Device and method for predicting inducible ischemic heart disease by using electrocardiogram-based artificial intelligence

Abstract

The present invention relates to a device and a method for predicting inducible ischemic heart disease by using electrocardiogram-based artificial intelligence. According to the present invention, the device or the method for predicting inducible ischemic heart disease by using electrocardiogram-based artificial intelligence comprises: a data input unit for receiving electrocardiogram data of a set time length observed in a subject; a data pre-processing unit for pre-processing the input electrocardiogram data according to a set method; and a prediction unit for predicting the presence or absence of inducible myocardial ischemia in the subject by inputting the pre-processed electrocardiogram data into a pre-trained artificial intelligence model.

Description

Device and method for predicting inducible ischemic heart disease using electrocardiogram-based artificial intelligence

[0001] The present invention relates to a device and method for predicting inducible ischemic heart disease using electrocardiogram-based artificial intelligence, and more specifically, to a device and method for predicting inducible ischemic heart disease that can predict the presence or absence of inducible myocardial ischemia by inputting electrocardiogram data of a set length of time observed from a patient into an artificial intelligence model.

[0002] Ischemic heart disease (IHD) is a chronic, progressive condition that accounts for the largest proportion of death and disability worldwide. Myocardial ischemia occurs when the supply of oxygen to the heart muscle is inadequate relative to its needs, resulting in relative oxygen deprivation. It can be triggered by factors such as exercise or emotional upheaval and is frequently associated with angina or myocardial infarction. However, myocardial ischemia can also occur in asymptomatic individuals.

[0003] Because inducible myocardial ischemia is associated with a poor prognosis, regardless of the presence or absence of symptoms, it is crucial for clinicians to detect patients at risk for inducible myocardial ischemia early. However, currently, there is no effective method for identifying inducible myocardial ischemia, especially in asymptomatic patients.

[0004] In recent years, the remarkable advancements in artificial intelligence (AI), combined with deep neural network architecture and powerful computing, have led to an increase in research and practical use cases utilizing AI in the medical field.

[0005] In particular, several studies using AI to interpret electrocardiograms (ECGs) have shown that AI performs well in identifying cardiac abnormalities that cardiologists cannot recognize from ECGs alone, such as severe aortic stenosis, left ventricular dysfunction, and asymptomatic or paroxysmal atrial fibrillation.

[0006] Myocardial ischemia is also associated with disease prognosis. Several differences have been reported between patients with and without provoked myocardial ischemia, even at rest, including various biomarkers and responses at the cellular level.

[0007] However, no practical screening tool exists yet to identify patients with probable myocardial ischemia, and a technology that can identify patients with probable myocardial ischemia from electrocardiogram data using artificial intelligence is needed.

[0008] The technology underlying the present invention is disclosed in Korean Patent Publication No. 10-2020-0084561 (published on July 13, 2020).

[0009] The purpose of the present invention is to provide a device and method for predicting inducible ischemic heart disease using electrocardiogram-based artificial intelligence, which can quickly and easily predict the presence or absence of inducible myocardial ischemia by inputting electrocardiogram data of a set length of time observed from a subject into an artificial intelligence model.

[0010] The present invention relates to a device for predicting inducible ischemic heart disease using electrocardiogram-based artificial intelligence, comprising: a data input unit for receiving electrocardiogram data of a set length of time observed from a subject; a data preprocessing unit for preprocessing the input electrocardiogram data according to a set method; and a prediction unit for predicting the presence or absence of inducible myocardial ischemia in the subject by inputting the preprocessed electrocardiogram data into a pre-trained artificial intelligence model.

[0011] Additionally, the data input unit can receive data including a 1-lead, 2-leads, 6-leads, or 12-leads electrocardiogram of 5 to 15 seconds in length.

[0012] In addition, the data preprocessing unit may perform at least one of preprocessing among preprocessing for removing power noise of a set frequency band used by a measuring device for input electrocardiogram data, preprocessing for removing baseline wander noise caused by the subject's breathing or movement, and preprocessing for removing muscle noise caused by muscle movement.

[0013] In addition, the above-mentioned inducible ischemic heart disease prediction device may further include a control unit that pre-builds the artificial intelligence model using a data set including electrocardiogram data collected for each of a plurality of patients and whether or not myocardial ischemia exists.

[0014] In addition, the control unit can train the artificial intelligence model by additionally utilizing data that has been enhanced using a data enhancement technique after preprocessing the electrocardiogram data for each of the plurality of patients.

[0015] In addition, the above-mentioned multiple patients are divided into a group with and without inducible myocardial ischemia, and the group with inducible myocardial ischemia may include patients who underwent percutaneous coronary intervention (PCI) excluding patients who underwent PCI for myocardial infarction, and who received PCI in addition to drug therapy (Guideline-Directed Medical Therapy (GDMT), and were diagnosed with silent myocardial ischemia, stable angina, or unstable angina, and who have an electrocardiogram (ECG) record within 2 weeks of the time of PCI.

[0016] In addition, the group without myocardial ischemia may be comprised of patients who have coronary artery stenosis = 0% and coronary artery calcium score (CACS) ≤ 100 points in all coronary arteries in coronary CT angiography, and who have an interval of less than 2 weeks between the time of CCTA and the time of electrocardiogram (ECG) observation, and who have no history of coronary artery intervention or bypass surgery (revascularization) in the past, myocardial infarction (MI), or other diseases that may clinically cause ischemia or inducible ischemia.

[0017] And, the present invention provides a method for predicting inducible ischemic heart disease performed by a device for predicting inducible ischemic heart disease using electrocardiogram-based artificial intelligence, the method including the steps of: receiving electrocardiogram data of a set length of time observed from a subject; preprocessing the input electrocardiogram data according to a set method; and inputting the preprocessed electrocardiogram data into a pre-trained artificial intelligence model to predict the presence or absence of inducible myocardial ischemia in the subject.

[0018] In addition, the method for predicting the inducible ischemic heart disease may further include a step of pre-constructing the artificial intelligence model using a data set including electrocardiogram data collected for each of multiple patients and whether or not myocardial ischemia exists.

[0019] According to the present invention, the presence or absence of induced myocardial ischemia can be quickly and easily predicted by inputting electrocardiogram data of a set length of time observed from a subject into an artificial intelligence model.

[0020] The present invention can be utilized as a screening tool for determining the possibility of myocardial ischemia that can be induced from electrocardiogram data.

[0021] FIG. 1 is a diagram showing the configuration of a system for predicting inducible ischemic heart disease using electrocardiogram-based artificial intelligence according to an embodiment of the present invention.

[0022] Figure 2 is a diagram showing the configuration of the inducible ischemic heart disease prediction device illustrated in Figure 1.

[0023] FIG. 3 is a diagram exemplarily showing the architecture of an artificial intelligence model according to an embodiment of the present invention.

[0024] FIG. 4 is a diagram showing an example of selecting two groups of patients (a group with myocardial ischemia and a group without myocardial ischemia) that provide data necessary for learning an artificial intelligence model in an embodiment of the present invention.

[0025] Figure 5 is a diagram showing the prediction performance of the proposed artificial intelligence model of the present invention.

[0026] Figure 6 is a diagram showing the distribution of prediction results by an artificial intelligence module trained with electrocardiogram data.

[0027] FIG. 7 is a diagram illustrating a method for predicting inducible ischemic heart disease using electrocardiogram-based artificial intelligence according to an embodiment of the present invention.

[0028] Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings so that those skilled in the art can easily practice the present invention. However, the present invention may be implemented in various different forms and is not limited to the embodiments described herein. In addition, in the drawings, parts irrelevant to the description have been omitted to clearly explain the present invention, and similar parts have been designated with similar reference numerals throughout the specification.

[0029] Throughout the specification, when a part is said to be "connected" to another part, this includes not only the cases where the parts are "directly connected" but also the cases where the parts are "electrically connected" with other elements intervening. Furthermore, when a part is said to "include" a component, this does not exclude other components, but rather includes other components, unless otherwise stated.

[0030] FIG. 1 is a diagram showing the configuration of a system for predicting inducible ischemic heart disease using electrocardiogram-based artificial intelligence according to an embodiment of the present invention.

[0031] As shown in Fig. 1, a system for predicting inducible ischemic heart disease using electrocardiogram-based artificial intelligence according to an embodiment of the present invention may include an inducible ischemic heart disease prediction device (100) and a user terminal (200).

[0032] The device (100) for predicting inducible ischemic heart disease can be connected to a user terminal (200) via a wired, wireless, or wired / wireless combination network to transmit and receive information with each other. In the case of a wireless network, at least one of RF, WLAN, Wi-Fi, and Bluetooth methods can be included, and various known wireless network methods can be used.

[0033] The proposed inducible ischemic heart disease prediction device (100) may be implemented as a web server or app server that receives electrocardiogram (ECG) data of a subject from a network-connected user terminal (200), performs deep learning analysis, and provides a prediction result on the presence or absence of inducible myocardial ischemia of the subject, or may be implemented in the form of an application program or application (Application) on a user terminal, etc.

[0034] In addition, the inducible ischemic heart disease prediction device (100) can provide a service platform implemented as an application or web to a network-connected user terminal (200). In addition, the inducible ischemic heart disease prediction device (100) can be provided in the form of a user terminal (200) built into the form of an application program or the like. The service platform can be an application program that runs in an app or web environment, and the user terminal (200) can receive related services by being connected to the device (100) through a network while the related application program is running, or by executing an internal application program.

[0035] The user terminal (200) may correspond to a terminal of a medical staff member or a related worker, and may include a device that can connect to a wired or wireless network to exchange information, such as a PC, desktop, smart phone, tablet, notebook, or pad.

[0036] This inducible ischemic heart disease prediction device (100) can be used as an artificial intelligence-based screening tool that can determine whether inducible myocardial ischemia exists from a patient's electrocardiogram data, and as a diagnostic and prediction tool that can be used in various clinical situations where inducible myocardial ischemia may have an effect, such as surgery or procedures.

[0037] Figure 2 is a diagram showing the configuration of the inducible ischemic heart disease prediction device illustrated in Figure 1.

[0038] As shown in FIG. 2, the device (100) for predicting induced ischemic heart disease according to an embodiment of the present invention includes a data input unit (110), a data preprocessing unit (120), a prediction unit (130), and may further include a control unit (140). Here, the operation of each unit (110 to 130) and the data flow between each unit may be controlled by the control unit (140).

[0039] This inducible ischemic heart disease prediction device (100) may be implemented as a computer device that is physically configured to include a processor, memory, a user interface input / output device and a storage device, a network input / output unit, etc., or may be implemented as an application program running on a computer device or a user terminal.

[0040] The data input unit (110) can receive electrocardiogram data of a set time length observed from the subject.

[0041] Here, the data input unit (110) can input electrocardiogram data including 1-lead, 2-leads, 6-leads, and 12-leads, each 5 to 15 seconds long, observed at a set sampling rate while the subject is in a resting state. For example, 12-lead electrocardiogram measurement data of 10 seconds sampled 500 times per second can be input.

[0042] When measuring an electrocardiogram, measurement data is obtained over time from single or multiple leads attached to different parts of the body. Accordingly, the electrocardiogram data input into the data input unit (110) may correspond to single or multiple lead data in a time series, from which spatiotemporal ECG characteristics can be obtained.

[0043] This data input unit (110) can receive electrocardiogram data through a network-connected user terminal (200), electrocardiogram measuring equipment, etc., and transmit the data to the data preprocessing unit (120).

[0044] The data preprocessing unit (120) can preprocess the input electrocardiogram data according to a setting method.

[0045] Here, the data preprocessing unit (120) can perform resampling or zero-padding when electrocardiogram data recorded at a different sampling rate or electrocardiogram data with a smaller number of leads is given as input.

[0046] Additionally, the data preprocessing unit (120) can perform various preprocessing processes to remove noise from input data.

[0047] For example, the data preprocessing unit (120) may perform at least one preprocessing process among preprocessing for removing power noise of a set frequency band used by the measuring equipment (e.g., power line noise of the 50 Hz or 60 Hz band) for the input electrocardiogram data, preprocessing for removing baseline wander noise caused by the breathing or movement of the subject of measurement, and preprocessing for removing muscle noise caused by muscle movement.

[0048] The process of removing power noise, baseline fluctuation noise, and muscle noise from these ECG signals can utilize known ECG preprocessing techniques. For example, baseline fluctuation noise removal can utilize methods such as linear filtering and interpolation. Muscle noise can also be processed during the baseline fluctuation noise removal process.

[0049] Additionally, since each preprocessing process is independent of the others, it can be applied individually to the input data, or multiple preprocessing processes can be applied simultaneously.

[0050] The prediction unit (130) can predict the presence or absence of inducible myocardial ischemia in a subject by inputting preprocessed electrocardiogram data of a set time length into a pre-trained artificial intelligence model.

[0051] This prediction unit (130) can derive a probability value for the presence or absence of myocardial ischemia, and can classify the presence or absence of myocardial ischemia, that is, myocardial ischemia presence (Ischemic) or no myocardial ischemia (Non-Ischemic), and provide the prediction result, and can provide the prediction result in the form of a probability value for the presence or absence of inducible myocardial ischemia.

[0052] Unlike doctors who interpret ECGs sequentially and time-dependently, AI models can extract and identify multiple nonlinear and interdependent features from ECGs using a convolutional approach. Furthermore, AI models can identify subtle changes in ECGs that are undetectable by humans.

[0053] Among models that learn ECG signals, models based on a CNN architecture are generally known to perform well. CNNs are commonly applied to image processing, feature extraction, and object identification. In an embodiment of the present invention, a CNN is used to extract both temporal and spatial features from time-series multi-read data.

[0054] FIG. 3 is a diagram exemplarily showing the architecture of an artificial intelligence model according to an embodiment of the present invention.

[0055] As shown in Figure 3, an AI model can be created by stacking layers based on a convolutional neural network (CNN). Furthermore, in addition to this basic AI model structure, the network can be expanded by applying derivative structures such as a residual network (ResNet), an attention mechanism, and a Transformer.

[0056] An AI model can be composed of a total of five convolutional blocks, each consisting of one convolutional layer, one batch normalization layer, one ReLU activation layer, and one dropout layer.

[0057] To prevent overfitting, batch normalization and dropout were applied as normalization techniques. Following five convolutional blocks, an average pooling layer was used to summarize the features. This process was repeated for the subsequent four convolutional blocks, after which a max-pooling layer was used to capture the most activated features within the window. Finally, two fully connected layers and a softmax activation layer were applied to obtain the probability of each class. Here, each class can represent ischemic or non-ischemic myocardial ischemia.

[0058] The control unit (140) can pre-build an artificial intelligence model using a data set including electrocardiogram data collected for each of multiple patients and the presence or absence of myocardial ischemia. Here, the multiple patients can be divided into groups with and without inducible myocardial ischemia.

[0059] At this time, the control unit (140) can pre-train the artificial intelligence model using data obtained by preprocessing the original electrocardiogram data for each patient by the data preprocessing unit (120).

[0060] Additionally, the control unit (140) may acquire data augmented by preprocessing multiple patient-specific electrocardiogram data using a data augmentation technique for robust model learning, and additionally utilize this data to train an artificial intelligence model. Here, the data augmentation process may be performed in the data preprocessing unit (120).

[0061] Data augmentation methods can include techniques such as removing noise from input data, adding Gaussian noise, time-shifted signal augmentation to minimize the influence of the absolute position of the signal, channel randomization signal augmentation to limit the influence of lead omission or placement order, and random masking signal augmentation to limit the influence of transient signal omission or mismeasurement.

[0062] The patient selection criteria for the two groups used to train the artificial intelligence model are as follows.

[0063] FIG. 4 is a diagram showing an example of selecting two groups of patients (a group with myocardial ischemia and a group without myocardial ischemia) that provide data necessary for learning an artificial intelligence model in an embodiment of the present invention.

[0064] The left side shows an example of selection in the group with myocardial ischemia, and the right side shows an example of selection in the group without myocardial ischemia.

[0065] First, the group with provable myocardial ischemia can be targeted by patients who underwent percutaneous coronary intervention (PCI) excluding patients who underwent PCI for myocardial infarction, received PCI in addition to drug therapy (Guideline-Directed Medical Therapy (GDMT), were diagnosed with silent myocardial ischemia, stable angina, or unstable angina by a cardiologist, and had an electrocardiogram (ECG) recorded within 2 weeks of PCI.

[0066] More specifically, referring to the left side of Figure 4, 66,730 patients who underwent coronary angiography (CAG) at the Cardiovascular Center of Seoul National University Hospital (SNUH) from October 2004 to October 2020 were first selected. From these, patients who underwent percutaneous coronary intervention (PCI) due to silent ischemia, stable angina, or unstable angina were selected. Furthermore, patients who underwent guideline-directed medical therapy (GDMT) and PCI were further selected. Finally, patients with available electrocardiogram data within two weeks prior to PCI were selected. ECG data from the date closest to the PCI examination date can be used for model training. Through this process, 6,070 patients were ultimately selected from the group with myocardial ischemia.

[0067] Next, the group without myocardial ischemia can be targeted to patients who have coronary artery stenosis = 0% and coronary artery calcium score (CACS) ≤ 100 points in all coronary arteries on coronary CT angiography, and the interval between the time of CCTA and the time of electrocardiogram (ECG) observation is less than 2 weeks, and who have no history of percutaneous coronary intervention or bypass surgery, myocardial infarction (MI), or other diseases that may clinically cause ischemia or inducible ischemia (e.g., moderate or severe valvular disease, hypertrophic cardiomyopathy, variant angina, etc.).

[0068] This process is clearly illustrated in the right-hand figure of Figure 4. The group without myocardial ischemia was sampled from 2003 to 2011 at two hospitals: SNUH and SNUH Healthcare System Gangnam Center Hospital. After the filtering process described above, 7,346 patients were ultimately selected for the group without myocardial ischemia. ECG data observed on the date closest to the CCTA were used for model training.

[0069] The data used for model training was directly evaluated by medical professionals with the highest level of expertise in each field, and it is distinguished by the fact that it is a large-scale data set comprehensively judged and labeled by a cardiology specialist. In other words, the data is excellent in both quality and quantity, and such high-quality data cannot be found in open access data, and related papers are lacking. Based on this data, the embodiment of the present invention not only secures data on patients who can be clinically determined to have or not have inducible myocardial ischemia with high reliability, but also, as shown in Figure 4, utilized various test results, patient medical histories, and additional data obtained through subsequent clinical follow-up observations, and underwent a very rigorous data cleansing process.

[0070] The AI ​​model was trained on 10-second 12-lead electrocardiograms acquired at a sampling rate of 500 Hz in the supine position. All data were acquired in XML format using the MUSE ECG system (GE Healthcare, Wauwatosa, WI, USA). ECGs obtained from groups with or without myocardial ischemia were randomly divided into three groups (80% training dataset, 10% validation dataset, and 10% test dataset). Patients who experienced multiple PCI events within the period were excluded from the test dataset to maintain its independence.

[0071] FIG. 5 is a diagram showing the prediction performance of the proposed artificial intelligence model of the present invention, and FIG. 6 is a diagram showing the distribution of prediction results in a test dataset by an artificial intelligence module trained with electrocardiogram data.

[0072] As shown in Fig. 5, the proposed AI model showed an AUROC of 0.90 (95% CI 0.88-0.91), an AUPRC of 0.87 (95% CI 0.84-0.90), and an F1 score of 0.82 in a test dataset using 12-lead ECG. In addition, it showed AUROC of 0.81 (lead I) and 0.78 (lead II) in a test dataset using one lead, and AUROC of 0.84 (95% CI 0.82-0.86) in a test dataset using both leads (i.e., a 6-lead system).

[0073] The AI ​​model generates an output value between 0 and 1 when given ECG signal data as input. The threshold was derived from the internal validation dataset and set at 0.43. However, this threshold can vary depending on the situation. As shown in Figure 6, using this threshold, the model achieved a sensitivity of 83.8% (95% CI 80.6-86.6) and a specificity of 79.6% (95% CI 76.5-82.5).

[0074] External validation included multicenter data from four different datasets, totaling 35,898 patients. For the positive control group, testing was performed on 788 patients (mean age 65.9 ± 9.5 years, 75.3% male) who underwent coronary artery bypass grafting (CABG) due to provoked myocardial ischemia. Based on these electrocardiograms, the model's sensitivity was 86.5% (95% CI 84.0–88.9%). The model's performance was consistent in 1,764 patients who underwent PCI at another cardiovascular center.

[0075] Furthermore, when tested on 12,564 healthy young adults aged 20-35 years (mean age 26.9±4.2 years, 39.4% male) considered free of myocardial ischemia, the proposed AI model showed a specificity of 92.2% (95% CI 91.8-92.7%). The area under the curve (AUROC) and area under the curve (AUPRC) for the combined dataset of positive and negative controls were 0.94 (95% CI 0.94-0.95) and 0.78 (95% CI 0.76-0.81), respectively.

[0076] As a result, we were able to confirm that the proposed AI model can accurately identify subtle differences in the electrocardiograms of patients with inducible myocardial ischemia.

[0077] Figure 7 is a diagram illustrating a method for predicting inducible ischemic heart disease using electrocardiogram-based artificial intelligence according to an embodiment of the present invention. It is assumed that the artificial intelligence model was built through pre-training based on patient big data.

[0078] First, the ischemic heart disease prediction device (100) can receive electrocardiogram data of a set time length observed from a subject (S710). At this time, measurement data including 1-lead, 2-lead, 6-lead, or 12-lead electrocardiograms sampled 500 times per second for 10 seconds can be received.

[0079] Next, the ischemic heart disease prediction device (100) can preprocess the input electrocardiogram data according to a setting method (S720).

[0080] Thereafter, the ischemic heart disease prediction device (100) inputs preprocessed electrocardiogram data into a pre-trained artificial intelligence model to classify the presence or absence of inducible myocardial ischemia in the subject and provides a prediction result (S730). At this time, a probability value for the presence or absence of myocardial ischemia may also be provided.

[0081] According to the present invention as described above, the presence or absence of induced myocardial ischemia can be quickly and easily predicted by inputting electrocardiogram data of a set length of time observed from a subject into an artificial intelligence model.

[0082] In addition, the present invention can be utilized as a screening tool for determining whether myocardial ischemia can be induced from electrocardiogram data.

[0083] While the present invention has been described with reference to the embodiments illustrated in the drawings, these are merely exemplary, and those skilled in the art will understand that various modifications and equivalent alternative embodiments are possible. Therefore, the true scope of technical protection of the present invention should be determined by the technical spirit of the appended claims.

Claims

1. In a device for predicting inducible ischemic heart disease using electrocardiogram-based artificial intelligence, A data input unit for receiving electrocardiogram data of a set time length observed from a subject; A data preprocessing unit that preprocesses the input electrocardiogram data according to the setting method; and A device for predicting inducible ischemic heart disease, comprising a prediction unit that inputs preprocessed electrocardiogram data into a pre-trained artificial intelligence model to predict the presence or absence of inducible myocardial ischemia in a subject.

2. In claim 1, The above data input section, An inducible ischemic heart disease prediction device that receives data including a 1-lead, 2-leads, 6-leads, or 12-leads electrocardiogram of 5 to 15 seconds in length.

3. In claim 1, The above data preprocessing unit, A device for predicting inducible ischemic heart disease that performs at least one of preprocessing among preprocessing for removing power noise of a set frequency band used by a measuring device for input electrocardiogram data, preprocessing for removing baseline wander noise caused by the subject's breathing or movement, and preprocessing for removing muscle noise caused by muscle movement.

4. In claim 1, A device for predicting inducible ischemic heart disease, further comprising a control unit for pre-establishing the artificial intelligence model using a data set including electrocardiogram data collected for each of multiple patients and the presence or absence of myocardial ischemia.

5. In claim 4, The above control unit, A device for predicting inducible ischemic heart disease that trains the artificial intelligence model by additionally utilizing data augmented by a data augmentation technique after preprocessing the electrocardiogram data of the above multiple patients.

6. In claim 4, The above multiple patients are divided into groups with and without inducible myocardial ischemia. The group with the above-mentioned probable myocardial ischemia is a device for predicting probable ischemic heart disease targeting patients who underwent percutaneous coronary intervention (PCI) among patients who excluding those who underwent PCI for myocardial infarction, received PCI in addition to drug treatment (Guideline-Directed Medical Therapy (GDMT), were diagnosed with silent ischemia, stable angina, or unstable angina, and had an electrocardiogram (ECG) record within 2 weeks of PCI.

7. In claim 6, The group without myocardial ischemia is a device for predicting inducible ischemic heart disease targeting patients who, among patients who have coronary artery stenosis = 0% and coronary artery calcium score (CACS) ≤ 100 points in coronary CT angiography (CCTA), have an interval of less than 2 weeks between the time of CCTA and the time of electrocardiogram (ECG) observation, and have no history of previous percutaneous coronary intervention or bypass surgery (revascularization), myocardial infarction (MI), or other diseases that may clinically cause ischemia or inducible ischemia.

8. In a method for predicting inducible ischemic heart disease performed by a device for predicting inducible ischemic heart disease using artificial intelligence based on electrocardiogram, A step of receiving electrocardiogram data of a set time length observed from a subject; A step of preprocessing the input electrocardiogram data according to the setting method; and A method for predicting inducible ischemic heart disease, comprising a step of inputting preprocessed electrocardiogram data into a pre-trained artificial intelligence model to predict the presence or absence of inducible myocardial ischemia in a subject.

9. In claim 8, The step of inputting the above electrocardiogram data is: A method for predicting probable ischemic heart disease, comprising inputting data including 1-lead, 2-leads, 6-leads, or 12-leads electrocardiograms of 5 to 15 seconds in length.

10. In claim 8, The above preprocessing step is, A method for predicting inducible ischemic heart disease, which performs at least one of preprocessing among preprocessing for removing power noise of a set frequency band used by a measuring device for input electrocardiogram data, preprocessing for removing baseline wander noise caused by the subject's breathing or movement, and preprocessing for removing muscle noise caused by muscle movement.

11. In claim 8, A method for predicting inducible ischemic heart disease, further comprising a step of pre-constructing the artificial intelligence model using a data set including electrocardiogram data collected for each of multiple patients and the presence or absence of myocardial ischemia.

12. In claim 11, The steps for pre-building the above artificial intelligence model are: A method for predicting inducible ischemic heart disease, wherein the artificial intelligence model is trained by additionally utilizing data augmented using a data augmentation technique after preprocessing the electrocardiogram data of the above multiple patients.

13. In claim 11, The above multiple patients are divided into groups with and without inducible myocardial ischemia. The group with the above-mentioned inducible myocardial ischemia is a method for predicting inducible ischemic heart disease targeting patients who received PCI in addition to drug treatment (Guideline-Directed Medical Therapy, GDMT), excluding patients who received PCI for myocardial infarction among patients who underwent percutaneous coronary intervention (PCI), and were diagnosed with silent ischemia, stable angina, or unstable angina, and had an electrocardiogram (ECG) record within 2 weeks of PCI.

14. In claim 13, The group without myocardial ischemia is a method for predicting inducible ischemic heart disease targeting patients who, among patients who have coronary artery stenosis = 0% and coronary artery calcium score (CACS) ≤ 100 points in coronary CT angiography (CCTA), have an interval of less than 2 weeks between the time of CCTA and the time of electrocardiogram (ECG) observation, and have no history of coronary artery intervention or bypass surgery (revascularization) in the past, myocardial infarction (MI), or other diseases that may clinically cause ischemia or inducible ischemia.

Images