of this disclosure, the method includes generating an electroanatomical mapping in response to the indication of the local activation time.

[0020] According to another embodiment of this disclosure, a system for finding local activation times of intracardiac electrogrammation (IEGM) signals is also provided, including a remote server comprising processing circuitry configured to: receive a first IEGM signal and a corresponding local activation time annotation of the first IEGM signal manually annotated by a corresponding annotator from an electrophysiology laboratory subsystem; train an artificial neural network in response to the first IEGM signal and the corresponding local activation time annotation to find the local activation times of the IEGM signal; receive a second IEGM signal; and apply the trained artificial neural network to the received second IEGM signal to provide an indication of the local activation times of the received second IEGM signal.

[0021] Further according to an embodiment of this disclosure, the system includes the electrophysiology laboratory subsystem, each electrophysiology laboratory subsystem including: a catheter configured to be inserted into at least one heart chamber of at least one living subject and to capture a corresponding IEGM signal from the first IEGM signal from the at least one heart chamber; a display; and processing circuitry configured to present the corresponding IEGM signal from the first IEGM signal to the display, receive the corresponding local activation time annotation from the local activation time annotation of the displayed first IEGM signal manually annotated by a corresponding annotator among the annotators, and provide the corresponding IEGM signal from the first IEGM signal and the corresponding local activation time annotation from the local activation time annotation to the remote server.

[0022] Furthermore, according to an embodiment of this disclosure, the processing circuit is configured to: calculate the weight of the annotation performed by the corresponding annotator among the annotators in response to the local activation time annotation experience level of the corresponding annotator among the annotators; and train the artificial neural network in response to the first IEGM signal and the corresponding local activation time annotation to find the local activation time of the IEGM signal, the corresponding local activation time annotation being weighted according to the corresponding weight of the corresponding annotator among the annotators who annotated the corresponding local activation time annotation in the local activation time annotation.

[0023] Furthermore, according to an embodiment of this disclosure, the processing circuit is configured to search a database of published scientific literature in response to a corresponding number of search matches indicating the local activation time annotation experience level of the corresponding annotator among the annotators; and to calculate the weight of the annotation performed by the corresponding annotator among the annotators in response to the corresponding number of search matches of the corresponding annotator among the annotators.

[0024] Furthermore, according to an embodiment of this disclosure, the processing circuit is configured to limit the search of the database to scientific publications describing local activation time annotations.

[0025] Further according to the embodiments of this disclosure, the corresponding number of search matches is the corresponding number of scientific publications that match the corresponding annotators among the annotators.

[0026] Furthermore, according to an embodiment of the present disclosure, the processing circuit is configured to input the first IEGM signal into the artificial neural network; and to iteratively adjust the parameters of the artificial neural network in response to the output of the artificial neural network and the local activation time annotation of the first IEGM signal.

[0027] Furthermore, according to an embodiment of this disclosure, the processing circuit is configured to: minimize a loss function, which is a function of the output of the artificial neural network and the local activation time annotation of the first IEGM signal weighted according to the corresponding weights in the calculated weights; and iteratively adjust the parameters of the artificial neural network in response to minimizing the loss function.

[0028] Furthermore, according to embodiments of this disclosure, the loss function includes a binary cross-entropy loss function.

[0029] Further according to an embodiment of the present disclosure, the processing circuit is configured to generate an electroanatomical mapping in response to the indication of the local activation time.

[0030] According to another embodiment of this disclosure, a software product is also provided, including a non-transitory computer-readable medium storing program instructions that, when read by a central processing unit (CPU), cause the CPU to: receive a first IEGM signal and a corresponding local activation time annotation of the first IEGM signal manually annotated by a corresponding annotator from an electrophysiology laboratory subsystem; train an artificial neural network in response to the first IEGM signal and the corresponding local activation time annotation to find the local activation time of the IEGM signal; receive a second IEGM signal; and apply the trained artificial neural network to the received second IEGM signal to provide an indication of the local activation time of the received second IEGM signal. Attached Figure Description

[0031] The invention will be understood from the following detailed description taken in conjunction with the accompanying drawings, wherein:

[0032] Figure 1 A schematic diagram illustrating a system for performing catheter insertion surgery on the heart, constructed and operated according to an exemplary embodiment of the present invention;

[0033] Figure 2 For includingFigure 1 A flowchart of the steps in the operation method of the electrophysiology laboratory subsystem in the system;

[0034] Figure 3 For use Figure 1 The system presents a view of the annotated intracardiac electrocardiogram signals.

[0035] Figure 4 for Figure 1 A block diagram of the system;

[0036] Figure 5 For including Figure 1 A flowchart of the steps involved in operating a remote server in the system;

[0037] Figure 6 For use with Figure 1 A schematic diagram of an artificial neural network used in a system;

[0038] Figure 7 For including Figure 5 A flowchart of the sub-steps in the method;

[0039] Figure 8 For including use Figure 6 A flowchart of the steps in a method for processing intracardiac electrocardiogram signals using a trained artificial neural network; and

[0040] Figure 9 For the reason Figure 1 The system presents a schematic diagram showing the electroanatomical mapping. Detailed Implementation

[0041] SUMMARY

[0042] One of the major challenges in electrophysiology (EP) is finding an accurate intracardiac electrogram (IEGM) signal annotation algorithm to select the correct Local Activation Time (LAT) from the IEMGM signal, for example, to generate a LAT mapping. One approach for automatically finding the LAT involves finding the maximum negative slope of the signal within each window of interest (WOI) and assigning the LAT to that point. The WOI is typically set by the system based on the QRS complexes of signals captured by one or more surface electrodes. Detecting the maximum negative slope does not provide a comprehensive solution to the aforementioned problem, as in some cases, many such slopes may exist within the WOI and the algorithm may select the incorrect slope. Therefore, physicians have the option to manually adjust the calculated LAT to different points within the WOI.

[0043] The embodiments of the present invention address the aforementioned problems by using deep learning techniques to train an artificial neural network (ANN) to find the corresponding LAT for the IEGM signal. The ANN is trained using IEGM signals provided by different EP laboratories (laboratories) and corresponding LAT annotations manually annotated by annotators (e.g., physicians or other medical professionals). Manual annotation can correct automatically calculated annotations or provide initial annotations for the IEGM signal.

[0044] In some implementations, different LAT annotations are not given the same weights during ANN training. For example, the loss function used to train the ANN may include weights that can be associated with each corresponding IEGM signal and LAT annotation pair to weight the contribution of each pair during training. In some implementations, the binary cross-entropy (BCE) loss function may be used.

[0045] In some implementations, the weights assigned to the IEGM signal and LAT annotation pair (e.g., for the loss function) can be calculated in response to the LAT annotation experience level of the annotator who manually determined the LAT annotation. In this way, the ANN can be trained by assigning greater weights to more experienced annotators.

[0046] The LAT annotation experience level of different annotators can be estimated by searching databases including published scientific literature to find the number of search matches (e.g., the number of published scientific literature) for different annotators in the LAT annotation field. For example, a search for "John Smith and LAT annotations" might yield 50 matches, while a search for "Tim Jones and LAT annotations" might yield 400 matches. In this case, LAT annotations provided by Tim Jones would receive a much larger weight in training than LAT annotations provided by John Smith. Any suitable database can be searched, such as Google Scholar, Scopus, or ScienceNet.

[0047] System Description

[0048] Now for reference Figure 1This is a schematic diagram illustrating a medical system 10 for performing catheter insertion surgery on the heart 12, constructed and operated according to an exemplary embodiment of the present invention. The medical system 10 can be configured to assess electrical activity on the heart 12 of a living subject and perform ablation surgery thereon. System 10 includes an EP laboratory subsystem 11 (only one is shown for simplicity) that captures EP data. The medical system 10 also includes a remote server 26 (e.g., a cloud computing device) from which the EP data is transmitted by the EP laboratory subsystem 11 via a network 27 for storage and / or processing. The EP data may be compressed in the EP laboratory subsystem 11 and transmitted in compressed form to the remote server 26 via the network 27.

[0049] The following description, by way of example only, details one of the EP laboratory subsystems 11. Different EP laboratory subsystems 11 may include the same or different EP laboratory equipment to provide EP data to a remote server 26 for processing. Figure 1 The EP laboratory subsystem 11 includes a catheter 14, which is inserted by an operator 16 through the skin across the patient's vascular system into a chamber or vascular structure of the heart 12. The operator 16 (typically a physician) brings the distal end 18 of the catheter into contact with the heart wall, for example, at the ablation target site. An electrical activity mapping can be prepared according to methods disclosed in U.S. Patents 6,226,542, 6,301,496, and 6,892,091. A commercially available embodiment of elements of system 10 may be... 3. This system is derived from Biosense Webster, Inc., Irvine, CA. This system can be modified by those skilled in the art to implement the principles of the invention described herein.

[0050] Areas identified as abnormal, for example, by assessment using an electrical activity mapping technique, can be ablated by applying thermal energy. This is achieved, for instance, by conducting radiofrequency current through a catheter via wires to one or more electrodes at the distal end 18, which apply radiofrequency energy to the myocardium. The energy is absorbed in the tissue, heating it to a temperature at which it permanently loses its electrical excitability. Upon successful completion of this procedure, non-conductive ablation foci are formed in the cardiac tissue, disrupting abnormal electrical pathways leading to arrhythmias. The principles of this invention can be applied to different cardiac chambers to diagnose and treat a variety of different arrhythmias.

[0051] The catheter 14 typically includes a handle 20 with suitable controls to allow the operator 16 to manipulate, position, and orient the distal end 18 of the catheter 14 as required for ablation. To assist the operator 16, the distal portion of the catheter 14 includes position sensors (not shown) that provide signals to processing circuitry 22 located in the console 24. Processing circuitry 22 can perform several processing functions as described below.

[0052] The lead connector 35 can connect the console 24 to the surface electrode 30 and other components of the positioning subsystem for measuring the position and orientation coordinates of the catheter 14. Processing circuitry 22 or another processor (not shown) may be a component of the positioning subsystem. The catheter electrode 31 and the surface electrode 30 can be used to measure tissue impedance at the ablation site, as taught in U.S. Patent No. 7,536,218. A temperature sensor (not shown), typically a thermocouple or thermistor, may be mounted on the ablation surface of the distal portion of the catheter 14 as described below.

[0053] The console 24 typically includes one or more ablation power generators 25. The catheter 14 may be adapted to deliver ablation energy (e.g., radiofrequency energy, ultrasound energy, irreversible electroporation, and laser-generated light energy) to the heart using any known ablation technique. Such methods are disclosed in U.S. Patents 6,814,733, 6,997,924, and 7,156,816.

[0054] In one embodiment, the positioning subsystem includes a magnetic positioning tracking arrangement that utilizes a magnetic field generating coil 28 to determine the position and orientation of the catheter 14 by generating a magnetic field at a predetermined working volume and sensing these magnetic fields at the catheter. This positioning subsystem is described in U.S. Patents 7,756,576 and 7,536,218.

[0055] As described above, catheter 14 is coupled to console 24, enabling operator 16 to observe and control the function of catheter 14. Console 24 includes processing circuitry 22, typically a computer with appropriate signal processing circuitry. Processing circuitry 22 is coupled to drive display 29 (e.g., a monitor). The signal processing circuitry typically receives, amplifies, filters, and digitizes signals from catheter 14, including signals generated by sensors such as electrical sensors, temperature sensors, and contact force sensors, and position sensing electrodes 31 located distal to catheter 14. The digitized signals are received by console 24 and the positioning system to calculate the position and orientation of catheter 14 and to analyze the electrical signals from the electrodes. In some embodiments, the digitized signals are sent (and optionally compressed before transmission) to remote server 26 to calculate position and orientation data, and / or analyze the electrical signals from electrodes 30, 31, and / or use the electrical signals and associated data to train an artificial neural network, as described in more detail below.

[0056] To generate electroanatomical mapping maps, processing circuitry 22 typically includes a mapping module that includes an electroanatomical mapping map generator, an image registration procedure, an image or data analysis procedure, and a graphical user interface configured to display graphical information on display 29. In some embodiments, some or all of the mapping module's functions are performed by a remote server 26.

[0057] Typically, system 10 includes other components, but these are not shown in the figures for simplicity. For example, system 10 may include an electrocardiogram (ECG) monitor coupled to receive signals from one or more surface electrodes 30 to provide ECG synchronization signals to console 24 or remote server 26. In some embodiments, some or all of the functions of the ECG monitor are performed by remote server 26. As mentioned above, system 10 typically also includes a reference position sensor, either located on an externally applied reference patch attached to the outside of the subject's body or on an internal catheter inserted into and held in a fixed position relative to heart 12. Conventional pumps and tubing may be provided to circulate fluid through catheter 14 to cool the ablation site. System 10 and / or remote server 26 may receive image data from external imaging modalities such as MRI units and include an image processor that may be incorporated into or invoked (e.g., by processing circuitry 22 and / or remote server 26) in processing circuitry and / or remote server 26 for generating and displaying images.

[0058] In implementation, some or all of the functions of processing circuitry 22 may be combined in a single physical component, or alternatively, implemented using multiple physical components. These physical components may include hardwired or programmable devices, or a combination of both. In some embodiments, at least some of the functions of processing circuitry 22 may be implemented by a programmable processor under the control of suitable software. This software may be downloaded to the device electronically via, for example, a network. Alternatively or otherwise, the software may be stored in a tangible, non-transitory computer-readable storage medium, such as optical, magnetic, or electronic memory.

[0059] Now for reference Figure 2 and Figure 3 . Figure 2 For including Figure 1 A flowchart 40 shows the steps of the operation method of one of the electrophysiology laboratory subsystems 11 in system 10. Figure 3 For use Figure 1 The system 10 presents a view of the displayed IEGM signal 48 with annotations.

[0060] Each EP laboratory subsystem 11 includes a catheter 14 configured to be inserted into at least one heart chamber of at least one living subject and to capture a corresponding IEGM signal 48 from at least one heart chamber. One or more catheters 14 of any suitable type may be used in each EP laboratory subsystem 11. Processing circuitry 22 is configured to receive the IEGM signal 48 and present (box 42) a representation of the IEGM signal 48 to a display 29, and to receive (box 44) a Local Activation Time (LAT) annotation 50 for the corresponding IEGM signal 48 manually annotated by an annotator of the EP laboratory subsystem 11. The IEGM signal 48 may initially be automatically annotated by an algorithm running on processing circuitry 22. The annotator may then correct the automatic annotation with manual annotation by marking the LAT annotation 50 on the displayed IEGM signal 48. In some embodiments, the annotator may view the displayed IEGM signal 48 on display 29 (where processing circuitry 22 does not calculate the automatic LAT annotation) and then mark the LAT annotation 50 on the displayed IEGM signal 48.

[0061] The processing circuitry 22 of the EP Lab subsystem 11 is configured to provide (box 46) the IEGM signal 48 and the corresponding local activation timing annotation 50 to the remote server 26. In some embodiments, the EP Lab subsystem 11 is configured to communicate via a suitable network interface through network 27. Figure 1 Provide the IEGM signal 48 and the corresponding local activation time annotation 50 to the remote server 26.

[0062] Now for reference Figure 4 andFigure 5 . Figure 4 for Figure 1 Block diagram of System 10. Figure 5 For including Figure 1 The flowchart 100 shows the steps in the operation method of the remote server 26 in system 10.

[0063] The remote server 26 includes processing circuitry 60, memory 62, data bus 64, and network interface 66. Processing circuitry 60 is configured to run software to perform various signal processing and computational tasks, including an artificial neural network 68, a training module 70, and a calibration module 72. Training module 70 is configured to train the artificial neural network 68, as described below. Figures 5 to 7 A more detailed description is provided. Mapping module 72 is configured to generate an EP mapping map in response to cardiac signals and other data captured from a living subject, as described in the reference... Figure 8 and Figure 9 To describe in more detail.

[0064] In implementation, some or all of the functions of the processing circuitry 60 may be combined in a single physical component, or alternatively, implemented using multiple physical components. These physical components may include hardwired or programmable devices, or a combination of both. In some embodiments, at least some of the functions of the processing circuitry 60 may be implemented by a programmable processor under the control of suitable software. This software may be downloaded to the device electronically via, for example, a network. Alternatively or otherwise, the software may be stored in a tangible, non-transitory computer-readable storage medium, such as optical, magnetic, or electronic memory.

[0065] The memory 62 is configured to store data used by the processing circuitry 60. The data bus 64 is configured to transfer data between various components of the remote server 26 (e.g., between the processing circuitry 60 and the network interface 66).

[0066] The training module 70, operating on the processing circuitry 60, is configured to receive (box 102) IEGM signals 48 (captured in the corresponding EP lab subsystem 11) and corresponding local activation time annotations 50 of the IEGM signals 48 manually annotated by the corresponding annotators. In other words, the training module 70 is configured to receive IEGM signals 48 and corresponding local activation time annotations 50 manually annotated by one annotator in one of the EP lab subsystems 11, as well as other IEGM signals 48 and corresponding local activation time annotations 50 manually annotated by another annotator in another of the EP lab subsystems 11, etc.

[0067] The training module 70, running on the processing circuit 60, is configured to train (box 104) the artificial neural network 68 in response to training data including the IEGM signal 48 and corresponding local activation time annotations 50 to find the local activation times of the IEGM signal. The steps in box 104 include the sub-steps in boxes 106 to 110, which are described in more detail below.

[0068] The training module 70, running on the processing circuit 60, is configured to search (box 106) databases of scientific publications (e.g., Google Scholar, Scopus, or ScienceNet) in response to a corresponding annotator (who supplies local activation time annotations 50) as a search string, thereby generating a corresponding number of search matches indicating the level of local activation time annotation experience of the corresponding annotator. In other words, the database is searched using different search strings for each annotator (e.g., “J. Smith”, “T. Jones”, etc.) to generate multiple search matches for each annotator (e.g., 5 matches for J. Smith and 40 matches for T. Jones, etc.). In some cases, a search for a given annotator may not produce any search matches. Any suitable software script can be used to perform the search, for example, using a web crawler such as pybliometrics 2.5.0 to access Scopus. The corresponding number of search matches can be the corresponding number of scientific publications that match the corresponding annotator (e.g., 5 publications for J. Smith and 40 publications for T. Jones, etc.).

[0069] In some implementations, the training module 70, running on the processing circuit 60, is configured to limit the search of the database to scientific publications describing local activation time annotations and / or IEG and / or ECG and / or EP annotations, etc. Limiting the search to one or more of the above is useful for preventing stray results. For example, there may be J. Smith who has published articles in nuclear physics, and therefore his experience is irrelevant to J. Smith who provides the local activation time annotation 50.

[0070] The training module 70, running on the processing circuit 60, is configured to calculate (box 108) the weights of annotations performed by the respective annotators in response to the local activation time annotation experience level of the annotator. For example, calculating the weights of annotations performed by J. Smith, and calculating another weight for annotations performed by T. Jones, etc.

[0071] In some implementations, a training module 70 running on processing circuitry 60 is configured to calculate the weight of an annotation performed by a given annotator in response to a corresponding number of search matches (e.g., the number of scientific publications). For example, a weight is calculated for an annotation performed by J. Smith in response to 5 search matches (e.g., 5 publications) found in the database for J. Smith, and another weight is calculated for an annotation performed by T. Jones in response to 40 search matches (e.g., 40 publications) found in the database for T. Jones. The weights can be calculated proportionally. For example, if there are N annotators, and the j-th annotator has P found in the search... j If a post is published, then the weight W of the i-th commenter is... i equal:

[0072]

[0073] The training module 70, running on the processing circuit 60, is configured to train (box 110) the artificial neural network 68 in response to training data to find the local activation times of the IEGM signal. This training data includes the IEGM signal 48 (received from the EP laboratory subsystem 11) and corresponding local activation time annotations 50 weighted according to the calculated weights of the corresponding annotators who provided the annotations. For example, annotations provided by J. Smith are weighted according to weights calculated for J. Smith, and annotations provided by T. Jones are weighted according to weights calculated for T. Jones, etc.

[0074] Now for reference Figure 6 and Figure 7 . Figure 6 For use with Figure 1 A schematic diagram of the artificial neural network 68 used in System 10. Figure 7 For including Figure 5 The flowchart of the sub-steps in the steps of the method in box 110.

[0075] A neural network is a network or circuit of neurons, or, in the modern sense, an artificial neural network composed of artificial neurons or nodes. The connections of biological neurons are modeled as weights. Positive weights reflect excitatory connections, while negative values ​​represent inhibitory connections. The input is modified by the weights and summed using a linear combination. Activation functions control the amplitude of the output. For example, an acceptable output range is typically between 0 and 1, or the range can be between -1 and 1.

[0076] These artificial networks can be used for predictive modeling, adaptive control, and applications, and can be trained on datasets. Experience-based self-learning can occur within the network, drawing conclusions from complex and seemingly unrelated groups of information.

[0077] For completeness, a biological neural network consists of one or more groups of chemically connected or functionally related neurons. A single neuron can connect to many other neurons, and the total number of neurons and connections in a network can be extensive. Connections (called synapses) typically form from an axon to a dendrite, but dendritic synapses and other connections are also possible. In addition to electrical signals, other forms of signaling, caused by the diffusion of neurotransmitters, exist.

[0078] Artificial intelligence, cognitive modeling, and neural networks are information processing paradigms inspired by how biological nervous systems process data. Artificial intelligence and cognitive modeling attempt to simulate some characteristics of biological neural networks. In the field of artificial intelligence, artificial neural networks have been successfully applied to speech recognition, image analysis, and adaptive control to build software agents or autonomous robots (in computers and video games).

[0079] In terms of artificial neurons, often referred to as artificial neural networks (ANNs) or simulated neural networks (SNNs), a neural network (NN) is a set of interconnected natural or artificial neurons that process information using mathematical or computational models based on computational connection methods. In most cases, an ANN is an adaptive system that adapts its structure based on the flow of external or internal information through the network. More practically, the term neural network is used for modeling nonlinear statistical data or as a decision-making tool. These terms can be used to model complex relationships between inputs and outputs or to find patterns in data.

[0080] In some embodiments, the artificial neural network 68 includes a fully connected neural network (e.g., a convolutional neural network). In other embodiments, the artificial neural network 68 may include any suitable ANN. The artificial neural network 68 may include processing circuitry 60 ( Figure 4 The software that executes and / or the hardware modules that are configured to perform the functions of the artificial neural network 68.

[0081] Artificial neural network 68 includes an input layer 80 from which input is received, and one or more hidden layers 82 that progressively process the input to an output layer 84 from which the output of artificial neural network 68 is provided. Artificial neural network 68 may include layer weights between layers 80, 82, and 84. Artificial neural network 68 manipulates data received at input layer 80 based on the values ​​of the respective layer weights between layers 80, 82, and 84.

[0082] During the training of the artificial neural network 68, the layer weights of the artificial neural network 68 are updated, so that the artificial neural network 68 can perform the data manipulation tasks that the artificial neural network 68 was trained to perform.

[0083] The number and width of layers in the artificial neural network 68 can be configured. As the number and width of layers increase, the accuracy with which the artificial neural network 68 can manipulate data according to the task at hand also increases. However, a larger number of layers and wider layers typically require more training data, more training time, and training may not converge. For example, the input layer 80 may include 400 neurons (e.g., to compress a batch of 400 samples), and the output layer may also include 400 neurons.

[0084] Training an artificial neural network 68 is typically an iterative process. One method for training the artificial neural network 68 is now described below. A training module 70 runs on the processing circuit 60. Figure 4 The parameters (e.g., layer weights) of the artificial neural network 68 are configured to iteratively adjust (box 112) to reduce the difference between the output of the artificial neural network 68 and the local activation time annotation 50 of the IEGM signal 48.

[0085] In some implementations, a training module 70 runs on the processing circuit 60. Figure 4 The configuration is set to minimize the loss function, which is the output of the artificial neural network 68 and the weights calculated based on the weights. Figure 5 The loss function is calculated in step 108 as a function of the local activation time annotations 50 of the IEGM signal 48 weighted by the corresponding weights in the calculation; and the parameters (e.g., layer weights) of the artificial neural network 68 are iteratively adjusted in response to minimizing the loss function. In some implementations, the loss function includes the binary cross-entropy (BCE) loss function. Examples of suitable BCE loss functions are provided on Pytorch.org.

[0086] The sub-steps of the steps in box 112 are now described below.

[0087] The training module 70 runs on the processing circuit 60 of the processing circuit 22. Figure 4 The IEGM signal 48 is configured to be input (box 114) into the input layer 80 of the artificial neural network 68. The training module 70, which runs on the processing circuitry 60, is also configured to... Figure 4 The system is configured to compare the output of the artificial neural network 68 (e.g., the output of the output layer 84) with the desired output (i.e., the corresponding local activation time annotation 50 of the IEGM signal 48) (box 116), for example, using a suitable loss function that takes into account the weights of the corresponding local activation time annotation 50 (calculated for the corresponding annotator).

[0088] The output of the artificial neural network 68 includes various vectors corresponding to the IEGM signal 48 input to the artificial neural network 68. Each output vector includes a component, each component having a floating-point value (e.g., between 0 and 1). Each desired output is represented as a one-hot vector, where all components of the vector have zero values, except for one of the components having a value of one corresponding to the time value of the corresponding local activation time annotation 50.

[0089] For example, if there exists a set of vectors A, B, C output by artificial neural network 68 and a set of corresponding vectors representing corresponding local activation time annotations A', B', and C', then based on the weights of the annotators who annotated A', B', and C', the training module 70 of processing circuit 60 ( Figure 4 Use loss functions to compare A with A', B with B', C with C', etc.

[0090] At decision box 118, training module 70 runs on processing circuitry 60. Figure 4 The system is configured to determine whether the difference between the output of the artificial neural network 68 and the desired output is small enough. If the difference between the output of the artificial neural network 68 and the desired output is small enough (branch 120), then the training module 70 running on the processing circuit 60... Figure 4 It is configured to save (box 122) the parameters (e.g., layer weights) of the artificial neural network 68 for future use.

[0091] If the difference is not small enough (branch 124), then the training module 70 runs on the processing circuit 60. Figure 4 The training module 70, configured to modify (box 126) the parameters (e.g., layer weights) of the artificial neural network 68 to reduce the difference between the outputs of the artificial neural network 68 and the desired outputs of the artificial neural network 68 according to a loss function. According to the loss function, the difference minimized in the above example is the total difference between all outputs of the artificial neural network 68 and all desired outputs (e.g., local activation time annotations 50). Figure 4 The parameters are configured to be modified using any suitable optimization algorithm (e.g., gradient descent algorithms such as the Adam optimization algorithm). Then repeat the steps in boxes 114-118.

[0092] Now for reference Figure 8 The diagram shows the use of Figure 6 A flowchart 150 shows the steps in a method for processing intracardiac electrocardiogram signals using a trained artificial neural network 68. See also: Figure 4 .

[0093] A mapping module 72 (or any other suitable module) operating on processing circuitry 60 is configured to receive (box 152) an IEGM signal from one of the EP laboratory subsystems 11. The mapping module 72 (or any other suitable module) operating on processing circuitry 60 is configured to apply a trained artificial neural network 68 (box 154) to the received IEGM signal to provide an indication of the local activation time of the received IEGM signal. The output of artificial neural network 68 may include a vector having components (e.g., 400 components), each component having a floating value (e.g., a decimal value) between, for example, 0 and 1. The floating value represents the probability that the corresponding vector component is the LAT value of the input IEGM signal. Therefore, the highest floating value is associated with the highest probability and thus indicates the LAT value that should be used for the input IEGM signal. The mapping module 72 operating on processing circuitry 60 may be configured to receive the output of artificial neural network 68, find the highest floating value of the vector component output by artificial neural network 68, and calculate the LAT value of the received IEGM signal in response to the location of the vector component with the highest floating value.

[0094] Now for reference Figure 9 The image is from Figure 1 The system 10 presents a schematic diagram of the electroanatomical mapping 160. See also: Figure 8 The mapping module 72 (or any other suitable module) operating on the processing circuit 60 is optionally configured to generate (box 156) an electroanatomical mapping map 160 in response to an indication of the local activation time and other similar EP data provided in the step of block 154. In some embodiments, the processing circuit 60 is configured to provide the electroanatomical mapping map 160 to the EP laboratory subsystem 11, which provides an IEGM signal for generating the electroanatomical mapping map 160. The mapping module 72 is optionally configured to provide (box 158) the LAT found in the step of block 154 to the EP laboratory subsystem 11, which provides an IEGM signal for the LAT it found.

[0095] As used herein, the term “about” or “approximately” for any numerical value or range indicates appropriate dimensional tolerances that allow a collection of parts or components to achieve the intended purpose as described herein. More specifically, “about” or “approximately” may refer to a range of values ​​±20% of the listed values; for example, “about 90%” may refer to a range of values ​​from 72% to 108%.

[0096] For clarity, the various features of the invention described in the context of individual embodiments may also be provided in combination in a single embodiment. Conversely, for brevity, the various features of the invention described in the context of individual embodiments may also be provided individually or in any suitable sub-combination.

[0097] The above embodiments are cited by way of example, and the invention is not limited to the specific examples shown and described above. Rather, the scope of the invention includes combinations and sub-combinations of the various features described above, as well as variations and modifications thereof, which will occur to those skilled in the art upon reading the above description and are not disclosed in the prior art.

Claims

1. A method for finding the local activation time of an intracardiac electrogram (IEGM) signal, comprising: Receive the first IEGM signal from the electrophysiology laboratory subsystem and the corresponding local activation time annotation of the first IEGM signal manually annotated by the relevant annotator; The artificial neural network is trained in response to the first IEGM signal and the corresponding local activation time annotation to find the local activation time of the IEGM signal; Receive the second IEGM signal; and The trained artificial neural network is applied to the received second IEGM signal to provide an indication of the local activation time of the received second IEGM signal. The weights of annotations performed by the corresponding annotator are calculated in response to the local activation time annotation experience level of the corresponding annotator among the annotators, wherein the training includes training the artificial neural network in response to the first IEGM signal and the corresponding local activation time annotation to find the local activation time of the IEGM signal, the corresponding local activation time annotation being weighted according to a corresponding weight among the calculated weights of the corresponding annotator among the annotators who annotated the corresponding local activation time annotation.

2. The method of claim 1, further comprising searching a database of published scientific literature in response to a corresponding number of search matches indicating the local activation time annotation experience level of the corresponding annotator among the annotators, and wherein the calculation includes calculating the weight of the annotation performed by the corresponding annotator among the annotators in response to the corresponding number of search matches of the corresponding annotator among the annotators.

3. The method of claim 2, wherein the search is limited to searching scientific literature publications that describe local activation time annotations.

4. The method of claim 2, wherein the corresponding number of search matches is the corresponding number of scientific publications that match the corresponding annotators among the annotators.

5. The method of claim 2, wherein the training comprises: The first IEGM signal is input into the artificial neural network; as well as The parameters of the artificial neural network are iteratively adjusted in response to the output of the artificial neural network and the local activation time annotation of the first IEGM signal.

6. The method of claim 5, further comprising minimizing a loss function, the loss function being a function of the output of the artificial neural network and the local activation time annotations of the first IEGM signal weighted according to corresponding weights in the calculated weights, wherein the iterative adjustment is performed in response to minimizing the loss function.

7. The method of claim 6, wherein the loss function comprises a binary cross-entropy loss function.

8. The method of claim 1, further comprising generating an electroanatomical mapping in response to the indication of the local activation time.

9. A system for finding the local activation time of an intracardiac electrogram (IEGM) signal, comprising a remote server, the remote server including processing circuitry configured to: Receive the first IEGM signal from the electrophysiology laboratory subsystem and the corresponding local activation time annotation of the first IEGM signal manually annotated by the relevant annotator; The artificial neural network is trained in response to the first IEGM signal and the corresponding local activation time annotation to find the local activation time of the IEGM signal; Receive the second IEGM signal; and The trained artificial neural network is applied to the received second IEGM signal to provide an indication of the local activation time of the received second IEGM signal. The processing circuit is configured as follows: The weight of the annotation performed by the corresponding annotationer among the annotationers is calculated in response to the local activation time annotation experience level of the corresponding annotationer among the annotationers; and The artificial neural network is trained in response to the first IEGM signal and the corresponding local activation time annotation to find the local activation time of the IEGM signal, wherein the corresponding local activation time annotation is weighted according to the corresponding weight calculated by the annotator of the corresponding local activation time annotation in the local activation time annotation.

10. The system of claim 9, further comprising the electrophysiology laboratory subsystem, each electrophysiology laboratory subsystem comprising: A catheter configured to be inserted into at least one heart chamber of at least one living subject, and to capture a corresponding IEGM signal from the first IEGM signal from the at least one heart chamber; monitor; as well as Processing circuit, the processing circuit being configured to: Present the corresponding IEGM signal from the first IEGM signal to the display; Receive the corresponding local activation time annotation in the local activation time annotation of the displayed first IEGM signal, which is manually annotated by the corresponding annotator among the annotators; and The remote server is provided with the corresponding IEGM signal from the first IEGM signal and the corresponding local activation time annotation from the local activation time annotation.

11. The system of claim 9, wherein the processing circuit is configured to: In response to the respective annotator among the annotators, a corresponding number of search matches are generated indicating the level of local activation time annotation experience of the respective annotator among the annotators to search the database of published scientific literature; and The weight of the annotation performed by the corresponding annotationer is calculated in response to the corresponding number of search matches of the corresponding annotationer among the annotationers.

12. The system of claim 11, wherein the processing circuitry is configured to limit the search of the database to scientific publications describing local activation time annotations.

13. The system of claim 11, wherein the corresponding number of search matches is the corresponding number of scientific publications that match the corresponding annotators among the annotators.

14. The system of claim 11, wherein the processing circuit is configured to: The first IEGM signal is input into the artificial neural network; and The parameters of the artificial neural network are iteratively adjusted in response to the output of the artificial neural network and the local activation time annotation of the first IEGM signal.

15. The system of claim 14, wherein the processing circuitry is configured to: To minimize the loss function, which is a function of the output of the artificial neural network and the local activation time annotations of the first IEGM signal weighted according to the corresponding weights in the calculated weights; and The parameters of the artificial neural network are iteratively adjusted in response to minimizing the loss function.

16. The system of claim 15, wherein the loss function comprises a binary cross-entropy loss function.

17. The system of claim 9, wherein the processing circuitry is configured to generate an electroanatomical mapping in response to the indication of the local activation time.

18. A software product comprising a non-transitory computer-readable medium storing program instructions that, when read by a central processing unit (CPU), cause the CPU to: Receive the first IEGM signal from the electrophysiology laboratory subsystem and the corresponding local activation time annotation of the first IEGM signal manually annotated by the relevant annotator; The weight of the annotation performed by the corresponding annotator among the annotators is calculated in response to the local activation time annotation experience level of the corresponding annotator among the annotators; The artificial neural network is trained in response to the first IEGM signal and the corresponding local activation time annotation to find the local activation time of the IEGM signal, wherein the corresponding local activation time annotation is weighted according to the corresponding weight calculated by the annotator of the corresponding local activation time annotation in the local activation time annotation. Receive the second IEGM signal; and The trained artificial neural network is applied to the received second IEGM signal to provide an indication of the local activation time of the received second IEGM signal.

Images