Program for predicting occurrence of steam pop

Abstract

To provide a program for outputting data indicative of prediction of an occurrence of steam pop due to ablation for arrhythmia treatment.SOLUTION: A program is configured for outputting data indicative of prediction of an occurrence of steam pop due to ablation for arrhythmia treatment in a subject. The program is generated by a method including the steps of: inputting, as training data, active indicators or passive indicators of ablation for arrhythmia treatment and a plurality of data representing that steam pop has not occurred even after a certain period or more of ablation for arrhythmia treatment, into an artificial intelligence model; and causing the artificial intelligence model to learn.SELECTED DRAWING: None

Description

[Technical field]

[0001] The present invention relates to a program for outputting data indicating a prediction of steam pop occurrence due to ablation for arrhythmia treatment, a computer storing the program, and a computer-readable recording medium storing the program. The present invention also relates to an apparatus for outputting data indicating a prediction of steam pop occurrence due to ablation for arrhythmia treatment in a subject. Furthermore, the present invention also relates to a method for outputting data indicating a prediction of steam pop occurrence due to ablation for arrhythmia treatment in a subject. Furthermore, the present invention also relates to a method for predicting steam pop occurrence due to ablation for arrhythmia treatment. [Background technology]

[0002] Radiofrequency catheter ablation (RFCA) is a medical technique that induces thermal damage by passing a high-frequency AC current of 350 kHz to 1 MHz through myocardial tissue to treat cardiac diseases such as arrhythmia (Non-Patent Document 1). During RFCA, steam popping, an audible sound caused by a steam explosion in myocardial tissue, can occur, which can cause ventricular septal defects and cardiac perforations, leading to sequelae and death (Non-Patent Document 2). The detailed cause of steam popping has not yet been elucidated. If the risk of steam popping could be quantitatively calculated, the safety of RFCA would be improved.

[0003] The steam pop phenomenon that occurs during radiofrequency catheter ablation for the treatment of arrhythmia is a serious event that can lead to serious complications such as cardiac tamponade. Active indices such as output, ablation time, and contact force, as well as passive indices such as local temperature and resistance at the ablation site, have been examined as indicators for predicting this serious event. However, it is difficult to accurately predict the steam pop phenomenon clinically using any of these indices alone, and it is also affected by the type of catheter and irrigation from the catheter tip.

[0004] Patent document 1 describes an invention relating to preventing steam pop during ablation, in which an apparatus is described for performing tissue ablation using a probe having a distal conductive cap, and describes that according to one aspect of the apparatus, analyzing the reflected optical radiation includes recognizing a characteristic signature in the reflected optical radiation that indicates an impending steam pop event.

[0005] Patent document 2 describes an apparatus for predicting a steam pop phenomenon during ablation, the apparatus comprising: an apparatus for performing a medical procedure, the apparatus comprising an invasive probe configured to be coupled to tissue within a patient's organ and to apply ablation energy to the tissue; and a processor coupled to the probe and configured to estimate a model of the evolution of vapor pressure within the tissue caused by the ablation energy as a function of time, to predict the time of occurrence of a steam pop phenomenon caused by the vapor pressure based on the model, and to indicate the predicted time of occurrence of the steam pop phenomenon to an operator.

[0006] Non-Patent Document 3 proposes Multivariate Time-series Anomaly Detection via Graph Attention Network (MATD-GAT) as a method for detecting anomalies in multivariate time-series data using a neural network. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] JP 2015-208684 A [Patent Document 2] JP 2015-009157 A [Non-patent literature]

[0008] [Non-Patent Document 1] DE Haines, “The biophysics of radiofrequency catheter ablation in the heart: The importance of temperature monitoring,” Pacing and Clinical Electrophysiology, vol. 16, no. 3, pp. 586-591, March 1993. [Non-Patent Document 2] J. Seiler, KC Roberts-Thomson, J.-M. Raymond, J. Vest, E. Delacretaz, and WG Stevenson, “Steam pops during irrigated radiofrequency ablation: Feasibility of impedance monitoring for prevention,” Heart Rhythm, vol. 5, no. 10, pp. 1411-1416, October 2008. [Non-Patent Document 3] H. Zhao, Y. Wang, J. Duan, C. Huang, D. Cao, Y. Tong, B. Xu, J. Bai, J. Tong, and Q. Zhang, “Multivariate time-series anomaly detection via graph attention network,” September 2020. Summary of the Invention [Problem to be solved by the invention]

[0009] Therefore, the present invention aims to build a real-time anomaly detection model that predicts the occurrence of steam pops using machine learning. That is, the present invention aims to provide a program for outputting data indicating a prediction of the occurrence of steam pops due to ablation for arrhythmia treatment; an apparatus for outputting data indicating a prediction of the occurrence of steam pops due to ablation for arrhythmia treatment in a subject; and a method for outputting data indicating a prediction of the occurrence of steam pops due to ablation for arrhythmia treatment in a subject. [Means for solving the problem]

[0010] The inventors recently reported that variables including both active and passive indices provide a better correlation with the size of the ablation lesion. Therefore, they hypothesized that more accurate prediction of the ablation effect and safety could be achieved by using real-time changes in all parameters (both active and passive indices) in the generator of the ablation system.

[0011] The generator of the ablation system records real-time changes in a large number of active and passive indicators per unit time. The time-series data on a large number of variables during electrical current were analyzed using machine learning to investigate the association with the occurrence of steam pop, a serious complication that occurs during ablation. As a result, the prediction algorithm for the occurrence of the steam pop phenomenon according to the present invention was significantly superior to predictions based on conventional parameters.

[0012] This algorithm, which uses machine learning from a large amount of time-series data recorded in the system's generator to accurately predict the results of ablation, is a new technology that has not been used in the field of catheter ablation to date. Although catheter treatment for arrhythmia may further advance in the future and energy other than radiofrequency may be used, it is possible to use the present invention to build an algorithm that predicts ablation complications from data similarly stored in the generator, and it is expected that the present invention will continue to be clinically applicable in the future.

[0013] That is, the present invention provides the following. [Aspect 1] A program for outputting data indicating a prediction of the occurrence of steam pops due to ablation for the treatment of arrhythmia in a subject, the program being generated by a method including a step of inputting active or passive indicators of ablation for the treatment of arrhythmia, and a plurality of data indicating that steam pops did not occur even after a certain period of cauterization by ablation for the treatment of arrhythmia, into an artificial intelligence model as learning data, and having the artificial intelligence model learn the data. [Aspect 2] The program according to embodiment 1, wherein the active or passive indicators of ablation for treating arrhythmia are selected from the group consisting of output, contact force, upper limit temperature, current application time, and saline concentration used for cooling. [Aspect 3] The program of aspect 2, wherein an active or passive indicator of ablation for treating arrhythmia comprises an output. [Aspect 4] A program as described in aspect 1, wherein the active or passive index of ablation for arrhythmia treatment is a variable representing the relative difference between a set upper limit value of the active or passive index and the active or passive index. [Aspect 5] A program as described in aspect 1, wherein an active or passive indicator of ablation for treating arrhythmia is a variable representing the relative difference between a set upper limit value of output and the output. [Aspect 6] 2. The program according to embodiment 1, wherein the certain period of time is a period of 30 seconds or more. [Aspect 7] 3. The program according to claim 2, wherein the energization time is 10 seconds or longer. [Aspect 8] 2. The program according to embodiment 1, wherein the plurality of data is 50 or more pieces of data. [Aspect 9] The program according to embodiment 1, wherein the algorithm used in the step of training the artificial intelligence model is MTAD-GAT. [Aspect 10] 2. The program according to embodiment 1, wherein in the step of training the artificial intelligence model, a plurality of data sets are used by dividing them into training: validation: test = 4-7: 1-3: 1-3. [Aspect 11] 2. The program according to claim 1, wherein in the step of training the artificial intelligence model, the plurality of data is multivariate time series data. [Aspect 12] The program according to embodiment 1, wherein in the step of training the artificial intelligence model, the plurality of data are standardized using the average and standard deviation of each feature within the plurality of data. [Aspect 13] The program according to embodiment 1, wherein the step of training the artificial intelligence model includes a step of processing the data into fixed-length data using a moving window. [Aspect 14] The program according to aspect 1, wherein the step of training the artificial intelligence model includes a step of convolving each feature of the plurality of data based on a specified kernel size. [Aspect 15] 15. The program according to embodiment 14, further comprising a step of filling data lost by the convolution step with zeros. [Aspect 16] 15. The program according to embodiment 14, wherein the convolution step is performed by taking a sum. [Aspect 17] The program of embodiment 1, wherein the step of training an artificial intelligence model incorporates multiple graph attention mechanisms. [Aspect 18] The program of embodiment 17, wherein the multiple graph attention mechanisms include graph attention for feature direction and graph attention for time direction. [Aspect 19] The program of embodiment 17, wherein the step of training the artificial intelligence model includes a step of combining the output of multiple graph attention mechanisms and the convolutional data and passing the combined data to a GRU layer. [Aspect 20] The program of aspect 1, wherein the step of training the artificial intelligence model includes a step of extracting features by passing them through an encoder based on a GRU layer and passing them to a decoder layer. [Aspect 21] The program of embodiment 1, wherein the step of training the artificial intelligence model includes a step of adding a self-attention layer and a fully connected layer to the GRU layer, weighting the output while returning it so as to restore the original input. [Aspect 22] The program according to embodiment 21, wherein the restored output extracts only the last column of the window. [Aspect 23] 2. The program according to claim 1, wherein the step of training the artificial intelligence model includes a step of calculating an anomaly degree at time stamp t, where i is each input feature, using the following formula:

number

[0014] According to the program, device, and method of the present invention, it is possible to output data indicating a prediction of the occurrence of steam pop due to ablation for the treatment of arrhythmia in a subject, and to predict the occurrence of steam pop due to ablation for the treatment of arrhythmia. [Brief description of the drawings]

[0015] [Figure 1] FIG. 1 is a diagram showing the proposed MTAD-GAT-based algorithm used in an embodiment of the present invention. [Diagram 2] FIG. 2 is a diagram showing preprocessing performed in an embodiment of the present invention. [Diagram 3] FIG. 3 is a diagram showing 1-D convolution performed in an embodiment of the present invention. [Figure 4] FIG. 4 is a diagram showing a data restoration method in a reconstruction-based model (decoder) performed in an embodiment of the present invention. [Diagram 5] FIG. 5 is a diagram showing an abnormality score of abnormal data in an embodiment of the present invention. [Figure 6] FIG. 6 is a diagram showing anomaly scores of Diff_Power in the embodiment of the present invention. [Figure 7] FIG. 7 is a diagram showing an abnormality score of abnormal data in an embodiment of the present invention. [Figure 8]FIG. 8 is a schematic diagram of a computer that may be used to implement an embodiment of the present invention. [Figure 9] FIG. 9 is a flow chart showing an example of a processing procedure performed in the device and method of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0016] program: The present invention provides a program for outputting data indicating a prediction of the occurrence of steam pops due to ablation for the treatment of arrhythmia in a subject, the program being generated by a method including a step of inputting active or passive indicators of ablation for the treatment of arrhythmia and a plurality of data indicating that steam pops did not occur even after ablation for the treatment of arrhythmia for a certain period of time or more into an artificial intelligence model as learning data, and having the artificial intelligence model learn the program.

[0017] Examples of ablation for treating arrhythmia include radiofrequency catheter ablation for treating arrhythmia as well as ablation by electroporation for treating arrhythmia, with radiofrequency catheter ablation for treating arrhythmia being preferred.

[0018] The data indicating the predicted occurrence of steam pops due to ablation for the treatment of arrhythmia may be numerical data indicating the predicted occurrence of steam pops due to ablation for the treatment of arrhythmia, or it may be symbolic data indicating the predicted occurrence of steam pops due to ablation for the treatment of arrhythmia.

[0019] The data indicative of an active or passive indicator of ablation for treating arrhythmia may be a feature vector generated based on the data indicative of an active or passive indicator of ablation for treating arrhythmia.

[0020] The plurality of data indicating active or passive indicators of ablation for arrhythmia treatment and no steam pops occurring even after a certain period of ablation for arrhythmia treatment is preferably 50 or more data, more preferably 60 or more data, more preferably 70 or more data, more preferably 80 or more data, more preferably 100 or more data, more preferably 1,000 or more data, and more preferably 10,000 or more data. Also, the plurality of data is preferably 1,000,000 or less data, more preferably 100,000 or less data, more preferably 10,000 or less data, and more preferably 1,000 or less data.

[0021] A plurality of data indicating active or passive indicators of ablation for arrhythmia treatment and the fact that steam pops did not occur even after ablation for arrhythmia treatment for a certain period of time or more means, for example, a plurality of data indicating, as a set, active or passive indicators when ablation for arrhythmia treatment is performed and the fact that steam pops did not occur even after ablation for arrhythmia treatment for a certain period of time or more.

[0022] The plurality of data indicating active or passive indicators of ablation for treating arrhythmia and the fact that no steam pop occurred even after ablation for treating arrhythmia for a certain period of time or more is preferably an active indicator of ablation for treating arrhythmia, a passive indicator of ablation for treating arrhythmia, and the fact that no steam pop occurred even after ablation for treating arrhythmia for a certain period of time or more.

[0023] A plurality of data indicating active indicators of ablation for treating arrhythmia, passive indicators of ablation for treating arrhythmia, and the fact that steam pops did not occur even when ablation for treating arrhythmia was performed for a certain period of time or more means, for example, a plurality of data indicating, as a set, active indicators when ablation for treating arrhythmia was performed, passive indicators when ablation for treating arrhythmia was performed, and the fact that steam pops did not occur even when ablation for treating arrhythmia was performed for a certain period of time or more.

[0024] The active index of ablation for arrhythmia treatment is preferably one, two, three, four, or five selected from the group consisting of ablation output, contact force, ablation time, catheter contact direction, and type of ablation reflux liquid. Here, the number is the number selected from the categories of ablation output, contact force, ablation time, catheter contact direction, and type of ablation reflux liquid. Therefore, for example, in cases where the number selected from the ablation output is two or more, the number is not limited to these.

[0025] The ablation output is preferably one or two selected from the group consisting of a time integral value of the output and an average value of the output. Here, the ablation output being selected from the group consisting of a time integral value of the output and an average value of the output includes the ablation output being a time integral value of the output and an average value of the output.

[0026] The contact force is preferably one, two, or three selected from the group consisting of the average value of the force pressing the catheter against the target, the maximum value of the force pressing the catheter against the target, and the minimum value of the force pressing the catheter against the target. Here, the number is the number selected from the categories of the average value of the force pressing the catheter against the target, the maximum value of the force pressing the catheter against the target, and the minimum value of the force pressing the catheter against the target. Therefore, for example, in cases where the number selected from the average value of the force pressing the catheter against the target is two or more, the number is not limited to these numbers.

[0027] The catheter contact direction is, for example, one or two selected from the group consisting of a vertical direction and a parallel direction. Here, the catheter contact direction being selected from the group consisting of a vertical direction and a parallel direction includes the catheter contact direction being a vertical direction and a parallel direction.

[0028] The type of ablation perfusion fluid is, for example, one or two selected from saline and semi-saline. Here, the type of ablation perfusion fluid selected from saline and semi-saline includes the type of ablation perfusion fluid being saline and semi-saline.

[0029] The passive indicators of ablation for arrhythmia treatment are preferably one or two selected from the group consisting of the local temperature of the ablation area and the local resistance value. Here, the number is the number selected from the categories of the local temperature of the ablation area and the local resistance value. Therefore, for example, in cases where the number selected from the local temperature of the ablation area is two or more, the number is not limited to these.

[0030] The local temperature of the cauterization area is preferably one or two selected from the group consisting of an average temperature and a maximum temperature. Here, the local temperature of the cauterization area being selected from the group consisting of an average temperature and a maximum temperature includes the local temperature of the cauterization area being the average temperature and the maximum temperature.

[0031] The local resistance value is preferably one, two, three, or four selected from the group consisting of the maximum resistance value, the minimum resistance value, the range of resistance value drop, and the rate of resistance value drop. Here, the number is the number selected from the categories of the maximum resistance value, the minimum resistance value, the range of resistance value drop, and the rate of resistance value drop. Therefore, for example, in cases where the number selected from the maximum resistance value is two or more, there is no restriction to these numbers.

[0032] The active or passive indicators of ablation for arrhythmia treatment are preferably one, two, three, four, or five indicators selected from the group consisting of output, contact force, upper limit temperature, current application time, and saline solution concentration used for cooling. Here, the number is the number selected from the categories of output, contact force, upper limit temperature, current application time, and saline solution concentration used for cooling. Therefore, for example, in cases where the number selected from output is two or more, the number is not limited to these. The active or passive indicators of ablation for arrhythmia treatment preferably include output.

[0033] The energization time is preferably 5 seconds or more, more preferably 10 seconds or more, even more preferably 15 seconds or more, and even more preferably 20 seconds or more. The active or passive indicators of ablation for arrhythmia treatment and the multiple data showing that steam pops did not occur even after ablation for arrhythmia treatment for a certain period of time or more are preferably less than 10% of the data with an energization time of less than 5, 10, 15 or 20 seconds, more preferably less than 5% of the data with an energization time of less than 5, 10, 15 or 20 seconds, even more preferably less than 1% of the data with an energization time of less than 5, 10, 15 or 20 seconds, and even more preferably less than 0.1% of the data with an energization time of less than 5, 10, 15 or 20 seconds. The multiple data are preferably substantially free of data with an energization time of less than 5 seconds, more preferably substantially free of data with an energization time of less than 10 seconds, even more preferably substantially free of data with an energization time of less than 15 seconds, and even more preferably substantially free of data with an energization time of less than 20 seconds.

[0034] The active index or passive index of ablation for arrhythmia treatment is preferably a variable representing a relative difference between a set upper limit value of the active index or passive index and the active index or passive index.The active index or passive index of ablation for arrhythmia treatment is preferably a variable representing a relative difference between a set upper limit value of output and the output.

[0035] In multiple data showing that no steam pops occurred even when ablation was performed for a certain period of time or longer by arrhythmia treatment, the certain period of time is preferably selected from a period of 30 seconds or more, more preferably selected from a period of 30 seconds to 3 minutes, even more preferably selected from a period of 40 seconds to 2 minutes, and even more preferably selected from a period of 50 seconds to 90 seconds.

[0036] In the program of the present invention, the multiple data input to the artificial intelligence model as learning data preferably contains 10% or less of data indicating that a steam pop has occurred, more preferably 5% or less, even more preferably 1% or less, even more preferably 0.1% or less, and even more preferably 0.01% or less. In the program of the present invention, the multiple data input to the artificial intelligence model as learning data preferably does not substantially contain data indicating that a steam pop has occurred, and more preferably does not contain data indicating that a steam pop has occurred.

[0037] In the program of the present invention, the multiple data input to the artificial intelligence model as learning data preferably includes data indicating that steam pops did not occur with cauterization for less than the certain period of time of 10% or less, more preferably 5% or less, even more preferably 1% or less, even more preferably 0.1% or less, and even more preferably 0.01% or less. In the program of the present invention, the multiple data input to the artificial intelligence model as learning data preferably does not substantially include data indicating that steam pops did not occur with cauterization for less than the certain period of time of the above, and more preferably does not include data indicating that steam pops did not occur with cauterization for less than the certain period of time of the above.

[0038] The algorithm used in the step of training the artificial intelligence model is preferably MTAD-GAT. One embodiment of the algorithm of the program of the present invention is shown below together with one embodiment of the data used as the experimental material.

[0039] A Experimental Materials (Data): The data used in this embodiment is ablation data from an ex-vivo experiment obtained using a Tacti-Flex catheter (Abbott Inc., USA). This data was recorded with a sampling period of 0.01 seconds, and the variables shown in Table 1 are obtained as those that can be used for machine learning. [Table 1]

[0040] In addition, when cauterizing, the upper limit of Power (Target Power), the upper limit of Contact Force (Target CF), and the upper limit of Temp (Target Temp) are set in advance. In order to perform cauterization so as not to exceed these upper limits, Diff Power, Diff CF (Diff Contact Force), and Diff Temp, which are expressed by the following formulas, are used as variables that represent the relative difference from the upper limits. Diff Power =Target Power - Power Diff CF = Target CF - CF Diff Temp = Target Temp - Temp

[0041] The following values ​​are set as parameters for cauterization. 1. Power: 30, 40, and 50 W 2. Contact force: 10 and 20 g 3. Upper-temperature limits: 40, 43, and 45°C 4. Energization time: 10, 20, 30, 60, 120, and 180 seconds 5. Irrigation settings: normal and half saline

[0042] In this embodiment, learning is performed only for current conduction times of 20 seconds or more. The reason for this is that the power reaches its upper limit 2 seconds after the start of cauterization, and this period is unsteady, there are cases where the upper limit is not reached in 2 seconds and takes about 5 seconds, and the impedance and temperature of the myocardium to be cauterized change drastically in the first few seconds, making it unsuitable for both learning and test data, and it is assumed that the first few seconds do not affect steam popping.

[0043] In this embodiment, 314 normal data in which no steam pop occurred and 87 data in which steam pop occurred are used.

[0044] B method: In this embodiment, the occurrence of a steam pop itself is not defined as an abnormality, but rather it is assumed that some kind of precursor occurs before the steam pop occurs, and this precursor is regarded as an abnormality and detected. In other words, the timing at which the steam pop occurs itself cannot be known in real time, and it is determined whether the model operated correctly by checking whether a steam pop occurs after the fact.

[0045] The steam pop occurrence prediction model constructed in this embodiment predicts the occurrence of a steam pop by detecting the signs, but because this model is premised on real-time operation, and because, as mentioned above, there is less data on the occurrence of steam pops and the data available for learning is imbalanced, it is necessary to use an unsupervised learning algorithm that uses time-series data as input and detects anomalies.

[0046] Therefore, in this embodiment, we use the new algorithm shown below based on Multivariate Time-series Anomaly Detection via Graph Attention Network (MATD-GAT), which has been proposed as a method for detecting anomalies in time-series multivariate data using a neural network.

[0047] Fig. 1 shows an overview of the algorithm. This algorithm learns only normal data in which no steam pop occurred. Therefore, the normal data is divided into learning:verification:test=6:2:2 for use. Note that the division of the normal data may be, for example, learning:verification:test=4-7:1-3:1-3, etc.

[0048] In Figure 1, Input Data is multivariate time series data that has been standardized in advance using the mean and standard deviation of each feature in the training data.

[0049] 1 Preprocessing: In preprocessing (circle 1 in Figure 1), the data is processed into fixed-length data using a moving window as shown in Figure 2.

[0050] 2. 1-D Convolution: In the 1-D Convolution shown in circle 2 in Figure 1, each feature is convolved based on the specified kernel size as shown in Figure 3. When the convolution is performed as shown in Figure 3, the time from t0 to t n-1 The values ​​up to are missing, but this algorithm fills them with 0. Also, convolution within the kernel size is performed by taking the sum.

[0051] 3 Two graph attention mechanisms: Regarding circle 3 in Figure 1, this algorithm incorporates two types of graph attention mechanisms. The difference between these is whether graph attention is implemented in the feature direction or in the time direction for the two-dimensional data of [Window Size, Feature Size] created in Figure 2, and it is possible to capture correlations between features and temporal dependencies. Since the cause of the anomalies that contribute to the occurrence of steam pops is completely unknown, we focus on this mechanism so that we can search for anomalies from two perspectives like this. The two outputs of the graph attention mechanism are combined with the post-convolution data and passed to the next layer.

[0052] Encoder based on 4 GRU layers: Regarding circle 4 in Figure 1, the GRU layer is a mechanism that can learn and memorize the temporal dependencies of time-series data. By passing it through the encoder, features (hidden states) are extracted and passed to the next decoder layer.

[0053] 5 Reconstruction-based model (decoder): Regarding circle 5 in Figure 1, by adding a self-attention layer and a fully connected layer to the GRU layer, it determines which extracted features are important and assigns weights to them, while returning an output that restores the original input (the data after preprocessing in circle 1 in Figure 1). In this embodiment, the data is created using a moving window, so the restored data will overlap. Therefore, only the last column of the window is extracted from the restored data, as shown in FIG.

[0054] 6 Calculation of Anomaly Score: For circle 6 in Figure 1, the anomaly degree at timestamp t is calculated as follows:

number

[0055] Here, i represents each input feature, and the root mean square error between the input and output is defined as the degree of anomaly. During learning, this degree of anomaly is used as an index to train the model so that the degree of anomaly decreases. The validation data is input at the same time as the training data, and is not used for learning. However, the validation data is used during learning to calculate the degree of anomaly, which is used to monitor the model output for unknown data and prevent overlearning.

[0056] Next, the logic for determining whether or not there is an abnormality from the calculated abnormality level will be described. First, the abnormality level is calculated using verification data. The 95th percentile value of the calculated abnormality level is obtained and set as a threshold value. If m seconds (n>m) out of the past n seconds exceed the abnormality threshold, it is determined that there is a sign of steam pop. Here, the 95th percentile value may be a percentile value of a predetermined numerical value, and the predetermined numerical value is preferably a numerical value selected from numerical values ​​included in the range of 80 to 100, more preferably a numerical value selected from numerical values ​​included in the range of 85 to 100, and even more preferably a numerical value selected from numerical values ​​included in the range of 90 to 100.

[0057] The present invention provides a computer having the above program stored thereon, and a computer-readable recording medium having the above program stored thereon.

[0058] Device: The present invention provides an apparatus for outputting data indicative of a prediction of occurrence of steam pop due to ablation for treating arrhythmia in a subject, the apparatus comprising: means for inputting data indicative of active or passive indicators of ablation for treating arrhythmia in a subject; A means for calculating data indicating an active indicator or a passive indicator of ablation for arrhythmia treatment in a subject using the input data and the program, the data indicating a prediction of the occurrence of steam pop due to ablation for arrhythmia treatment; and means for outputting data indicative of a predicted occurrence of steam pop due to ablation for treating arrhythmia in the subject. An apparatus is provided.

[0059] Examples of ablation for treating arrhythmia include radiofrequency catheter ablation for treating arrhythmia as well as ablation by electroporation method for treating arrhythmia, with radiofrequency catheter ablation for treating arrhythmia being preferred.

[0060] The data indicating the predicted occurrence of steam pops due to ablation for the treatment of arrhythmia in a subject may be numerical data indicating the predicted occurrence of steam pops due to ablation for the treatment of arrhythmia, or it may be symbolic data indicating the predicted occurrence of steam pops due to ablation for the treatment of arrhythmia.

[0061] The means for inputting data indicative of an active index or a passive index of ablation for arrhythmia treatment in the subject may be, for example, a means for inputting data indicative of an active index or a passive index of ablation for arrhythmia treatment in the subject. Note that the data indicative of an active index or a passive index of ablation for arrhythmia treatment in the subject may be a feature vector generated based on the data indicative of an active index or a passive index of ablation for arrhythmia treatment in the subject.

[0062] Means for inputting data indicative of active or passive indicators of ablation to treat arrhythmia in a subject include, for example, a keyboard, a mouse, a sensor in a catheter, and the like.

[0063] Means for calculating data indicating a prediction of the occurrence of steam pop due to ablation for treating arrhythmia in a subject using input data indicating active or passive indicators of ablation for treating arrhythmia in a subject and the above-mentioned program include, for example, a central processing unit (CPU) connected to a keyboard, mouse, sensor in the catheter, etc. for inputting the above-mentioned data, connected to a hard disk, flash memory, etc. that stores the above-mentioned program, and connected to memory (storage means) such as ROM and RAM.

[0064] Examples of the means for outputting the calculated data indicating the predicted occurrence of steam pops due to ablation for arrhythmia treatment in a subject include a monitor, a printer, etc. Also, examples of the means for outputting the calculated data indicating the predicted occurrence of steam pops due to ablation for arrhythmia treatment in a subject include a means for storing the calculated data indicating the predicted occurrence of steam pops due to ablation for arrhythmia treatment in a storage means such as a hard disk, a flash memory, a ROM, or a RAM.

[0065] The device of the present invention may also include a means for stopping an operation such as ablation by the device of the present invention, such as a means for turning off the power supply for an operation such as ablation by the device of the present invention, such as a catheter, based on data indicating a prediction of the occurrence of steam pop due to ablation for treating arrhythmia in a subject, stored in a storage means such as a hard disk, flash memory, ROM, RAM, etc.

[0066] The device of the present invention may include a means for storing the program. Examples of the means for storing the program include a hard disk and a flash memory. However, it is not essential that the device of the present invention includes the means for storing the program. For example, the device of the present invention may not include the means for storing the program, but may be connected to a device including the means for storing the program via a communication line. In other words, the device of the present invention may be a part of a system obtained by connecting to a device including the means for storing the program via a communication line.

[0067] The device of the present invention may include a means for determining whether to continue ablation based on the output data indicating the predicted occurrence of steam pop due to ablation for arrhythmia treatment in the subject. For example, the means may include a CPU connected to a hard disk, flash memory, etc. that stores a program including conditions for determining whether to continue ablation, and connected to a memory such as a ROM or RAM.

[0068] The device of the present invention may include a means for inputting data indicating active or passive indices of ablation for arrhythmia treatment in the subject up to that point in time based on a decision to continue ablation by the means for deciding whether to continue ablation. Examples of such means include a CPU connected to a hard disk, flash memory, or the like that stores a program including conditions for deciding whether to continue ablation, and connected to a hard disk, flash memory, sensor in a catheter, or the like that stores or inputs data indicating active or passive indices of ablation for arrhythmia treatment in the subject up to that point in time, and connected to memories such as ROM and RAM.

[0069] Fig. 8 is a schematic diagram showing an exemplary embodiment of the device of the present invention. In Fig. 8, reference numeral 100 denotes a computer, which includes a control unit 101, a storage unit 102, a peripheral device I / F unit 103, an input unit 104, a display unit 105, and a communication unit 106, which are connected by a bus 110. Note that this configuration is merely an example, and various configurations can be adopted as appropriate.

[0070] The control unit 101 is composed of a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), etc. The CPU calls up and executes programs stored in the storage unit 102, the ROM, a recording medium, etc., into a work memory area on the RAM, drives and controls each device connected via the bus 110, and realizes the processing performed by the computer. The ROM is a non-volatile memory, and holds programs such as the boot program and BIOS of the computer 100, data, etc. The RAM is a volatile memory, and temporarily holds programs, data, etc. loaded from the storage unit 102, the ROM, a recording medium, etc., and has a work area used when the control unit 101 performs various processing. The storage unit 102 is, for example, a HDD (Hard Disk Drive), and stores the programs executed by the control unit 101 and various other data.

[0071] The peripheral device I / F (interface) unit 103 is a port for connecting the computer 100 to a peripheral device. The peripheral device I / F unit 103 is configured with USB, IEEE1394, RS-232C, etc. The connection with the peripheral device may be wired or wireless. The input unit 104 has input devices such as a keyboard, a pointing device such as a mouse, and a numeric keypad, and performs operation instructions, operation instructions, data input, etc. to the computer 100. The display unit 105 is a logic circuit or a device driver for displaying videos, images, etc. on a display device such as a liquid crystal panel. The input unit 104 and the display unit 105 can also be configured integrally as a touch display.

[0072] The communication unit 106 has a communication control device, a communication port, etc., and is a wired or wireless communication interface that mediates communication with the network 120. The bus 110 is a communication path that mediates the transmission and reception of control signals, data signals, etc. between the devices. The network 120 may further be connected to an external server 130 and a net storage 140.

[0073] FIG. 9 is a flowchart showing an example of a processing procedure performed in the device of the present invention. In an example of the processing procedure performed in the device of the present invention, after the start of the processing procedure, time-series ablation data up to the present time is input. Then, a feature vector is generated based on the input time-series ablation data up to the present time. The generated feature vector is input to a trained model, and the presence or absence of occurrence of steam pop due to ablation for arrhythmia treatment is predicted. Whether or not to continue ablation is determined based on the predicted presence or absence of occurrence of steam pop. If it is determined that ablation is to be continued, time-series ablation data up to the present time is input again, and the same processing procedure as before is repeated. If it is determined that ablation is not to be continued, the processing procedure is terminated.

[0074] method: The present invention provides a method for outputting data indicative of a prediction of occurrence of steam pop due to ablation for treating arrhythmia in a subject, the method comprising: inputting data into the device indicative of active or passive indicators of ablation for treating arrhythmia in a subject; A step of causing the device to calculate data indicating a prediction of the occurrence of steam pop due to ablation for arrhythmia treatment in the subject, using the input data indicating active or passive indicators of ablation for arrhythmia treatment in the subject and the program; and outputting the calculated data indicating a prediction of occurrence of steam pop due to ablation for treating arrhythmia in the subject to the device. A method is provided.

[0075] Examples of ablation for treating arrhythmia include radiofrequency catheter ablation for treating arrhythmia as well as ablation by electroporation method for treating arrhythmia, with radiofrequency catheter ablation for treating arrhythmia being preferred.

[0076] The data indicating the predicted occurrence of steam pops due to ablation for the treatment of arrhythmia in a subject may be numerical data indicating the predicted occurrence of steam pops due to ablation for the treatment of arrhythmia, or it may be symbolic data indicating the predicted occurrence of steam pops due to ablation for the treatment of arrhythmia.

[0077] The step of inputting data indicative of active or passive indices of ablation for treating arrhythmia in a subject into the device is performed using, for example, a keyboard, a mouse, a sensor in a catheter, etc. Note that the data indicative of active or passive indices of ablation for treating arrhythmia in a subject may be a feature vector generated based on data indicative of active or passive indices of ablation for treating arrhythmia in a subject.

[0078] The step of inputting data indicative of active or passive indicators of ablation for treating arrhythmia in a subject into the device may, for example, be a step of inputting data indicative of active indicators of ablation for treating arrhythmia in a subject and data indicative of passive indicators of ablation for treating arrhythmia in a subject into the device.

[0079] The process of having the device calculate data indicating a prediction of the occurrence of steam pop due to ablation for treating arrhythmia in a subject using input data indicating active or passive indicators of ablation for treating arrhythmia in a subject and the program is carried out by, for example, having a CPU connected to a keyboard, mouse, sensor in the catheter, etc. for inputting the data, connected to a hard disk, flash memory, etc. that stores the program, and connected to memory such as ROM and RAM, perform calculations in accordance with the data and the program.

[0080] The process of having the device output data indicating the calculated prediction of the occurrence of steam pop due to ablation for the treatment of arrhythmia in a subject is performed, for example, using a monitor, printer, hard disk, flash memory, ROM, RAM, etc.

[0081] The method of the present invention may include, for example, a step of stopping an operation such as ablation by a device of the present invention, such as a step of turning off a power source for an operation such as ablation by a device of the present invention, such as a catheter, based on calculated data indicating a prediction of the occurrence of steam pop in a subject due to ablation for treating arrhythmia.

[0082] The method of the present invention may include a step of determining whether to continue ablation based on the output data indicating a prediction of occurrence of steam pop due to ablation for arrhythmia treatment in the subject. This step is performed, for example, by using a CPU connected to a hard disk, flash memory, etc. that stores a program including conditions for determining whether to continue ablation, and connected to a memory such as a ROM or RAM.

[0083] The method of the present invention may include a step of inputting data indicating active or passive indices of ablation for arrhythmia treatment in the subject up to that point in time based on the determination to continue ablation in the step of determining whether to continue ablation. This step is performed, for example, by using a CPU or the like connected to a hard disk, flash memory, or the like storing a program including conditions for determining whether to continue ablation, and connected to a hard disk, flash memory, sensor in a catheter, or the like storing or inputting data indicating active or passive indices of ablation for arrhythmia treatment in the subject up to that point in time, and connected to a memory such as a ROM or RAM.

[0084] The method of the present invention may further include a step of acquiring data indicating active or passive indicators of ablation for arrhythmia treatment in the subject before the step of inputting the data indicating active or passive indicators of ablation for arrhythmia treatment in the subject into the device. Here, the step of acquiring data indicating active or passive indicators of ablation for arrhythmia treatment in the subject may be, for example, a step of acquiring data indicating active indicators of ablation for arrhythmia treatment and data indicating passive indicators of ablation for arrhythmia treatment. The step of acquiring these data may be, for example, performed by identifying active or passive indicators expected when ablation for arrhythmia treatment is performed in the subject.

[0085] FIG. 9 is a flowchart showing an example of a processing procedure performed in the method of the present invention. In an example of the processing procedure performed in the method of the present invention, after the start of the processing procedure, time-series ablation data up to the present time is input. Next, a feature vector is generated based on the input time-series ablation data up to the present time. The generated feature vector is input to a trained model, and the presence or absence of occurrence of steam pop due to ablation for arrhythmia treatment is predicted. Whether or not to continue ablation is determined based on the predicted presence or absence of occurrence of steam pop. If it is determined that ablation is to be continued, time-series ablation data up to the present time is input again, and the same processing procedure as before is repeated. If it is determined that ablation is not to be continued, the processing procedure is terminated.

[0086] The present invention provides a method for predicting the occurrence of steam pops due to ablation for arrhythmia treatment, including the above-mentioned method. When predicting the occurrence of steam pops due to ablation for arrhythmia treatment, the prediction of the occurrence of steam pops due to ablation for arrhythmia treatment may be based only on the meaning of the data indicating the prediction of the occurrence of steam pops due to ablation for arrhythmia treatment in a subject output by the method of the present invention, or the prediction of the occurrence of steam pops due to ablation for arrhythmia treatment may be based on the meaning of the data indicating the prediction of the occurrence of steam pops due to ablation for arrhythmia treatment in a subject output by the method of the present invention, further taking into account other information. EXAMPLES

[0087] Example 1: In this example, a porcine myocardial ablation experiment was performed using a TactiFlex catheter (Abbot Inc. USA). The TactiFlex catheter collected time-series data at a sampling rate of 0.01 seconds during ablation. In the time-series data of this example, the occurrence of steam pops could not be visually confirmed, but the specific location of the abnormality that caused the steam pop was not clear. If abnormal data that causes steam pops can be detected in real time during ablation, the ablation process can be stopped before steam pops occur. Ablation data that does not cause steam pops even after ablation for 65 seconds or more was defined as normal data. The experimental data collected in this example consisted of 91 normal data without steam pops and 80 abnormal data with steam pops.

[0088] In this example, we adopted MTAD-GAT, which is a proposed algorithm for identifying abnormal patterns in multivariate time series data based on neural networks. This methodology has the ability to detect not only correlations between variables but also interdependencies over time.

[0089] As a result, of the 80 abnormal steam pop data, 43 were predicted before occurrence. The anomaly scores of the abnormal data are shown in Figure 5. In Figure 5, the colored areas represent the occurrence of anomalies, and the horizontal line at the anomaly score of 0.04 represents the threshold for detecting anomalies. Furthermore, no erroneous abnormalities were identified in the normal data. Therefore, the performance of this model was 54% sensitivity and 100% specificity. Data in which no steam pop occurred even after cauterization for 65 seconds or more was defined as normal data, but when all data, including data in which no steam pop occurred after cauterization for 65 seconds or less, were used for learning, the model was unable to predict the occurrence of steam pops. Therefore, it was shown that the steam pop prediction model using only data in which no steam pop occurred even after cauterization for a certain period of time is effective.

[0090] Example 2: In this example, experimental data was collected from an in vitro experiment in which radiofrequency ablation was applied to porcine myocardium. The conditions for the ablation experiment were as follows: 1. Power: 30, 40, and 50 W 2. Contact force: 10 and 20 g 3. Upper-temperature limits: 40, 43, and 45°C 4. Energization time: 10, 20, 30, 60, 120, and 180 seconds 5. Irrigation settings: normal and half saline We collected 87 data points with steam pop and 314 data points without steam pop, the latter of which were used for machine learning.

[0091] It is believed that an anomaly occurred during the cauterization process, causing the steam pop. Therefore, the following algorithm based on MTAD-GAT was used to detect anomalies during the cauterization process. The anomaly detection process is as follows: First, the data was scaled. Second, the trained model was used to calculate the anomaly score. Third, a predefined threshold was used to determine anomalies. Finally, it was determined whether the input data was normal or abnormal.

[0092] The resulting performance was 58% sensitivity and 83% specificity. 50 of 87 abnormal steam pops were predicted before they occurred. 11 of 63 normal test data without steam pops were incorrectly predicted to be abnormal.

[0093] The second row from the top in Figure 6 is the anomaly score for Diff_Power. Diff_Power is a variable with a narrow distribution because it is the difference from the upper limit, and it clearly captures the fluctuations in Power. The anomaly score for Diff_Power was larger before the occurrence of a steam pop. Power is a variable with a wide distribution because the upper limit changes due to cauterization. Diff_Power is a variable with a narrow distribution because it is the difference from the upper limit, and it clearly captures the fluctuations in Power. A large Diff_Power score suggests that fluctuations in Power may affect the occurrence of a steam pop. Figure 7 shows the anomaly score for abnormal data. Colored areas indicate the occurrence of anomalies. The horizontal line at approximately 0.15 anomaly score indicates the threshold for detecting anomalies.

Claims

1. A program for outputting data indicating a prediction of the occurrence of steam pops due to ablation for the treatment of arrhythmia in a subject, the program being generated by a method including a step of inputting active or passive indicators of ablation for the treatment of arrhythmia, and a plurality of data indicating that steam pops did not occur even after a certain period of cauterization by ablation for the treatment of arrhythmia, into an artificial intelligence model as learning data, and having the artificial intelligence model learn the data.

2. The program according to claim 1 , wherein the active or passive indicator of ablation for treating arrhythmia is selected from the group consisting of output, contact force, upper limit temperature, current application time, and saline concentration used for cooling.

3. The program of claim 2 , wherein an active or passive indicator of ablation for treating arrhythmia comprises an output.

4. The program according to claim 1 , wherein the active index or passive index of ablation for arrhythmia treatment is a variable representing a relative difference between a set upper limit value of the active index or passive index and the active index or passive index.

5. The program according to claim 1 , wherein the active or passive indicator of ablation for treating arrhythmia is a variable representing a relative difference between a set upper limit value of output and the output.

6. 2. The program according to claim 1, wherein the certain period is selected from a period of 30 seconds or more.

7. 3. The program according to claim 2, wherein the current application time is 10 seconds or more.

8. The program according to claim 1 , wherein the plurality of data is 50 or more pieces of data.

9. 2. The program according to claim 1, wherein the algorithm used in the step of training the artificial intelligence model is MTAD-GAT.

10. 2. The program according to claim 1, wherein in the step of training the artificial intelligence model, a plurality of data are used by dividing them into training: verification: test = 4-7 : 1-3 : 1-3.

11. The program according to claim 1 , wherein in the step of training the artificial intelligence model, the plurality of data are multivariate time series data.

12. 2. The program according to claim 1, wherein in the step of training the artificial intelligence model, the plurality of data are standardized using the average and standard deviation of each feature within the plurality of data.

13. 2. The program according to claim 1, wherein the step of training the artificial intelligence model includes a step of processing the data into fixed-length data using a moving window.

14. 2. The program according to claim 1, wherein the step of training the artificial intelligence model includes a step of convolving the feature quantities of each of the plurality of data based on a specified kernel size.

15. 15. The program according to claim 14, further comprising a step of filling data lost by the convolution step with zeros.

16. 15. The program of claim 14, wherein the convolution step is performed by taking a sum.

17. The program according to claim 1 , wherein the step of training the artificial intelligence model incorporates a plurality of graph attention mechanisms.

18. 20. The computer-readable medium of claim 17, wherein the plurality of graph attention mechanisms include graph attention for feature direction and graph attention for time direction.

19. 20. The program of claim 17, wherein the step of training the artificial intelligence model includes a step of combining outputs of multiple graph attention mechanisms and the convolutional data and passing the combined data to a GRU layer.

20. 2. The program of claim 1, wherein the step of training the artificial intelligence model includes a step of extracting features by passing them through an encoder based on a GRU layer and passing them to a decoder layer.

21. The program of claim 1, wherein the step of training the artificial intelligence model includes a step of adding a self-attention layer and a fully connected layer to the GRU layer, weighting the output while returning the output so as to restore the original input.

22. 22. The program of claim 21, wherein the reconstructed output extracts only the last column of the window.

23. 2. The program according to claim 1, wherein the step of training the artificial intelligence model includes a step of calculating an anomaly degree at a timestamp t using the following formula, where i is each input feature. [0010]

24. 2. The program of claim 1, wherein outputting data indicating a prediction of the occurrence of a steam pop due to ablation for treating arrhythmia in a subject includes a step of calculating an abnormality degree using verification data, obtaining a percentile value of a predetermined numerical value of the calculated abnormality degree and setting it as an abnormality threshold, and determining that there is a sign of the occurrence of a steam pop if m seconds (where n>m) out of the past n seconds exceed the abnormality threshold.

25. A computer storing the program according to claim 1.

26. A computer-readable recording medium storing the program according to claim 1.

27. 1. An apparatus for outputting data indicative of a prediction of occurrence of steam pop due to ablation for treating arrhythmia in a subject, comprising: means for inputting data indicative of active or passive indicators of ablation for treating arrhythmia in a subject; A means for calculating data indicating an active indicator or a passive indicator of ablation for arrhythmia treatment in a subject using the input data and the program according to claim 1, the data indicating a prediction of occurrence of steam pop due to ablation for arrhythmia treatment in the subject; and means for outputting data indicative of a predicted occurrence of steam pop due to ablation for treating arrhythmia in the subject. Device.

28. 18. Apparatus according to claim 17, comprising means for storing a program according to claim 1.

29. 1. A method for outputting data indicative of a prediction of occurrence of steam pop due to ablation for treating arrhythmia in a subject, comprising: inputting data indicative of active or passive indications of ablation for treating arrhythmia in a subject into the device of claim 27; A step of causing the device to calculate data indicating a prediction of occurrence of steam pop due to ablation for arrhythmia treatment in the subject, using the input data indicating active or passive indicators of ablation for arrhythmia treatment in the subject and the program according to claim 1; and outputting the calculated data indicating a prediction of occurrence of steam pop due to ablation for treating arrhythmia in the subject to the device. method.

30. 30. The method of claim 29, further comprising obtaining data indicative of active or passive indicators of ablation for treating arrhythmia in the subject prior to inputting the data indicative of active or passive indicators of ablation for treating arrhythmia in the subject into the device of claim 27.

31. 30. A method for predicting the occurrence of steam pop due to ablation for treating arrhythmia, comprising the method of claim 29.

Images