阅读说明：本技术 基于ecg的心壁厚度估计 (ECG-based heart wall thickness estimation ) 是由 L·措雷夫 S·奥尔巴赫 M·阿米特 Y·A·阿莫斯 A·沙吉 于 2020-09-22 设计创作，主要内容包括：本发明题为“基于ECG的心壁厚度估计”。本发明公开了一种系统,所述系统包括接口和处理器。所述接口被配置为接收在患者的心脏中执行的多个电生理(EP)测量结果。所述处理器被配置为基于所述EP测量结果来估计位于所述心脏的指定位置处的壁厚。(The invention is entitled "ECG-based heart wall thickness estimation". A system includes an interface and a processor. The interface is configured to receive a plurality of Electrophysiological (EP) measurements performed in a heart of a patient. The processor is configured to estimate a wall thickness at a specified location of the heart based on the EP measurements.)

1. A system for estimating a property of heart wall tissue, the system comprising:

an interface configured to receive a plurality of Electrophysiological (EP) measurements performed in a heart of a patient; and

a processor configured to estimate a wall thickness at a specified location of the heart based on the EP measurements.

2. The system according to claim 1, wherein one or more of the EP measurements comprise an intracardiac Electrogram (EGM).

3. The system according to claim 2, wherein the EP measurements further include respective locations in the heart at which the EGMs were acquired.

4. The system of claim 1, wherein one or more of the EP measurements comprise a body surface Electrocardiogram (ECG).

5. The system of claim 1, wherein the processor is configured to estimate the wall thickness using a model defined by the EP measurements and refine the model based on results of an ablation procedure applied at the specified location of the heart.

6. The system of claim 5, wherein the model is a trained Machine Learning (ML) model.

7. The system of claim 6, wherein the ML model comprises at least one type of auto-encoder comprising an encoder coupled to a decoder.

8. The system of claim 7, wherein the at least one auto-encoder comprises a first auto-encoder configured to operate on the EGM and a second auto-encoder configured to operate on the ECG.

9. The system of claim 5, wherein the results of the ablation procedure include one or more of: (i) a temperature rise associated with the ablation procedure, and (ii) a change in tissue impedance associated with the ablation procedure.

10. A method for estimating a property of heart wall tissue, the method comprising:

receiving a plurality of Electrophysiological (EP) measurements performed in a heart of a patient; and

estimating a wall thickness at a specified location of the heart based on the EP measurements.

11. The method according to claim 10, wherein one or more of the EP measurements comprise an intracardiac Electrogram (EGM).

12. The method of claim 11, wherein the EP measurements further comprise respective locations in the heart at which the EGMs were acquired.

13. The method of claim 10, wherein one or more of the EP measurements comprise a body surface Electrocardiogram (ECG).

14. The method of claim 10, wherein estimating the wall thickness comprises: using a model defined by the EP measurements, and refining the model based on results of an ablation procedure applied at the specified location of the heart.

15. The method of claim 14, wherein the model is a trained Machine Learning (ML) model.

16. The method of claim 15, wherein the ML model comprises at least one type of auto-encoder comprising an encoder coupled to a decoder.

17. The method of claim 16, wherein the at least one auto-encoder comprises a first auto-encoder configured to operate on the EGM and a second auto-encoder configured to operate on the ECG.

18. The method of claim 14, wherein the results of the ablation procedure include one or more of: (i) a temperature rise associated with the ablation procedure, and (ii) a change in tissue impedance associated with the ablation procedure.

Technical Field

The present invention relates generally to the processing of electrophysiological signals and ablation, and in particular to estimating characteristics of heart wall tissue using Machine Learning (ML).

Background

Various methods may be used to estimate the thickness of the heart wall, such as ultrasound, fluoroscopy, and MRI imaging. The estimated wall thickness may be further correlated with the electrophysical signal to estimate damage to the heart wall tissue. For example, Takeshi Sasaki et al describe significant Associations between left Ventricular wall thickness, Post-Infarct scar thickness and intramural scar locations seen in MRI and Local endocardial electrogram bipolar/unipolar voltages, durations and electroanatomical deviations in "Myocardial Structural Associations with Local Electrograms: A Study of Post-resistant ventt ventural TACHMARD physiology and Magnetic Resonance Based Non-Invasive Mapping" (Circulation Arrhytmia and Electrophysiology, 12 months 2012; 5 th edition, volume 6: pages 1081 to 1090).

Disclosure of Invention

Embodiments of the invention described below provide a system that includes an interface and a processor. The interface is configured to receive a plurality of Electrophysiological (EP) measurements performed in a heart of a patient. The processor is configured to estimate a wall thickness at a specified location of the heart based on the EP measurements.

In some embodiments, one or more of the EP measurements comprise an intra-cardiac Electrogram (EGM).

In some embodiments, the EP measurements further include respective locations in the heart at which the EGMs were acquired.

In one embodiment, one or more of the EP measurements comprise a body surface Electrocardiogram (ECG).

In some embodiments, the processor is configured to estimate the wall thickness using a model defined by the EP measurements, and refine the model based on results of an ablation procedure applied at the specified location of the heart.

In another embodiment, the model is a trained Machine Learning (ML) model. In yet another embodiment, the ML model includes at least one type of auto-encoder, the auto-encoder including an encoder coupled to a decoder.

In one embodiment, the at least one auto-encoder includes a first auto-encoder configured to operate on the EGM and a second auto-encoder configured to operate on the ECG.

In some embodiments, the results of the ablation procedure include one or more of: (i) a temperature rise associated with the ablation procedure, and (ii) a change in tissue impedance associated with the ablation procedure.

There is also provided, in accordance with another embodiment of the present invention, a method including receiving a plurality of Electrophysiological (EP) measurements performed in a heart of a patient. Estimating a wall thickness at a specified location of the heart based on the EP measurements.

Drawings

The invention will be more fully understood from the following detailed description of embodiments of the invention taken together with the accompanying drawings, in which:

fig. 1 is a schematic illustration of a catheter-based Electrophysiological (EP) sensing, signal analysis, and IRE ablation system according to an exemplary embodiment of the present invention;

FIG. 2 is a flow diagram of the training and use of inference of a machine learning model for estimating heart wall thickness according to an exemplary embodiment of the present invention;

FIG. 3 illustrates a depth learning algorithm based on an automatic encoder and wall thickness estimation of fully-connected layers according to an exemplary embodiment of the present invention; and is

Fig. 4 is a schematic diagram of an auto-encoder architecture used in the deep learning algorithm of fig. 3, according to an exemplary embodiment of the present invention.

Detailed Description

SUMMARY

Cardiac ablation is a common procedure for treating cardiac arrhythmias by forming lesions in the cardiac tissue of a patient. Such lesions may be formed by irreversible electroporation (IRE) or other types of ablative energy such as Radio Frequency (RF), both of which may be applied using a catheter. In IRE ablation, the catheter is manipulated such that an electrode disposed on the distal end of the catheter is in contact with or in close proximity to the tissue. A high voltage bipolar pulse is then applied between the electrodes and the strong electric field pulse generated in the tissue results in cell death and lesion production. In RF ablation, an alternating RF current is applied to the tissue through one or more electrodes, causing cell death by heat.

To be effective, tissue ablation must be transmural, i.e., penetrate into the depth of the tissue. However, "over ablation" can cause undesirable damage to tissue (including few heart wall perforations) or damage to adjacent structures that may be behind the heart tissue, such as the esophagus. Therefore, it is important to be able to assess heart wall thickness (e.g., atrial or ventricular walls) just before and/or during ablation in order to use the optimal ablation parameters during the procedure.

Different imaging modalities may be employed to assess heart wall thickness, including Magnetic Resonance Imaging (MRI), Computed Tomography (CT), ultrasound, and the like. However, the use of these modalities adds cost and complexity to the procedure. Furthermore, the spatial resolution of these modalities may be on the order of the tissue thickness, which may yield a less accurate estimate of the actual wall thickness during ablation.

Embodiments of the present invention described below provide systems and methods for estimating the thickness of a heart wall (i.e., an atrial wall or a ventricular wall) using currently limited information just prior to and/or during ablation. In some embodiments, a Machine Learning (ML) model such as an Artificial Neural Network (ANN) is provided to allow this estimation using only EP data such as a multi-channel (e.g., 12-lead) surface Electrocardiogram (ECG) and intracardiac Electrogram (EGM) acquired by the electrodes of the catheter just prior to or during the ablation procedure.

In some embodiments, the ML model is trained using data including EP data (multichannel ECG and EGM, the latter also known as intracardiac ECG (icecg)), patient medical history, and 3D location information of the collected data. The model is optimized to obtain predictive capability of wall thickness via training using base live data, such as atrial/ventricular wall thickness assessed by an imaging modality such as ultrasound, CT, MRI, or similar imaging modalities.

The training data may also include (e.g., in combination with) the above data items, data collected after ablation initiation, and additional initial ablation data, such as a temperature rise profile and/or impedance changes during ablation. (in ablations that typically require between four (4) and sixty (60) seconds, the temperature rise profile can be detected very quickly, typically 10 or 100 milliseconds).

Just before and/or during the new ablation procedure (i.e., during the inference), the model uses only the EP data, including ECG and EGMS, and subsequent ablation data (i.e., any subsequent data acquired during the ablation procedure) for the particular patient as described above to further assess the patient's wall thickness.

Although an ANN model is used as an example herein, one skilled in the art can select from other ML models available, such as decision tree learning, Support Vector Machines (SVMs), and bayesian networks. ANN models include, for example, convolutional NN (CNN), autoencoder, and Probabilistic Neural Network (PNN). Typically, the processor or processors used (hereinafter collectively referred to as "processors") are programmed in software containing specific algorithms that enable the processor to perform each of the processor-related steps and functions listed above. Typically, training is performed using a computing system that includes multiple processors, such as a Graphics Processing Unit (GPU) or Tensor Processing Unit (TPU). However, any of these processors may also be a Central Processing Unit (CPU).

The ability to assess the heart wall thickness just before ablation as well as during ablation (i.e., in real-time), based on at least a portion of the data mentioned above using the ML algorithm, allows for a simple assessment of the heart wall thickness and may result in a more accurate ablation time and, in general, also in improved results of the ablation procedure.

Description of the System

Fig. 1 is a schematic illustration of a catheter-based Electrophysiological (EP) sensing, signal analysis, and IRE ablation system 20 according to an exemplary embodiment of the present invention. System 20 may be, for example, manufactured by Biosense-Webster corporationAnd 3, system. As shown, system 20 includes a catheter 21 having a shaft 22 that is navigated by a physician 30 into a heart 26 (inset 25) of a patient 28. In the illustrated example, the physician 30 inserts the shaft 22 through the sheath 23 while manipulating the shaft 22 with a manipulator 32 near the proximal end of the catheter.

In the embodiments described herein, the catheter 21 may be used for any suitable diagnostic purpose and/or tissue ablation, such as electrophysiological mapping and IRE ablation of the heart 26, respectively. ECG recorder 35 may receive various types of ECG signals sensed by system 20 during the procedure.

As shown in the inset 25, the distal end of the shaft 22 of the catheter 21 is provided with a multi-electrode basket catheter 40. Inset 45 shows the arrangement of the plurality of electrodes 48 of basket catheter 40. The proximal end of catheter 21 is connected to console 24 for transmission of an electrogram, for example, taken by electrodes 48.

Console 24 includes a processor 41 (typically a general purpose computer) having suitable front end and interface circuitry 38 for receiving EP signals (e.g., ECG and EGM) and non-EP signals (such as position signals) from electrodes 48 of catheter 21. To this end, processor 41 is connected to electrode 48 via a wire extending within shaft 22. The interface circuit 38 is also configured to receive ECG signals, such as from a 12 lead ECG device, which may be an ECG recorder 35, and non-ECG signals from the body surface electrodes 49. Typically, the electrodes 49 are attached to the skin around the chest and legs of the patient 28. Processor 41 is connected to electrode 49 by wires extending through cable 39 to receive signals from electrode 49.

Four of the body surface electrodes 49 are named according to standard ECG protocols: RA (right arm), LA (left arm), RL (right leg), and LL (left leg). The Wilson Central Terminal (WCT) may be formed by three of the four named body surface electrodes 49 and the resulting ECG signal V_WCTReceived by the interface circuit 38.

During the EP mapping procedure, the position of the electrodes 48 is tracked while the electrodes are within the patient's heart 26. Such tracking may be performed using an active current position (ACL) system manufactured by Biosense-Webster corporation, which is described in U.S. patent No. 8,456,182, the disclosure of which is incorporated herein by reference.

Thus, the processor may associate any given signal (such as an EGM) received from the electrode 48 with the location at which the signal was acquired. Processor 41 uses the information contained in these signals to construct an EP map, such as a Local Activation Time (LAT) map, for presentation on a display. In the illustrated embodiment, processor 41 estimates heart wall thickness using an algorithm including an ML algorithm applied to EP and other data, as described in fig. 2 and 3.

To perform IRE ablation, electrodes 48 are connected (e.g., switched) to an IRE pulse generator 47 in console 24 that includes processor-controlled switching circuitry (e.g., a relay array, not shown). Using the wall thickness information, processor 41 or the physician may select the electrodes connected to pulse generator 47 to apply the IRE pulses (via the switching circuitry).

During RF ablation, the initial ablation data and the subsequent ablation data include at least one of: IRE energy profile, temperature rise and impedance change. They can be used to further assess (e.g., in real time) the wall thickness of the patient, as depicted in fig. 2.

The processor 41 is typically programmed in software to perform the functions described herein. For example, the software may be downloaded to the processors in electronic form, over a network, or alternatively or in addition to, the software may be provided and/or stored on non-transitory tangible media, such as magnetic, optical, or electronic memory. In particular, processor 41 runs a dedicated algorithm, included in fig. 3, as disclosed herein, which enables processor 41 to perform the steps disclosed herein, as further described below.

ECG-based heart wall thickness estimation using ML

FIG. 2 is a flow diagram of the training and use of inference of a machine learning model for estimating heart wall thickness according to an exemplary embodiment of the present invention.

According to the exemplary embodiment presented, the algorithm is divided into two parts: algorithm preparation 101 and algorithm usage 102.

The algorithm prepares to perform a process starting from the ML modeling step 70 to generate an ML algorithm for estimating the heart wall thickness.

Next, at an ML algorithm training step 72, the processor trains the algorithm (e.g., ANN and preprocessing portion) using a database including ECG and EGM. In step 72, the processor trains the ML model using training data (including the underlying live data). The training data is formed by:

1. multi-channel (e.g., 12-lead) ECG data

2. Electrograms with 3D information of cardiac tissue collection locations

3. Anatomical location of each intracardiac electrode-based on another ML model/3D segmentation of the atrium using MFAM.

4. Diagnostic catheter details

5. Patient demographic information (e.g., gender, age, height, weight)

6. History of patients

The base live data is formed from:

7. atrial/ventricular wall thickness assessed by an imaging modality such as ultrasound, CT, MRI, or similar imaging modalities.

The further training data may also comprise ablation delivery energy profiles and ablation related parameters such as temperature rise and/or impedance change and/or elasticity change and/or stiffness change during ablation. The above data items #1 to #6 and/or the additionally collected training data collected after when ablation begins are collectively referred to herein as "ablation data".

At a trained model storage step 74, the algorithm preparation ends by storing the trained model in a non-transitory computer readable medium, such as a keyboard (memory stick). In an alternative embodiment, the model is sent in advance and its optimization parameters (such as weights for the ANN) are sent separately after training.

The algorithm uses 102 to execute a process that begins at algorithm upload step 76, during which the user uploads the entire ML model or its optimization parameters (e.g., weights) to the processor. Next, at a patient data receiving step 78, a processor, such as processor 28, receives patient inferred data, such as the aforementioned ECG and EGM, from electrodes 49 and 48, respectively.

Next, using the trained ML model for inference, the processor inputs data from the selected patient to the trained model and implements an algorithm on the model such that the model is able to output the patient's atrial wall thickness or ventricular wall thickness from only the limited data available (such as the aforementioned EP data) at a heart wall thickness estimation step 80. After installation on the processor, the trained models may be used for multiple patients.

In some embodiments, the NN model outputs a statistical distribution of thicknesses, and the peaks of this distribution, i.e., beyond those included in the NN model, may be selected in a subsequent step to determine the most likely wall thickness values.

The exemplary flow chart shown in fig. 2 was chosen solely for conceptual clarity. This embodiment may also include additional steps of the algorithm, such as receiving an indication of the degree of physical contact of the electrodes with the tissue being diagnosed. This and other possible steps have been purposely omitted from the disclosure herein in order to provide a more simplified flow diagram.

ML algorithm description

FIG. 3 illustrates a deep learning algorithm 300 based on an automatic encoder and wall thickness estimation of fully-connected layers, according to an exemplary embodiment of the invention. The method includes providing a deep learning supervised framework for estimation, and using electrograms, 12-lead ECGs, anatomical data, catheter details, patient demographics, and patient medical history (corresponding to training data items #1 through #6 above). The method may also use temperature rise and/or impedance change during ablation.

In this method, two automated encoders 302 and 304 (described in more detail below) are applied to perform dimensionality reduction on a set of features from a 12-lead ECG and/or from an intracardiac ECG. The method uses a fully-connected layer that is based on those characteristics and based on medical history information, including but not limited to NYHA (new york heart association) score, CHA2DS2-VASc score, and AF duration, as well as demographic data (e.g., age, gender, height, and weight). Regression analysis is then performed to estimate the thickness of the heart wall.

As described above, the method uses two auto-encoders 302 and 304. The automatic encoder comprises two components: an encoder and a decoder. The encoder maps the input (in fig. 3 the ECG signal and/or the EGM signal) to the hidden representation (h or u, respectively) via a non-linear transformation. The decoder then maps the hidden representation back to the reconstructed data via another non-linear transformation. Equations 1 and 2 represent the mapping:

equation 1 h ═ f (ECG, θ)_{Encoder for encoding a video signal})，ECG'＝g(h，θ_Decoder)，

Equation 2

Wherein theta is_{Encoder for encoding a video signal}、θ_DecoderFor the weights used in the reconstruction of the ECG signal,are weights used for EGM signal reconstruction.

The same network architecture is used for ECG and EGM reconstruction so that the nonlinear functions f and g are substantially the same. Using a minimized L2 normalization function among a set of autoencoders may provide a set of theta for ECG signal reconstruction_{Encoder for encoding a video signal}、θ_DecoderWeights and a set for EGM reconstruction And (4) weighting.

Fig. 4 is a schematic diagram of an auto-encoder architecture used in the deep learning algorithm of fig. 3, according to an exemplary embodiment of the present invention. In particular, an automated encoder architecture is used to compress and learn the feature space of an electrogram and/or 12-lead ECG. Each auto-encoder is implemented using a fully-connected convolutional neural network (FCN) of encoders and decoders, and has a predefined number of layers, as shown. In the encoder, the EGM/ECG signals are reduced in size and encoded into low-dimensional features. The decoder attempts to reconstruct the output from the low-dimensional features. Embodiments of the present invention employ a rectifier linear unit (ReLU) as an excitation function for the hidden layer. In the FCN model, there is no excitation function for the output layer. In addition, each hidden layer was subjected to batch normalization.

The encoder comprises a series of layers, wherein each individual layer is made up of a convolutional layer, a batch normalization layer, and an excitation layer. The input layer is defined by the original signal having a size of 1024 × N, where N represents the number of input channels. Thus, for a 12-lead ECG and for intracardiac ECG signals, N-12 corresponds to the number of electrodes on the catheter acquisition signal. For example, N-20 is used for the PentaRay or Lasso catheter, and N-64 is used for the basket catheter of fig. 1. It will be appreciated that the catheters mentioned above are examples and that the scope of the invention includes any cardiac catheter.

A convolution process with 40 filters of size 16 x N and step size 2 is applied on the first layer. The next three convolutional layers all have 20 filters with size 16 × N and step size 2. The next layer then consists of 40 filters of size 16 × N and step size 2. The last layer has a filter with size 16 x 1 and step size 1. The down-sampling process is implemented using step 2. Through the encoding process, a 32 × N dimensional feature map is obtained. The signature also represents the compressed data and is 1/32 the size of the original data.

The decoder part of the auto-encoder is inversely symmetrical to the encoder. Here, the deconvolution layer continues to upsample the feature map in order to recover the structural details. As for the output layer, the final deconvolution layer with a filter size of 16N and step size of 1 produces the output signal.

Returning to fig. 3, the hidden representations h and u, patient history information (NYHA score, CHA2DS2-VASc score, AF duration, and duration AF duration), and patient demographics (age, gender, height, and weight) are used as a feature space (light grey circles in fig. 3) for a fully connected neural network with four hidden layers (dark grey circles). In some implementations, the feature space further includes at least one of a temperature rise and an impedance change input.

The output from the hidden layer is then inserted into output neurons that estimate the wall thickness of the heart.

The entire network is trained using a back propagation algorithm that attempts to minimize the L2 regularization function, as shown in equation 3.

Equation 3

WhereinIn order to be a function of the loss,_iTW represents the atrial wall thickness/ventricular wall thickness obtained from ultrasound or CT, MRI or similar imaging modality of subject i,_iTW' is the estimated heart wall thickness based on the proposed method,beta is the regularization parameter, which is the weight of the fully connected layer. In the disclosed embodiments, β is set to 0.01.

Performing a back propagation algorithm to make a loss function JAnd (4) minimizing.

In an exemplary embodiment of the invention, a set of parameters is learnedh、u、θ_{Encoder for encoding a video signal}、θ_Decoder、After the optimum value of (in the L2 regularization sense), a deep learning regression for the heart wall thickness is obtained.

The disclosed embodiments provide a specific number, such as the number of filters, by way of example only. Generally, such numbers can be modified. While the above description refers to an ablation procedure, and to measuring the tissue wall thickness of the procedure, it should be understood that the description may be applied mutatis mutandis to measuring the tissue wall thickness in the absence of an ablation procedure. Accordingly, the scope of the present invention includes cardiac surgery with or without an ablation procedure.

Thus, it will be appreciated that operating the above algorithm enables the processor to approximate the thickness of the heart wall. This value may be incorporated into the GUI of the ablation system. Alternatively or additionally, the thickness values may be presented numerically on a "wall" map, or the wall thickness may be displayed graphically according to a scale of the cardiac image presented on the ablation system display.

It will thus be appreciated that the embodiments described above are cited by way of example, and that the present invention is not limited to what has been particularly shown and described hereinabove. Although the embodiments described herein are primarily directed to cardiac diagnostic applications, the methods and systems described herein may also be used in other cardiac medical applications where estimation of heart wall thickness is desired.

It will thus be appreciated that the embodiments described above are cited by way of example, and that the present invention is not limited to what has been particularly shown and described hereinabove. Rather, the scope of the present invention includes both combinations and subcombinations of the various features described hereinabove, as well as variations and modifications thereof which would occur to persons skilled in the art upon reading the foregoing description and which are not disclosed in the prior art. Documents incorporated by reference into this patent application are considered an integral part of the application, except that definitions in this specification should only be considered if any term defined in these incorporated documents conflicts with a definition explicitly or implicitly set forth in this specification.