Computer program, intracardiac electrocardiogram evaluation device, intracardiac electrocardiogram evaluation system, intracardiac electrocardiogram evaluation method, and learning model generation method
The system uses a learning model to convert body surface electrocardiogram data into evaluation indices for intracardiac electrocardiogram data, addressing the limitations of surface electrocardiograms in predicting arrhythmia diseases, achieving accurate and cost-effective diagnosis of arrhythmia conditions.
Patent Information
- Application Number
- PCT/JP2024/018932
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-23
- Filing Date
- 2024-05-23
- Publication Date
- 2024-11-28
AI Technical Summary
Current surface electrocardiogram methods are inadequate for predicting arrhythmia diseases like atrial fibrillation, as they primarily reflect momentary information and lack the ability to predict predisposition or minute heart details, while invasive intracardiac electrocardiograms offer high diagnostic ability but at the cost of invasiveness and expense.
A computer program and system that uses a first learning model to generate evaluation indices for intracardiac electrocardiogram data from body surface electrocardiogram data, incorporating features like convolution layers and trained models to predict the presence and mechanism of arrhythmia diseases with high accuracy.
Enables accurate prediction of arrhythmia diseases from body surface electrocardiograms, providing a non-invasive and cost-effective solution for diagnosing conditions like atrial fibrillation, thereby reducing medical costs and maintaining individual productivity.
Smart Images

Figure JP2024018932_28112024_PF_FP_ABST
Abstract
Description
Computer program, intracardiac electrocardiogram evaluation device, intracardiac electrocardiogram evaluation system, intracardiac electrocardiogram evaluation method, and learning model generation method
[0001] The present invention relates to a computer program, an intracardiac electrocardiogram evaluation device, an intracardiac electrocardiogram evaluation system, an intracardiac electrocardiogram evaluation method, and a learning model generation method.
[0002] Atrial fibrillation is an arrhythmia that can cause irreversible diseases such as cerebral infarction and heart failure, which have a significant impact on society, and has a high prevalence among the elderly. Therefore, there is an increasing need for arrhythmia diagnosis in order to reduce medical costs across society and aim for a healthy longevity society where individuals can maintain their productivity.
[0003] Such arrhythmias are usually diagnosed using a surface electrocardiogram (ECG) such as a 12-lead ECG. Patent Document 1 discloses an ECG data processing device that automatically detects a part of a 12-lead ECG with high accuracy.
[0004] JP 2011-72542 A
[0005] However, surface electrocardiograms mainly reflect information at the moment of recording and do not necessarily reflect the predisposition to arrhythmia. In other words, surface electrocardiograms cannot obtain information at times other than the time of recording, and even if electrocardiogram findings are obtained during non-attack periods, it is difficult to predict the onset of arrhythmia disorders such as atrial fibrillation.
[0006] While surface electrocardiograms are common and non-invasive, they cannot interpret minute information inside the heart. On the other hand, intracardiac electrocardiograms directly record electrical signals obtained from an electrode catheter inserted into the heart. This provides detailed structural information about the heart and has high diagnostic capabilities, but requires the insertion of a catheter into the body and cardiac chambers, making it an expensive and highly invasive procedure.
[0007] The present invention has been made in consideration of the above circumstances, and aims to provide a computer program, an intracardiac electrocardiogram evaluation device, an intracardiac electrocardiogram evaluation system, an intracardiac electrocardiogram evaluation method, and a learning model generation method that can predict with high accuracy the presence or absence and mechanism of arrhythmia diseases from a surface electrocardiogram.
[0008] The present application includes multiple means for solving the above-mentioned problems. As an example, a computer program causes a computer to execute a process of acquiring a subject's surface electrocardiogram data, inputting the acquired surface electrocardiogram data into a first learning model that outputs an index for evaluating the intracardiac electrocardiogram data when the surface electrocardiogram data is input, and outputting an index for evaluating the subject's intracardiac electrocardiogram data.
[0009] According to the present invention, the presence or absence of arrhythmia and its mechanism can be predicted with high accuracy from a surface electrocardiogram.
[0010] 1 is a diagram showing an example of the configuration of an intracardiac electrocardiogram evaluation system. FIG. 2 is a diagram showing an example of body surface electrocardiogram data. FIG. 3 is a diagram showing an example of the configuration of a first learning model. FIG. 4 is a diagram showing an example of intracardiac electrocardiogram data. FIG. 5 is a diagram showing an example of creation of training data for generating a first learning model. FIG. 6 is a diagram showing an example of a method for generating a first learning model. FIG. 7 is a diagram showing a first example of labels. FIG. 8 is a diagram showing a second example of labels. FIG. 9 is a diagram showing an example of a pattern of coronary sinus potentials. FIG. 10 is a diagram showing a third example of labels. FIG. 11 is a diagram showing an example of an atrial tachycardia pattern. FIG. 12 is a diagram showing an example of an atrioventricular reciprocating tachycardia pattern. FIG. 13 is a diagram showing an example of visualization of prediction grounds using Grad-CAM. FIG. 14 is a diagram showing an example of the configuration of a second learning model. FIG. 15 is a diagram showing an example of arrhythmia diagnosis information. FIG. 16 is a diagram showing an example of processing procedures performed by an intracardiac electrocardiogram evaluation device. FIG. 17 is a diagram showing an example of a processing procedure for generating a first learning model.
[0011] The present invention will be described below with reference to the drawings showing embodiments thereof. Fig. 1 is a diagram showing an example of the configuration of an intracardiac electrocardiogram evaluation system. The intracardiac electrocardiogram evaluation system includes an intracardiac electrocardiogram evaluation device 50 and an electrocardiography device 10. The intracardiac electrocardiogram evaluation device 50 is connected to the electrocardiography device 10 by wired or wireless communication. The intracardiac electrocardiogram evaluation device 50 is also connected to a data server 20 via a communication network 1.
[0012] The electrocardiogram examination device 10 records the surface electrocardiogram data of a subject (e.g., a patient) and outputs the surface electrocardiogram data to the intracardiac electrocardiogram evaluation device 50. The electrocardiogram examination device 10 may be a 12-lead electrocardiogram examination device, a Holter electrocardiogram monitor, a portable electrocardiogram monitor, a wearable terminal, or the like. Note that the function of the intracardiac electrocardiogram evaluation device 50 may be incorporated into the electrocardiogram examination device 10.
[0013] Figure 2 shows an example of body surface electrocardiogram data. Figure 2 shows 12-lead electrocardiogram data as an example of body surface electrocardiogram data. Electricity generated in the sinoatrial node flows from the atrium to the atrioventricular node to the ventricles (right and left bundle branches). A body surface electrocardiogram is a time-series data recorded from the body surface, recording the potential difference between each site. In principle, the atria and ventricles contract during depolarization and expand during repolarization. A 12-lead electrocardiogram observes the heart from 12 different directions on the body surface through this electrical activity. Leads used in 12-lead electrocardiograms include limb leads and chest leads. Limb leads are obtained by attaching electrodes to the wrists and ankles, and six waveforms are obtained from four electrodes: I: leftward, II: downward-left, III: downward-right, aVR: upward-right, aVL: upward-left, and aVF: downward. Chest leads reflect the electrical direction in the horizontal plane, and by attaching six electrodes at predetermined positions on the chest, six waveforms, "V1, V2, V3, V4, V5, and V6," can be obtained from the six electrodes.
[0014] As shown in FIG. 1 , the intracardiac electrocardiogram evaluation device 50 includes a control unit 51 that controls the entire device, a communication unit 52, a memory 53, a display unit 54, an operation unit 55, an evaluation index generation unit 56, an electrocardiogram feature extraction unit 57, an arrhythmia diagnosis information generation unit 58, and a storage unit 60.
[0015] The control unit 51 is configured by incorporating a required number of central processing units (CPUs), micro-processing units (MPUs), graphics processing units (GPUs), general-purpose computing on graphics processing units (GPGPUs), etc. The control unit 51 may also be configured by combining digital signal processors (DSPs), field-programmable gate arrays (FPGAs), etc.
[0016] The communication unit 52 includes a communication module and has a function of communicating with the electrocardiogram examination apparatus 10. The communication unit 52 also has a function of communicating with the data server 20 via the communication network 1.
[0017] The memory 53 can be configured with a semiconductor memory such as a static random access memory (SRAM), a dynamic random access memory (DRAM), or a flash memory.
[0018] The display unit 54 can be configured with a liquid crystal display, an organic EL display, etc. Note that instead of the display unit 54, an external display device may be connected to the intracardiac electrocardiogram evaluation device 50.
[0019] The operation unit 55 is configured with, for example, a keyboard, a mouse, a touchpad, or a touch panel, and can accept operations on the information displayed on the display unit 54 .
[0020] The storage unit 60 can be configured, for example, with a hard disk or a semiconductor memory, and can store a computer program (program product) 61, a first learning model 62, a second learning model 63, and required information. The first learning model 62 and the second learning model 63 include a model before learning, a model in the middle of learning, or a trained model.
[0021] The computer program 61 is deployed in the memory 53 and executed by the control unit 51. The computer program 61 may be downloaded from an external device via the communication unit 52 and stored in the storage unit 60. Alternatively, the computer program 61 may be recorded on a recording medium (for example, an optically readable disk storage medium such as a CD-ROM) and read by a recording medium reading unit and stored in the storage unit 60. The computer program 61 may be deployed so as to be executed on a single computer, or on multiple computers located at one site, or distributed across multiple sites and interconnected by a communications network.
[0022] The evaluation index generating unit 56 and the electrocardiogram feature extracting unit 57 can execute their functions by performing processing using a first learning model 62 with a computer program 61. The arrhythmia diagnostic information generating unit 58 can execute its functions by performing processing using a second learning model 63 with a computer program 61.
[0023] 3 is a diagram showing an example of the configuration of the first learning model 62. The first learning model 62 includes an input layer 621, a convolutional layer 622, and an output layer 623. The input layer 621 has a predetermined number of channels. When the input surface electrocardiogram data is 12-lead electrocardiogram data, 12 pieces of time-series data of the 12-lead electrocardiogram data are input to the one-dimensional 12 channels of the input layer 621.
[0024] The convolutional layer 622 has multiple layers, extracts surface electrocardiogram features that are features of the surface electrocardiogram data, and outputs the extracted surface electrocardiogram features to the output layer 623. The surface electrocardiogram features are extracted as features by combining, for example, the shape of each waveform in the surface electrocardiogram data, the order in which waves (e.g., P waves, QRS waves, T waves, etc.), the intervals between characteristic portions of the waveform, and the like.
[0025] The output layer 623 can be configured, for example, as a fully connected layer, and outputs a probability indicating to which label the input surface ECG data belongs by combining the surface ECG feature values output from the convolution layer 622 into a one-dimensional numerical value. Here, the label is an index (evaluation index) for evaluating the intracardiac ECG data.
[0026] The evaluation index is information obtained by abstracting the intracardiac electrocardiogram data, and is artificially labeled by, for example, an arrhythmia specialist using the intracardiac electrocardiogram data.
[0027] The first learning model 62 can be configured with, for example, a model such as EfficientNet, ResNet, DenseNet, Vision Transformer, etc. As described above, by performing processing using the first learning model 62, the function of the evaluation index generation unit 56 that generates an evaluation index for intracardiac electrocardiogram data based on the surface electrocardiogram data and the function of the electrocardiogram feature extraction unit 57 that extracts surface electrocardiogram features, which are features of the surface electrocardiogram data, based on the surface electrocardiogram data, are executed.
[0028] FIG. 4 illustrates an example of intracardiac electrocardiogram data. Sources of intracardiac electrocardiogram data that serve as evaluation indices for intracardiac electrocardiogram data include, for example, pulmonary vein potentials, coronary sinus potentials, right ventricular potentials, and ablation catheter potentials. Different intracardiac electrocardiogram data sources can be used depending on the type of arrhythmia being evaluated (e.g., paroxysmal atrial fibrillation, persistent atrial fibrillation, postoperative recurrent atrial fibrillation, paroxysmal supraventricular tachycardia, ventricular tachycardia, ventricular premature contractions, atrial tachycardia, ventricular fibrillation, WPS syndrome, etc.). The example in FIG. 4 illustrates intracardiac electrocardiogram data obtained from electrode catheters inserted into predetermined locations in the right atrium (MAP), coronary sinus (CS), and pulmonary veins (SPV). An arrhythmia specialist can abstract the intracardiac electrocardiogram data illustrated in FIG. 4 and artificially classify it as evaluation indices (labels).
[0029] FIG. 5 shows an example of how training data is generated to generate the first learning model 62. Patients with arrhythmia (P1, P2, ..., Px) are registered. Their surface electrocardiogram data (B1, B2, ..., Bx) and intracardiac electrocardiogram data (I1, I2, ..., Ix) are collected, and the intracardiac electrocardiogram data of each patient is abstracted and labeled. In the example of FIG. 5 , for convenience, labels 1, 2, ..., N are used. Depending on the patient's arrhythmia, different labels may be assigned to different patients, or the same label may be used. In this way, the abstracted intracardiac electrocardiograms are prepared as correct labels. Training data can be generated using an external device (e.g., a personal computer). The training data generated by the external device can be stored in the data server 20.
[0030] 6 is a diagram showing an example of a method for generating the first learning model 62. The control unit 51 acquires training data including indices (correct labels) for evaluating surface electrocardiogram data and intracardiac electrocardiogram data from the data server 20. Based on the acquired training data, the control unit 51 can generate the first learning model 62 so as to output indices for evaluating the intracardiac electrocardiogram data of a subject when the subject's surface electrocardiogram data is input.
[0031] Specifically, surface electrocardiogram data as learning input data is input to the first learning model 62. The parameters in the first learning model 62 are adjusted so that the labels (evaluation indices) output by the first learning model 62 match the correct labels as training data. When the degree of match between the labels as output data and the correct labels falls within an acceptable range, the parameters of the first learning model 62 can be stored in the storage unit 60, and the first learning model 62 can be generated.
[0032] 7 is a diagram showing a first example of labels. Examples of labels that abstract intracardiac electrocardiogram data can be classified as follows: Label 1 indicates pulmonary vein potential, Label 2 indicates no pulmonary vein potential, Label 3 indicates coronary sinus potential, Label 4 indicates no coronary sinus potential, Label 5 indicates right ventricular potential, Label 6 indicates no right ventricular potential, Label 7 indicates ablation catheter potential, Label 8 indicates no ablation catheter potential, etc. Note that the label numbers are assigned for convenience and are not limited to the example shown in FIG. 7.
[0033] As described above, the evaluation index (label) includes the presence or absence of a pulmonary vein potential. The evaluation index may also include at least one of the presence or absence of a coronary sinus potential, the presence or absence of a right ventricular potential, and the presence or absence of an ablation catheter potential.
[0034] The control unit 51 acquires the subject's surface electrocardiogram data, and when the surface electrocardiogram data is input, it can input the acquired surface electrocardiogram data into the first learning model 62, which outputs an index (evaluation index, label) for evaluating the intracardiac electrocardiogram data, and output an index for evaluating the subject's intracardiac electrocardiogram data.
[0035] Atrial fibrillation is an arrhythmia that occurs when electrical signals are generated disorderly near blood vessels called "pulmonary veins" in the left atrium of the heart. It is thought that aging causes cells with abnormal electrical currents to appear in the pulmonary veins, and that these cells cause a strong electrical current to flow into the atrium, causing it to lose balance and enter a state of fibrillation.
[0036] That is, when the surface electrocardiogram data of a subject is input to the first learning model 62, if the first learning model 62 outputs label 1, it is possible to predict with high accuracy that the subject is at risk of arrhythmia due to atrial fibrillation and the associated risk of serious events such as cerebral infarction. Thus, according to this embodiment, the presence or absence and mechanism of arrhythmia can be predicted with high accuracy from the surface electrocardiogram.
[0037] 8 is a diagram showing a second example of labels. As an example of labels that abstract intracardiac electrocardiogram data, for example, label 11 is a potential pattern aaa1 of region AAA, label 12 is a potential pattern aaa2 of region AAA, ... label 21 is a potential pattern bbb1 of region BBB, label 22 is a potential pattern bbb2 of region BBB, ... The label numbers, regions, and potential patterns are assigned for convenience and are not limited to the example of FIG. 8.
[0038] As described above, the evaluation index (label) may include clusters classified by clustering based on potential patterns observed by multiple electrodes placed at required locations on the heart. A correct label may be assigned to each cluster classified by clustering.
[0039] FIG. 9 shows an example of a coronary sinus potential pattern. The region AAA is the coronary sinus, label 11 is the normal label, and label 12 is the abnormal label. Of the electrodes CS, CS1 and CS2 are on the left side of the heart, and CS9 and CS10 are on the right side of the heart. In the normal label (label 11), the potential pattern flows from CS9 and CS1 to CS2 (i.e., from the right side to the left side of the heart) over time. On the other hand, in the abnormal label (label 12), the potential pattern flows from CS1 and CS2 to CS9 and CS10 (i.e., from the left side to the right side of the heart) over time.
[0040] That is, when the surface electrocardiogram data of a subject is input to the first learning model 62, if the first learning model 62 outputs the label 12, it is possible to predict with high accuracy that the subject is at risk of arrhythmia due to an imbalance in the potential in the coronary sinus, and the associated risk of serious events such as cerebral infarction. Thus, according to this embodiment, the presence or absence and mechanism of arrhythmia can be predicted with high accuracy from the surface electrocardiogram.
[0041] FIG. 10 is a diagram showing a third example of labels. Intracardiac electrocardiogram data from multiple patients with arrhythmia is collected, and the intracardiac electrocardiogram data is clustered using the arrangement of intracardiac electrocardiogram waveforms at site XXX and the intracardiac electrocardiogram waveforms as classification axes. Site XXX may include multiple sites. The arrangement of intracardiac electrocardiogram waveforms is the temporal order of the waveforms at each electrode. The intracardiac electrocardiogram waveforms include the waveform shape of each electrode, the magnitude of the potential, the spacing between characteristic waveform points, etc. In the example of FIG. 10 , for convenience, the data is classified into four clusters labeled 31, 32, 33, and 34. Note that clustering may be performed using either the arrangement of intracardiac electrocardiogram waveforms or the intracardiac electrocardiogram waveforms as classification axes.
[0042] As described above, the evaluation index (label) may include a cluster classified by clustering based on at least one of a plurality of intracardiac electrocardiogram waveforms observed with a plurality of electrodes placed at desired locations on the heart and an arrangement of a plurality of intracardiac electrocardiogram waveforms.
[0043] Fig. 11 shows an example of an atrial tachycardia pattern. Fig. 11 shows intracardiac electrocardiogram data obtained from electrode catheters inserted into predetermined locations in the right atrium (RA) and coronary sinus (CS). The waveform and waveform arrangement indicated by symbol A represent a case of sinus rhythm (normal pattern). The waveform and waveform arrangement indicated by symbol B represent a case of an atrial tachycardia pattern. Atrial tachycardia is a tachycardia caused by the presence of cardiomyocytes with abnormal excitability in the atrium, which excite the heart at a frequency higher than that of the sinus node.
[0044] When the label (evaluation index) of the atrial tachycardia pattern is label 31 and the subject's surface electrocardiogram data is input to the first learning model 62, if the first learning model 62 outputs label 31, it is possible to predict with high accuracy that the subject is at risk of arrhythmia due to atrial tachycardia and the associated risk of serious events such as cerebral infarction. Thus, according to this embodiment, the presence or absence and mechanism of arrhythmia disease can be predicted with high accuracy from the surface electrocardiogram.
[0045] FIG. 12 is a diagram showing an example of an atrioventricular reciprocating tachycardia pattern. FIG. 12 shows intracardiac electrocardiogram data obtained from an electrode catheter inserted into a predetermined location in the coronary sinus (CS). The waveform and waveform arrangement within the framed time period indicate an atrioventricular reciprocating tachycardia pattern. Atrioventricular reciprocating tachycardia (WPS syndrome) is a type of tachycardia that occurs when an accessory pathway exists in addition to a normal pathway in the impulse conduction system connecting the atria and ventricles, and electrical signals circulate between the normal pathway and the accessory pathway.
[0046] When the label (evaluation index) of the atrioventricular reciprocating tachycardia pattern is label 41 and the subject's surface electrocardiogram data is input to the first learning model 62, if the first learning model 62 outputs label 41, it is possible to predict with high accuracy that the subject is at risk of arrhythmia due to atrioventricular reciprocating tachycardia and the associated risk of serious events such as cerebral infarction. Thus, according to this embodiment, the presence or absence and mechanism of arrhythmia disease can be predicted with high accuracy from the surface electrocardiogram.
[0047] Next, visualization of the prediction basis using the first learning model 62 will be described.
[0048] 13 is a diagram showing an example of visualization of prediction grounds using Grad-CAM. Grad-CAM inputs surface electrocardiogram data into a first learning model 62, extracts a feature map from the first learning model 62, calculates weights for each channel of the extracted feature map, and adds up the weighted channels to generate a saliency map. Since the more important the part of the surface electrocardiogram data, the greater the saliency, making it possible to visualize the prediction grounds.
[0049] In Figure 13, the areas that the first learning model 62 judged to be important are indicated by different shades of gray for both the surface electrocardiogram data with and without pulmonary vein potential. The darker the area, the higher the importance. It was found that the prediction basis was concentrated in time phases close to the appearance position of the pulmonary vein potential. This is also useful for identifying the pulmonary vein potential on the surface electrocardiogram data.
[0050] 14 is a diagram showing an example of the configuration of the second learning model 63. The second learning model 63 includes an input layer 631, an intermediate layer 632, and an output layer 633. The intermediate layer 632 may be a convolutional layer, but does not have to be a convolutional layer. When the surface electrocardiogram features extracted by the first learning model 62 are input to the second learning model 63, the second learning model 63 outputs arrhythmia diagnosis information.
[0051] That is, the control unit 51 acquires the subject's surface electrocardiogram features, and when the surface electrocardiogram features are input, the control unit 51 can input the acquired surface electrocardiogram features into the second learning model 63, which outputs arrhythmia diagnosis information, and output arrhythmia diagnosis information for the subject.
[0052] By performing processing using the second learning model 63, the function of the arrhythmia diagnostic information generating unit 58 that generates arrhythmia diagnostic information based on the body surface electrocardiogram features is executed.
[0053] The second learning model 63 may be generated as follows. That is, the control unit 51 acquires training data including surface electrocardiogram features and arrhythmia diagnosis information, and based on the acquired training data, generates the second learning model 63 so that, when the surface electrocardiogram features of a subject are input, the second learning model 63 outputs arrhythmia diagnosis information of the subject. The training data may be stored in advance in the data server 20.
[0054] Although not shown, the second learning model 63 may be generated as follows. That is, the control unit 51 may acquire training data including surface electrocardiogram data and arrhythmia diagnosis information, and generate the second learning model 63 based on the acquired training data so that, when the surface electrocardiogram data of a subject is input, the arrhythmia diagnosis information of the subject is output. The training data may be stored in advance in the data server 20.
[0055] FIG. 15 is a diagram showing an example of arrhythmia diagnosis information. As shown in FIG. 15 , the predicted occurrence of atrial fibrillation is, for example, △△% within XX years. The risk of cerebral infarction is, for example, low, medium, or high. The necessity of anticoagulant therapy is, for example, recommended or not recommended. The type of arrhythmia is, for example, atrial fibrillation, ventricular tachycardia, atrial tachycardia, premature ventricular contraction, etc. The necessity of ablation is, for example, recommended or not recommended. Ablation is a treatment for atrial fibrillation using pulmonary vein isolation (PVI). Different second learning models 63 may be used for different items of arrhythmia diagnosis information, or the same second learning model 63 may be used for multiple items.
[0056] To generate the second learning model 63, medical information such as surface electrocardiogram data, diagnosis information, treatment information, and intracardiac electrocardiogram data of multiple patients with arrhythmia is collected for each patient. Based on the medical information of multiple patients, the second learning model 63 is trained to learn, for example, the correlation between the time when the surface electrocardiogram data was acquired and the time when atrial fibrillation occurred, thereby making it possible to predict the onset of atrial fibrillation in a certain subject from the subject's surface electrocardiogram data.
[0057] Furthermore, by having the second learning model 63 learn, based on the medical information of multiple patients, for example, the correlation between the time when surface electrocardiogram data was acquired and whether or not cerebral infarction has occurred, it becomes possible to predict the risk of cerebral infarction of a subject from the subject's surface electrocardiogram data.
[0058] Furthermore, by having the second learning model 63 learn the correlation between, for example, surface electrocardiogram data and treatment information indicating whether or not anticoagulant therapy was required based on the medical information of multiple patients, it becomes possible to determine whether or not a subject requires anticoagulant therapy based on the subject's surface electrocardiogram data.
[0059] Furthermore, by having the second learning model 63 learn the correlation between, for example, surface electrocardiogram data and the type of arrhythmia based on intracardiac electrocardiogram data based on the medical information of multiple patients, it becomes possible to determine the type of arrhythmia of a subject from the surface electrocardiogram data of the subject.
[0060] Furthermore, by having the second learning model 63 learn, for example, the correlation between surface electrocardiogram data and symptoms after ablation based on the medical information of multiple patients, it becomes possible to determine whether a subject needs ablation based on the subject's surface electrocardiogram data.
[0061] 16 is a diagram showing an example of a processing procedure performed by the intracardiac electrocardiogram evaluation device 50. For convenience, the processing will be described as being performed by the control unit 51. The control unit 51 acquires the subject's surface electrocardiogram data (S11) and inputs the acquired surface electrocardiogram data into the first learning model 62 (S12). The control unit 51 acquires an evaluation index for the intracardiac electrocardiogram data from the output layer 623 of the first learning model 62 (S13).
[0062] The control unit 51 extracts surface electrocardiogram features from the convolutional layer 622 of the first learning model 62 (S14). The control unit 51 inputs the extracted surface electrocardiogram features to the second learning model 63 (S15). The control unit 51 acquires arrhythmia diagnosis information from the second learning model 63 (S16).
[0063] The control unit 51 outputs the acquired evaluation index (S17), outputs the acquired arrhythmia diagnosis information (S18), and ends the process. The arrhythmia diagnosis information may be displayed on the display unit 54 or an external display device.
[0064] 17 is a diagram showing an example of a generation process procedure for the first learning model 62. The control unit 51 acquires training data including evaluation indices (correct labels) for the surface electrocardiogram data and the intracardiac electrocardiogram data (S21). The control unit 51 inputs the surface electrocardiogram data as learning input data into the first learning model 62 and acquires evaluation indices (output labels) output by the first learning model 62 (S22).
[0065] The control unit 51 adjusts the parameters of the first learning model 62 so that the acquired evaluation index (output label) approaches the evaluation index (correct label) included in the training data (S23). The control unit 51 determines whether the prediction accuracy of the first learning model 62 is within an acceptable range (S24), and if the prediction accuracy is not within the acceptable range (NO in S24), continues the processing from step S22 onwards.
[0066] If the prediction accuracy is within the acceptable range (YES in S24), the control unit 51 stores the parameters of the first learning model 62 in the memory unit 60 (S25), generates the first learning model 62 (S26), and terminates the processing.
[0067] According to this embodiment, the pulmonary vein potential (a minute potential confirmed by an intracardiac electrocardiogram), which is believed to be the cause of atrial fibrillation, can be detected at an appropriate time phase from surface electrocardiogram data, and can be used as a basic technology for using a clinical arrhythmia prediction model.
[0068] (Appendix 1) The computer program acquires surface electrocardiogram data of a subject, and when the surface electrocardiogram data is input, causes the computer to execute a process of inputting the acquired surface electrocardiogram data into a first learning model that outputs an index for evaluating the intracardiac electrocardiogram data of the subject, and outputting an index for evaluating the intracardiac electrocardiogram data of the subject.
[0069] (Supplementary Note 2) In the computer program according to Supplementary Note 1, the indicators include the presence or absence of a pulmonary vein potential.
[0070] (Supplementary Note 3) In the computer program according to Supplementary Note 1 or Supplementary Note 2, the indicators include at least one of the presence or absence of coronary sinus potential, the presence or absence of right ventricular potential, and the presence or absence of ablation catheter potential.
[0071] (Supplementary Note 4) In the computer program according to any one of Supplementary Note 1 to Supplementary Note 3, the index includes clusters classified by clustering based on potential patterns observed at multiple electrodes placed at required locations on the heart.
[0072] (Supplementary Note 5) In the computer program according to any one of Supplementary Note 1 to Supplementary Note 4, the index includes clusters classified by clustering based on at least one of a plurality of intracardiac electrocardiogram waveforms observed by a plurality of electrodes placed at required locations on the heart and an arrangement of the plurality of intracardiac electrocardiogram waveforms.
[0073] (Appendix 6) The computer program in any one of Appendices 1 to 5 causes a computer to execute a process of extracting electrocardiogram features based on acquired surface electrocardiogram data, and when the electrocardiogram features are input, inputting the extracted electrocardiogram features into a second learning model that outputs arrhythmia diagnosis information, and outputting arrhythmia diagnosis information for the subject.
[0074] (Supplementary Note 7) In the computer program according to Supplementary Note 6, the arrhythmia diagnostic information includes at least one of atrial fibrillation risk, cerebral infarction risk, and type of arrhythmia.
[0075] (Appendix 8) The intracardiac electrocardiogram evaluation device comprises a control unit and a first learning model that, when surface electrocardiogram data is input, outputs an index for evaluating the intracardiac electrocardiogram data, and the control unit acquires the surface electrocardiogram data of a subject, inputs the acquired surface electrocardiogram data into the first learning model, and outputs an index for evaluating the intracardiac electrocardiogram data of the subject.
[0076] (Supplementary Note 9) An intracardiac electrocardiogram evaluation system includes the above-mentioned intracardiac electrocardiogram evaluation device and an electrocardiogram examination device, and the intracardiac electrocardiogram evaluation device acquires the subject's surface electrocardiogram data from the electrocardiogram examination device.
[0077] (Appendix 10) The intracardiac electrocardiogram evaluation method acquires a subject's surface electrocardiogram data, and when the surface electrocardiogram data is input, inputs the acquired surface electrocardiogram data into a first learning model that outputs an index for evaluating the intracardiac electrocardiogram data, and outputs an index for evaluating the intracardiac electrocardiogram data of the subject.
[0078] (Appendix 11) The learning model generation method acquires training data including indices for evaluating surface electrocardiogram data and intracardiac electrocardiogram data, and generates a first learning model based on the acquired training data so that, when a subject's surface electrocardiogram data is input, an indices for evaluating the subject's intracardiac electrocardiogram data are output.
[0079] The matters described in each embodiment can be combined with each other. Furthermore, the independent claims and dependent claims described in the claims can be combined with each other in any and all combinations, regardless of the reference format. Furthermore, the claims use a format in which a claim references two or more other claims (multiple claim format), but this is not limited to this. A multiple claim (multi-multi claim) that references at least one other multiple claim may also be used.
[0080] REFERENCE SIGNS LIST 1 Communication network 10 Electrocardiogram examination device 20 Data server 50 Intracardiac electrocardiogram evaluation device 51 Control unit 52 Communication unit 53 Memory 54 Display unit 55 Operation unit 56 Evaluation index generation unit 57 Electrocardiogram feature extraction unit 58 Arrhythmia diagnosis information generation unit 60 Storage unit 61 Computer program 62 First learning model 63 Second learning model
Claims
1. A computer program that causes a computer to execute a process of acquiring surface electrocardiogram data of a subject, inputting the acquired surface electrocardiogram data into a first learning model that outputs an index for evaluating intracardiac electrocardiogram data when surface electrocardiogram data is input, and outputting an index for evaluating the intracardiac electrocardiogram data of the subject.
2. The computer program according to claim 1, wherein the index includes the presence or absence of a pulmonary vein potential.
3. The computer program according to claim 1, wherein the indicators include at least one of the presence or absence of a coronary sinus potential, the presence or absence of a right ventricular potential, and the presence or absence of an ablation catheter potential.
4. The computer program according to claim 1, wherein the index includes clusters classified by clustering based on potential patterns observed at multiple electrodes placed at required locations on the heart.
5. The computer program of claim 1, wherein the index includes clusters classified by clustering based on at least one of a plurality of intracardiac electrocardiogram waveforms observed by a plurality of electrodes placed at required locations of the heart and a sequence of the plurality of intracardiac electrocardiogram waveforms.
6. A computer program as claimed in any one of claims 1 to 5, which causes a computer to execute a process of extracting electrocardiogram features based on the acquired surface electrocardiogram data, and inputting the extracted electrocardiogram features into a second learning model which outputs arrhythmia diagnosis information for the subject when the electrocardiogram features are input.
7. The computer program according to claim 6, wherein the arrhythmia diagnosis information includes at least one of atrial fibrillation risk, cerebral infarction risk, and type of arrhythmia.
8. An intracardiac electrocardiogram evaluation device comprising: a control unit; and a first learning model that outputs an index for evaluating intracardiac electrocardiogram data when surface electrocardiogram data is input, wherein the control unit acquires the surface electrocardiogram data of a subject, inputs the acquired surface electrocardiogram data into the first learning model, and outputs an index for evaluating the intracardiac electrocardiogram data of the subject.
9. An intracardiac electrocardiogram evaluation system comprising the intracardiac electrocardiogram evaluation device according to claim 8 and an electrocardiogram examination device, wherein the intracardiac electrocardiogram evaluation device acquires surface electrocardiogram data of a subject from the electrocardiogram examination device.
10. A method for evaluating an intracardiac electrocardiogram, comprising: acquiring a subject's surface electrocardiogram data; inputting the acquired surface electrocardiogram data into a first learning model which, when the surface electrocardiogram data is input, outputs an index for evaluating the intracardiac electrocardiogram data; and outputting an index for evaluating the subject's intracardiac electrocardiogram data.
11. A learning model generation method comprising: acquiring training data including indicators for evaluating surface electrocardiogram data and intracardiac electrocardiogram data; and generating a first learning model based on the acquired training data so as to output indicators for evaluating the intracardiac electrocardiogram data of a subject when the subject's surface electrocardiogram data is input.
Citation Information
Patent Citations
Device, method and program for processing electrocardiographic data
JP2011072542A
Intracardiac defibrillation system and cardiac defibrillator
JP2019058350A
Systems and methods for mapping functional nerves innervating arterial walls, 3D mapping therefor, and catheters therefor
JP2019505249A
Reducing noise of intracardiac electrocardiograms using autoencoder, and utilizing and refining intracardiac and body surface electrocardiograms using deep learning training loss functions
JP2021186685A
Reconstruction of a surface electrocardiogram from an endocardial electrogram using non-linear filtering
US20130345580A1
