Method and apparatus for ventricular tachycardia exit site identification
Patent Information
- Application Number
- CN202610367868.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2025-03-27
- Filing Date
- 2026-03-24
- Publication Date
- 2026-09-29
Smart Images
Figure CN122827697A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present teachings relate generally to treating ventricular tachycardia, and more particularly to identifying a ventricular tachycardia exit site location in a patient's heart. BACKGROUND
[0002] Ventricular tachycardia is a cardiac condition that can be fatal if left untreated. There are multiple treatment options, including antiarrhythmic drugs, invasive ablation procedures, and non-invasive radiotherapy.
[0003] As with catheter-based ablation and radiotherapy, treatment options can play a role in patients who have experienced ventricular tachycardia following a myocardial infarction. The latter is used to locally alter cardiac tissue to disable the responsible reentry circuit. Treatment success can highly depend on how accurately the responsible circuit is identified. In particular, it can be useful to locate an exit site location of a ventricular tachycardia reentry circuit in the heart.
[0004] One of the best existing methods to assess the functional electrophysiological behavior of cardiac tissue is electroanatomical mapping (EAM). Unfortunately, EAM presents its own set of challenges. For example, the measurement process itself is invasive, and obtaining high resolution maps is difficult and time consuming. BRIEF DESCRIPTION OF DRAWINGS
[0005] The methods and apparatuses for providing ventricular tachycardia exit site locations described in the detailed description below, at least partially address the above-mentioned needs and other needs, particularly in connection with the appended drawings, of which:
[0006] Figure 1 include block diagrams configured in accordance with various embodiments of these teachings;
[0007] Figure 2 include flow diagrams configured in accordance with various embodiments of these teachings; and
[0008] Figure 3 include schematic flow diagrams configured in accordance with various embodiments of these teachings.
[0009] The elements in the figures are illustrated for the purpose of example and clarity, and are not necessarily drawn to scale. For example, some of the elements in the figures can be exaggerated relative to other elements to help improve the understanding of various embodiments of the present teachings. Also, for the purpose of clarity, not all of the elements of a commercially feasible embodiment of the present teachings are necessarily included in a diagram. Certain acts and / or steps can be described or depicted in a particular, chronological sequence in various embodiments of the present teachings, and it is not necessary that a particular sequence be followed. The terminology used herein has its ordinary technical meaning as is accorded to such terminology in the technical field of the present teachings, unless otherwise defined here. The word "or" as used herein is intended to mean an inclusive "or" rather than an exclusive "or". That is, unless specifically indicated otherwise, the word "or" is used herein in the sense that it is used to provide either disjunctive or conjunctive meaning. DETAILED DESCRIPTION
[0010] In general, the various embodiments provide methods of non-invasively identifying a ventricular tachycardia exit location in a heart of a particular patient using electrocardiogram signals and medical images for the particular patient. With one method, a control circuit can be configured to access a neural network that has been trained using a plurality of synthetic data examples and input information about electrocardiogram signals and the aforementioned medical images for the particular patient to the neural network. The control circuit can then output from the neural network information identifying a predicted ventricular tachycardia exit location in the heart of the particular patient.
[0011] With one method, the aforementioned information about the electrocardiogram signals can include information about physical placement of electrocardiogram electrodes on the particular patient. Alternatively or in combination with the aforementioned approach, with one method, the electrocardiogram signals can include electrocardiogram signals from no more than twelve electrocardiogram leads (signals from more than twelve electrocardiogram leads can also be accommodated, if desired).
[0012] With one method, the aforementioned medical images can include three-dimensional images of at least a portion of the heart. For example, the medical images can include at least one computed tomography image of at least a portion of the heart and / or at least one magnetic resonance imaging image of at least a portion of the heart.
[0013] The present teachings are flexible in practice. As one example of these aspects, the aforementioned predicted ventricular tachycardia exit location in the heart of the particular patient can include one or more of: a particular region that includes a predetermined partition of the heart; a heat map that indicates at least one likelihood of the ventricular tachycardia exit location; and / or three-dimensional location coordinates.
[0014] The present teachings also allow the foregoing neural network to be trained using synthetic data examples. By one approach, at least some of the synthetic data examples can include electrogram information. By one approach, the foregoing inputting information about an electrocardiogram signal and a medical image for a particular patient to the neural network can further include inputting information about electrogram information for the particular patient to the neural network.
[0015] By one approach, at least some of the synthetic data examples can include examples each including: information about an electrocardiogram signal for a representative patient, a medical image for the representative patient, and at least one labeled ventricular tachycardia exit location. By one approach, at least some of these synthetic data examples can be generated using statistical shape modeling to increase variability in at least one of cardiac and torso anatomy instantiation, medical image instantiation, and / or labeled ventricular tachycardia exit location instantiation.
[0016] If desired, in addition to training the neural network using synthetic data examples, the training corpus can be expanded to include at least one non-synthetic data example.
[0017] By one approach, the present teachings can include a computer program that itself includes instructions that, when executed by a computer, cause the computer to perform: accessing a neural network that has been trained using a plurality of synthetic data examples, and inputting information about an electrocardiogram signal and the foregoing medical image for a particular patient to the neural network, with subsequent output from the neural network of information identifying a predicted ventricular tachycardia exit location in the particular patient's heart.
[0018] So configured, the present teachings provide for identifying ventricular tachycardia exit locations in a non-invasive manner, but with a precision that can rival electroanatomic mapping.
[0019] These and other benefits will become more apparent after a reading of the following detailed description in conjunction with the drawings. Reference is now made to the drawings, wherein: Figure 1 An illustrative device 100 compatible with many of the present teachings will first be presented.
[0020] In this particular example, the enabled device 100 includes a control circuit 101. As a "circuit," the control circuit 101 thus includes structure that includes at least one (and typically a plurality of) electrically conductive paths (such as paths constructed of an electrically conductive metal such as copper or silver) that conduct electricity in an orderly manner, which path(s) typically also include corresponding electrical components (both passive components such as resistors and capacitors and active components such as any of a variety of semiconductor-based devices) to allow the circuit to implement the control aspects of the present teachings.
[0021] Such control circuitry 101 can comprise a fixed-purpose, hardwired hardware platform (including, but not limited to, an application-specific integrated circuit (ASIC) (which is an integrated circuit customized for a particular use, rather than intended for general-purpose use), a field-programmable gate array (FPGA), and the like), or can comprise a partially or wholly programmable hardware platform (including, but not limited to, a microcontroller, a microprocessor, and the like). These architectural options for such structures are well-known and understood in the art, and need not be further described here. The control circuitry 101 is configured (e.g., through the use of corresponding programming, as would be readily understood by those skilled in the art) to perform one or more of the steps, actions, and / or functions described herein.
[0022] It will be appreciated that the control circuitry 101 can comprise a single integrated platform, or can comprise multiple such circuits that cooperate with one another.
[0023] The control circuitry 101 is operably coupled to a memory 102. The memory 102 can be integrated to the control circuitry 101, or can be physically separate (in whole or in part) from the control circuitry 101, as desired. The memory 102 can also be local (where, for example, both share a common circuit board, chassis, power supply, and / or housing) or partially or wholly remote with respect to the control circuitry 101 (where, for example, the memory 102 is physically located at another facility, city, or even country as compared to the control circuitry 101). As with the control circuitry 101, the memory 102 can comprise a single structure, or can comprise multiple memory platforms that collectively comprise the “memory” of the apparatus 100.
[0024] The memory 102 can be used, for example, to non-transitorily store computer instructions that, when executed by the control circuitry 101, cause the control circuitry 101 to function as described herein. (As used herein, a reference to “non-transitory” will be understood to refer to a non-temporal state of the contents being stored (and thus excludes the case that the stored contents merely constitute a signal or wave), as opposed to the volatility of the storage medium itself, and thus includes non-volatile memory such as read-only memory (ROM) as well as volatile memory such as dynamic random-access memory (DRAM).
[0025] The control circuitry 101 is also operably coupled to a user interface 103. The user interface 103 can comprise various user input mechanisms (such as, but not limited to, a keyboard and keypad, a cursor-controlled device, a touch-sensitive display, a voice-recognition interface, a gesture-recognition interface, and the like) and / or user output mechanisms (such as, but not limited to, a visual display, an audio transducer, a printer, and the like) to facilitate the receipt of information and / or instructions from and / or the provision of information to a user.
[0026] If desired, the control circuit 101 can also be operatively coupled to a network interface (not shown). So configured, the control circuit 101 can communicate with other elements (both internal and external to the apparatus 100) via the network interface. Network interfaces, including both wireless and non-wireless platforms, are well understood in the art and need not be elaborated upon here.
[0027] The control circuit 101 is also operatively coupled to, or itself is at least partially configured as, a neural network 104.
[0028] In this illustrative example, the control circuit 101 can be operatively coupled to an electrocardiogram platform 105 and one or more medical imaging systems 106.
[0029] The electrocardiogram platform 105 is used to deliver an electrocardiogram. The electrocardiogram platform is a non-invasive diagnostic tool used to record the electrical activity of the heart over a period of time. By placing electrodes on the skin of a patient 107 at specific points on the body, the platform captures the timing and intensity of the electrical impulses that pass through the heart with each heartbeat. The resulting graph, with its characteristic P-wave, QRS-wave, and T-wave, provides information about such things as the heart rhythm, the presence of any irregularities, and the condition of the heart muscle.
[0030] The aforementioned electrodes are connected to an electrocardiogram recording device. This device reports the electrical signals of multiple leads, each representing the electrical activity between a positive and negative electrode. There are twelve standard leads in a typical electrocardiogram platform: three bipolar limb leads (I, II, III), three augmented unipolar limb leads (aVR, aVL, aVF), and six unipolar chest leads (V1 through V6). Each lead provides unique information about the electrical axis of the heart, the heart rhythm, and any potential abnormalities such as arrhythmias, ischemia, or infarction.
[0031] In this illustrative example, the electrocardiogram platform 105 provides electrocardiogram information for the corresponding patient 107 to the aforementioned control circuit 101.
[0032] The aforementioned imaging systems 106 capture and provide medical images for a particular patient 107. Examples of such systems include computed tomography systems and magnetic resonance imaging systems. In general, for the sake of this illustrative example, the present description assumes that the medical images captured by the imaging systems 106 include at least a three-dimensional image of at least a portion of the patient’s heart. However, these teachings will also apply to medical images that capture the patient’s entire heart as well as a portion or all of the patient’s torso (and some or all of any electrocardiogram electrodes placed on the patient’s body).
[0033] Reference is now made to Figure 2A process 200 will be described, e.g., that can be performed in connection with the above-described application setup (more specifically via the aforementioned control circuit 101 and neural network 104). Generally speaking, this process 200 serves to non-invasively identify a ventricular tachycardia exit location in a patient's heart using electrocardiogram signals and medical images for that patient.
[0034] At optional block 201, the process 200 can generate one or more synthetic data examples. These synthetic data examples can later be used as part of (or all of) a training corpus for the above-described neural network 104.
[0035] By one approach, at least some of the synthetic data examples can be generated using statistical shape modeling to increase variability with respect to one or more represented features. Examples of this include, but are not limited to, variations with respect to electrocardiogram signal information, cardiac and torso anatomical instantiations, medical image instantiations, and labeled ventricular tachycardia exit location instantiations. Statistical shape modeling is a mathematical approach for analyzing and interpreting shape variability within a given class of objects. The approach can involve capturing geometric information for a collection of similar shapes (such as human organs) and creating a statistical model that can describe a typical shape (mean) and major variation patterns within the dataset.
[0036] Optional block 202 provides for training the neural network 104. Training material includes synthetic data examples 203, which can (or can not) be at least partially generated at block 201. The synthetic data examples 103 can include examples that each include information with respect to: information about electrocardiogram signals for a representative patient (where the representative patient can correspond to a real flesh-and-blood patient or a virtual patient), medical images for the representative patient, and at least one labeled ventricular tachycardia exit location. If desired, the synthetic data examples 103 can also include instantiations of physical placement of electrocardiogram electrodes on the representative patient's body.
[0037] If desired, these teachings also allow for use of at least one or more non-synthetic data examples 204 in the training corpus.
[0038] At block 205, the control circuit 101 accesses the trained neural network 104, which was trained at least in part using a plurality of synthetic data examples (e.g., as described above). At block 206, the control circuit 101 inputs information about electrocardiogram signals and medical images for a particular patient 107 to the neural network 104. This input information is generally true and relatively current / contemporaneous. In many cases, both modalities of information are captured on the same day that the information is input to the neural network 104.
[0039] Generally speaking, the information regarding electrocardiogram signals for the particular patient 107 can include information regarding the physical placement of electrocardiogram electrodes on the particular patient 107, and in many application settings, the electrocardiogram signals include electrocardiogram signals from no more than twelve electrocardiogram leads. In at least some application settings, it can be beneficial if the electrocardiogram signals happen to be from twelve electrocardiogram leads. These teachings are readily adapted, if desired, to application settings that utilize more than twelve electrocardiogram leads.
[0040] Likewise, generally speaking, the foregoing information regarding the patient medical images will include, at least in part, a three-dimensional image of at least a portion of the patient's heart, and can include at least one of: a computed tomography image of at least a portion of the heart and at least one magnetic resonance imaging image of at least a portion of the heart.
[0041] At block 207, the neural network 104 outputs information identifying a predicted ventricular tachycardia exit location in the heart of the particular patient 107. This identifying information can take any of a variety of forms, as desired. By one approach, the identifying information includes an image of a particular region, the particular region comprising a predetermined partition of the heart. By another approach, in the alternative to or in combination with the foregoing, the identifying information can include a heat map indicating at least one likelihood of a ventricular tachycardia exit location. (A heat map is a data visualization tool that typically uses a color gradient to represent various values of a metric of interest. Different colors on the heat map correspond to the magnitude of the observed metric, with certain colors indicating, for example, an increased likelihood that a particular region in the heart is a ventricular tachycardia exit location.) By yet another approach, again in the alternative to or in combination with the foregoing, the identifying information includes three-dimensional position coordinates specifying a particular three-dimensional location on the patient's heart.
[0042] Further details consistent with these teachings will now be presented. It should be appreciated that the specific details of these examples are intended to serve illustrative purposes only and are not intended to suggest any particular limitations regarding these teachings.
[0043] Figure 3 One approach is presented that uses 12-lead electrocardiogram traces and anatomical information derived from medical images to accurately and non-invasively estimate ventricular tachycardia exit locations. In this example, deep learning is used to fuse the 12-lead electrocardiogram traces and information about the patient's anatomy.
[0044] The 12-lead electrocardiogram traces 301 for the particular patient (in the form of printed output and / or digitized signals) are pre-processed at block 302 to remove, normalize, and / or mitigate baseline wander, power line interference, respiratory effects, etc.
[0045] At block 303, the pre-processed electrocardiogram information is processed to extract desired features. By one approach, the aforementioned features can include discriminative features specified by a particular user / clinic. By one approach, some or all of these features can be defined by literature and / or expert knowledge of cardiologists, and extracted using traditional electrocardiogram processing algorithms, such as peak detection or waveform tracing algorithms. Exemplary features can be amplitude of each of the 12 leads, QRS duration, QRS waveform, electrical axis, etc. (where the "QRS" complex will be understood to refer to a series of waves on an electrocardiogram representing ventricular depolarization). By another approach, particular features can be learned directly from the raw signal using deep learning methods, such as temporal convolutional networks or recurrent networks.
[0046] Three-dimensional cardiac and torso images 304 for the same patient and information 305 about placement of electrocardiogram electrodes on the patient are pre-processed at block 306. This pre-processing can include applying an automatic segmentation algorithm to identify particular portions of the heart or patient body. By one approach, the resulting segmentation mask can represent the biventricular heart cavities, the skin surface of the torso, and the electrodes if present in the three-dimensional medical images 304. By another approach, where electrocardiogram electrodes are not shown in the three-dimensional medical images 304, but capture by an RGB-D camera (which is a camera that typically provides both color and depth information) is available, the precise electrocardiogram electrode positions can be inferred by solving a registration problem between the depth map and the segmented torso surface. Then, the segmentation algorithm can extract the electrocardiogram electrodes from the RGB-D images and project the electrode positions onto the torso surface using the registration matrix. By yet another approach, the torso and electrocardiogram electrodes are given by an atlas model with standardized electrocardiogram placement that is registered to the segmented cardiac anatomy, which can facilitate application of the algorithm without three-dimensional medical images showing the complete torso and electrocardiogram electrode positions. And by yet another approach, the segmentation mask can be further pre-processed to extract a mesh. The heart can be represented by a dense mesh, such as a tetrahedral mesh. The torso can be represented by a triangulated surface, and the electrocardiogram electrodes can be integrated via an annotated set of vertices.
[0047] At block 307, the pre-processed image content has one or more features extracted therefrom. By one approach, when the anatomical information includes a segmentation mask, such as, for example, mesh structure data, a convolutional neural network can be used to extract the features. By another approach, when the data is given as a dense or surface mesh, a graph convolutional neural network can be used to extract the features.
[0048] The extracted features from block 303 and block 307 are input to a prediction head 308, such as the neural network described above. By one approach, the neural network comprises a fully connected neural network (sometimes also referred to as a multilayer perceptron neural network). By one approach, information fusion can be achieved by concatenating the aforementioned feature vectors before inputting them to the neural network. By another approach, information fusion can be achieved by conditioning the neural network on electrocardiogram features (e.g., by applying feature linear modulation). The neural network can then infer the ventricular tachycardia exit location. By one approach, the latter can be implemented as a classification task, in particular by classifying which region of a given ventricular tachycardia exit location originated in a predetermined partition of the heart. By another approach, the neural network can be adapted to infer a heat map reflecting the likelihood of the ventricular tachycardia exit location originating. And in yet another embodiment, the neural network can output an exact location, e.g., expressed as a three-dimensional coordinate in a local coordinate system.
[0049] By one approach, depending on the particular training utilized, the prediction head 308 outputs classification information 309 about whether the ventricular tachycardia exit location is within an American Heart Association ventricular segment of the patient’s heart. By another approach, the prediction head 308 outputs location information 310 about the ventricular tachycardia exit location, such as local coordinate information and / or a corresponding heat map.
[0050] The present teachings are highly flexible in practice and allow for various modifications and / or supplemental features.
[0051] As one example of these aspects, the aforementioned training can comprise joint training. By one approach, for example, the neural network can be trained on a large dataset of actual 12-lead electrocardiogram traces (printed output or digitized signals), three-dimensional medical images, torso imaging (three-dimensional medical images or RGB-D camera images), and ground truth labels (such as high-resolution electroanatomical maps) derived from electrophysiology data (where the true labels refer to clinical information in which experts labeled the exit site).
[0052] As another example of these aspects, when a large amount of real-world data is lacking, a cardiac computational model can be used to generate a large dataset of noise-free ground truth. To this end, statistical shape modeling can be used to increase variability of the cardiac and torso anatomy. Scar and border zone distribution can be added by utilizing available three-dimensional late gadolinium enhancement magnetic resonance imaging images or applying corresponding rule-based algorithms.
[0053] To reflect the imperfections of electrocardiogram electrode placement, the standard virtual electrocardiogram electrode placement on the torso model can be augmented by moving the placement pattern or randomly moving individual electrodes. Given the anatomical data, various electro-physiological cardiac activation sequences and associated virtual electrocardiograms can be generated by randomizing inhomogeneous tissue electrical conductivity and pacing locations, etc. Moreover, to enable robust feature extraction from the printed output of 12-lead electrocardiogram traces, these teachings allow rendering virtual electrocardiograms, e.g., using pre-defined templates that capture the variability of clinical printed outputs.
[0054] And as yet another example of these aspects, these teachings allow incorporating electrogram information (which can be available in some cases, e.g., from devices implanted in the patient) to improve the prediction accuracy provided by the neural network. As one example of these aspects, the electrogram information can be provided as a form of signal features on a subset of locations in the myocardium as input to the neural network. In the case where the myocardium is represented as a triangular or tetrahedral point mesh, the electrogram information can be provided as a set of features in the mesh points that represent the locations of measured electrode leads from an implanted device, such as, e.g., an implantable cardioverter defibrillator. One example of such features includes the time from pacing to sensed R-wave in each of the measured channels. The electrodes can be used for pacing, such as in a non-invasive program stimulation, in which case the features in the pacing leads would be zero values. Alternative definitions of input features based on measured electrograms can be the polarity of the QRS complex in each lead, the rising and falling slopes of the QRS complex in each lead, the R-wave peak time in each lead, etc. (note that these features can be available in standard electrocardiogram data as well as data from implanted devices). The neural network can be trained on a large database of training samples that are generated by cardiac electro-physiological computational models that include the presence of an implanted device, such as an implantable cardioverter defibrillator, that is capable of delivering pacing stimulation and sensing local electrical activity.
[0055] These teachings provide for the automatic determination of ventricular tachycardia exit locations, which can be crucial information for any local treatment of ventricular tachycardia affecting cardiac tissue, such as, e.g., radiofrequency ablation or radiation therapy. Since these teachings only require a 12-lead electrocardiogram trace (or fewer or more) and corresponding three-dimensional anatomical information derived from medical imaging methods, these teachings are completely non-invasive. In particular, these teachings allow integrating the information present in the 12-lead electrocardiogram trace printed output and specific electrocardiogram electrode placements derived, e.g., from RGB-D camera images, which allows easy integration into existing clinical settings. Moreover, by integrating the patient's anatomy into the estimation process, these teachings remove a large amount of uncertainty about the relative orientation of the heart, torso, electrocardiogram electrode directions. The latter can lead to more accurate estimates.
[0056] Other aspects of the present teachings are provided by the subject matter of the following clauses (wherein it should be understood that any one of these clauses can be combined with any one or more of the other clauses as desired).
[0057] Clause 1. A method to identify, non-invasively, a ventricular tachycardia exit location in a heart of a particular patient using electrocardiogram signals and medical images for the particular patient, the method comprising: by control circuitry: accessing a neural network that has been trained using a plurality of synthetic data examples; inputting information about the electrocardiogram signals and the medical images for the particular patient to the neural network; outputting, from the neural network, information identifying a predicted ventricular tachycardia exit location in the heart of the particular patient.
[0058] Clause 2. The method of clause 1, wherein the electrocardiogram signals comprise electrocardiogram signals from no more than twelve electrocardiogram leads.
[0059] Clause 3. The method of clause 1, wherein the information about the electrocardiogram signals comprises information about physical placement of electrocardiogram electrodes on the particular patient.
[0060] Clause 4. The method of clause 1, wherein the medical images comprise at least one of: at least one computed tomography image of at least a portion of the heart and at least one magnetic resonance imaging image of at least a portion of the heart.
[0061] Clause 5. The method of clause 1, wherein the medical images comprise three- dimensional images of at least a portion of the heart.
[0062] Clause 6. The method of clause 1, wherein the predicted ventricular tachycardia exit location in the heart of the particular patient comprises at least one of: a particular region comprising a predetermined partition; a heat map indicating at least one likelihood of a ventricular tachycardia exit location; and three-dimensional position coordinates.
[0063] Clause 7. The method of clause 1, wherein the synthetic data examples comprise examples each comprising: information about electrocardiogram signals for a representative patient; a medical image for the representative patient; and at least one labeled ventricular tachycardia exit location.
[0064] Clause 8. The method of clause 1, wherein the neural network has also been trained using at least one non-synthetic data example.
[0065] Clause 9. The method of clause 1, further comprising: generating at least some of the synthetic data examples using statistical shape modeling to increase variability in at least one of: cardiac and torso anatomic instantiations; medical image instantiations; and labeled ventricular tachycardia exit site location instantiations.
[0066] Clause 10. The method of clause 1, wherein at least some of the synthetic data examples include electrogram information, and wherein inputting information about the electrocardiogram signal and the medical image for the particular patient to the neural network further comprises inputting information about electrogram information for the particular patient to the neural network.
[0067] Clause 11. A device to non-invasively identify a ventricular tachycardia exit site location in a heart of a particular patient using an electrocardiogram signal and a medical image for the particular patient, the device comprising: control circuitry configured to: access a neural network that has been trained using a plurality of synthetic data examples; input information about the electrocardiogram signal and the medical image for the particular patient to the neural network; and output, from the neural network, information identifying a predicted ventricular tachycardia exit site location in the heart of the particular patient.
[0068] Clause 12. The device of clause 11, wherein the electrocardiogram signal comprises electrocardiogram signals from no more than twelve electrocardiogram leads.
[0069] Clause 13. The device of clause 11, wherein the information about the electrocardiogram signal comprises information about physical placement of electrocardiogram electrodes on the particular patient.
[0070] Clause 14. The device of clause 11, wherein the medical image comprises at least one of: at least one computed tomography image of at least a portion of the heart and at least one magnetic resonance imaging image of at least a portion of the heart.
[0071] Clause 15. The device of clause 11, wherein the medical image comprises a three- dimensional image of at least a portion of the heart.
[0072] Clause 16. The device of clause 11, wherein the predicted ventricular tachycardia exit site location in the heart of the particular patient comprises at least one of: a particular region comprising the predetermined partition; a heat map indicating at least one likelihood of a ventricular tachycardia exit site location; and three-dimensional position coordinates.
[0073] Clause 17. The apparatus of clause 11, wherein the synthetic data examples comprise examples each comprising: information about electrocardiogram signals for a representative patient; medical images for the representative patient; and at least one labeled ventricular tachycardia exit location.
[0074] Clause 18. The apparatus of clause 11, wherein the neural network has also been trained using at least one non-synthetic data example.
[0075] Clause 19. The apparatus of clause 11, wherein the control circuit is further configured to: generate at least some of the synthetic data examples using statistical shape modeling to increase variability in at least one of: cardiac and torso anatomical instantiations; medical image instantiations; and labeled ventricular tachycardia exit location instantiations.
[0076] Clause 20. The apparatus of clause 11, wherein at least some of the synthetic data examples comprise electrogram information, and wherein inputting information about the electrocardiogram signals and the medical images for the particular patient to the neural network further comprises inputting information about electrogram information for the particular patient to the neural network.
[0077] Clause 21. A non-transitory computer readable medium for non-invasively identifying a ventricular tachycardia exit location in a heart of a particular patient using electrocardiogram signals and medical images for the particular patient, the non-transitory computer readable medium comprising instructions stored thereon that, when executed on a processor (101), perform the following steps: accessing a neural network that has been trained using a plurality of synthetic data examples; inputting information about the electrocardiogram signals and the medical images for the particular patient to the neural network; outputting from the neural network information identifying a predicted ventricular tachycardia exit location in the heart of the particular patient.
[0078] Those skilled in the art will realize that various modifications, alterations and combinations can be made with regard to the above described embodiments without departing from the scope of the application, and that the above described embodiments are to be regarded as illustrative only and not restrictive.
Claims
1. A method (200) for non-invasively identifying the exit location of ventricular tachycardia in the heart of a specific patient (107) using electrocardiogram signals and medical images, the method comprising: Control circuit (101): Access (205) to the neural network (104) that has been trained (202) using multiple synthetic data examples (203); Information about the electrocardiogram signal and the medical image for the specific patient is input (206) into the neural network; Information from the neural network output (207) identifies the predicted ventricular tachycardia exit location in the heart of the particular patient.
2. An apparatus (100) for non-invasively identifying the exit location of ventricular tachycardia in the heart of a specific patient (107) using electrocardiogram signals and medical images, the apparatus comprising: Control circuit (101), the control circuit being configured to: Access (205) to the neural network (104) that has been trained (202) using multiple synthetic data examples (203); Information about the electrocardiogram signal and the medical image for the specific patient is input (206) into the neural network; as well as Information from the neural network output (207) identifies the predicted ventricular tachycardia exit location in the heart of the particular patient.
3. A nontransitory computer-readable medium (102) for non-invasively identifying the exit location of ventricular tachycardia in the heart of a particular patient (107) using electrocardiogram signals and medical images for that particular patient (107), the nontransitory computer-readable medium including instructions stored thereon that, when executed on a processor (101), perform the following steps: Access (205) to the neural network (104) that has been trained (202) using multiple synthetic data examples (203); Information about the electrocardiogram signal and the medical image for the specific patient is input (206) into the neural network; Information from the neural network output (207) identifies the predicted ventricular tachycardia exit location in the heart of the particular patient.
4. The method, apparatus or computer-readable medium according to any one of claims 1 to 3, wherein the electrocardiogram signal comprises electrocardiogram signals from no more than twelve electrocardiogram leads (301).
5. The method, apparatus, or computer-readable medium according to any one of claims 1 to 4, wherein the information regarding the electrocardiogram signal includes information regarding the physical placement of the electrocardiogram electrodes on the particular patient.
6. The method, apparatus, or computer-readable medium according to any one of claims 1 to 5, wherein the medical image comprises at least one of: at least one computed tomography image of at least a portion of the heart and at least one magnetic resonance imaging image of at least a portion of the heart.
7. The method, apparatus, or computer-readable medium according to any one of claims 1 to 6, wherein the medical image comprises a three-dimensional image (304) of at least a portion of the heart.
8. The method, apparatus, or computer-readable medium according to any one of claims 1 to 7, wherein the predicted ventricular tachycardia exit location in the heart of the particular patient comprises at least one of the following: A specific region including the predetermined partitions of the heart; A heat map indicating at least one possible exit location for ventricular tachycardia; and Three-dimensional position coordinates.
9. The method, apparatus, or computer-readable medium according to any one of claims 1 to 8, wherein the synthetic data examples include examples, each of which comprises: Information regarding electrocardiogram signals from representative patients; Medical images of the representative patients; as well as At least one marked exit location for ventricular tachycardia.
10. The method, apparatus or computer-readable medium according to any one of claims 1 to 9, wherein the neural network has been trained (202) using at least one non-synthetic data example (205).
11. The method, apparatus, or computer-readable medium according to any one of claims 1 to 10, further comprising: Statistical shape modeling is used to generate at least some of the synthetic data examples to increase the variability of at least one of the following: Instantiation of the heart and trunk anatomy; Medical image instantiation; and Instantiate the marked ventricular tachycardia exit location.
12. The method, apparatus, or computer-readable medium according to any one of claims 1 to 11, wherein at least some of the synthetic data examples include electrocardiogram information, and wherein inputting information about the electrocardiogram signal and the medical image for the particular patient into the neural network further includes inputting information about the electrocardiogram information for the particular patient into the neural network.