Variational autoencoder-based electronic device using exercise stress electrocardiogram characteristics, and control method therefor

Abstract

The present disclosure relates to a variational autoencoder-based electronic device using exercise stress electrocardiogram characteristics, and a control method therefor. The electronic device may comprise: a memory in which a process for developing a variational autoencoder-based artificial intelligence model by using exercise stress electrocardiogram characteristics is stored; and a processor for performing an operation according to the process, wherein the processor can acquire first ECG data including exercise stress electrocardiogram information, preprocess the first ECG data, learns the preprocessed first ECG data on the basis of the variational autoencoder, and generate an artificial intelligence model by using the result of learning.

Description

Electronic device based on variational autoencoder using exercise stress electrocardiogram characteristics and its control method

[0001] The present disclosure relates to an electronic device, and more particularly, to the development of an artificial intelligence model based on a variational autoencoder using exercise stress electrocardiogram characteristics, and an electronic device and a control method thereof for predicting vascular remodeling surgery using the same.

[0002] With the recent rapid advancements in medical technology, the integration of medical devices and artificial intelligence is on the rise. The Exercise Stress Test (or Treadmill Test) is a test used to assess cardiac function. This test primarily records the electrical activity of the heart under stress, thereby identifying the presence of heart disease.

[0003] The purpose of the test is to assess coronary artery disease (CAD), assess exercise capacity, evaluate treatment effectiveness, and detect arrhythmias. A normal test result indicates no abnormalities on the electrocardiogram (ECG) during exercise, and increases in blood pressure and heart rate within normal limits. Abnormal results indicate ST-segment changes, arrhythmias, or severe blood pressure changes on the ECG during exercise, which may indicate coronary artery disease or other heart problems.

[0004] Coronary artery disease (CAD) is a condition in which the coronary arteries within the heart become narrowed and blocked. When the degree of blockage causes hematological problems with blood flow, revascularization surgery is performed to widen the coronary arteries.

[0005] However, in the case of conventional technology, exercise stress testing can identify patients with a possibility of coronary artery disease at an early stage by putting a load on the heart, but the accuracy of predicting coronary artery disease is not high, and the amount of collected data is limited, making it difficult to train an artificial intelligence model, which causes inconvenience to users.

[0006] In addition, there was a problem that users felt inconvenienced because the accuracy of predicting coronary artery disease was not high and the amount of data collected was limited, which identified patients with a possibility of coronary artery disease by putting a load on the heart through exercise stress testing, and the accuracy of predicting the need for revascularization surgery was low.

[0007] The present disclosure aims to provide an artificial intelligence model development device capable of learning various characteristics of exercise stress electrocardiograms by making maximum use of collected data when learning exercise stress electrocardiogram data using a variational autoencoder algorithm.

[0008] In addition, the present disclosure aims to provide an artificial intelligence model development device that can identify electrical characteristics of the heart based on electrocardiograms that could not be identified from existing resting ECGs by using exercise electrocardiograms as learning data.

[0009] In addition, the present disclosure aims to provide an artificial intelligence model development device capable of producing interpretable results by utilizing a variational autoencoder with a decoder.

[0010] In addition, the present disclosure aims to provide an artificial intelligence model development device that can flexibly deal with non-learning data using a variational autoencoder algorithm.

[0011] In addition, the present disclosure aims to provide an artificial intelligence model development device that can identify electrical characteristics of the heart based on electrocardiograms that could not be identified from existing resting ECGs by using exercise electrocardiograms as learning data.

[0012] In addition, the present disclosure aims to provide a device for predicting revascularization surgery that determines whether revascularization surgery is necessary by using an artificial intelligence model that utilizes only systolic and diastolic blood pressure during exercise stress electrocardiography and exercise stress electrocardiography tests.

[0013] In addition, the present disclosure aims to provide a revascularization surgery prediction device that can predict patients requiring revascularization surgery and visually determine the need for surgery in a specific area on an exercise stress electrocardiogram.

[0014] In addition, the present disclosure aims to provide a revascularization surgery prediction device that can noninvasively and efficiently select patients with coronary artery disease who are predicted to require revascularization surgery by using only exercise stress electrocardiogram and blood pressure data.

[0015] In addition, the present disclosure aims to provide a device for predicting revascularization surgery by applying artificial intelligence and visually identifying specific patterns indicating the need for revascularization surgery, thereby enabling medical professionals to intuitively understand and evaluate the same.

[0016] The problems to be solved by the present disclosure are not limited to the problems mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art from the description below.

[0017] An electronic device based on a variational autoencoder using exercise stress electrocardiogram characteristics according to the present disclosure for achieving the above-described technical task comprises: a memory storing a process for developing an artificial intelligence model based on a variational autoencoder using exercise stress electrocardiogram characteristics; and

[0018] A processor for performing an operation according to the above process is included, wherein the processor obtains first ECG data including exercise load electrocardiogram information, preprocesses the first ECG data, learns the preprocessed first ECG data based on the variational autoencoder, and generates an artificial intelligence model using the learned result.

[0019] At this time, the processor can obtain data corresponding to the Median beat ECG Tag from XML, which is the original data of exercise electrocardiography, through the input module.

[0020] In addition, the processor includes a preprocessing module, and the preprocessing module can set the length of the median beat ECG to 300, map the stage information, and divide the stage information into stage 1, stage 2, stage 3, stage 4, and stage recovery to build a pre-learning dataset.

[0021] Additionally, the processor may include a variational autoencoder, and the variational autoencoder may include an encoder, a latent space, and a decoder.

[0022] Additionally, the encoder can compress the features of the electrocardiogram into 32 continuous values ​​using a convolutional neural network (CNN) into the latent space.

[0023] Additionally, the decoder can reconstruct an electrocardiogram using 32 consecutive values.

[0024] Additionally, the processor can perform learning by going through a step of reducing the difference between the input ECG and the reconstructed ECG.

[0025] Additionally, the performance of the variational autoencoder may mean the difference between the input ECG and the reconstructed ECG.

[0026] Additionally, the performance of the variational autoencoder can be measured by the Pearson correlation coefficient.

[0027] In addition, the memory further stores a process for predicting revascularization surgery based on the variational autoencoder using the exercise stress electrocardiogram characteristics, and the processor can create a revascularization surgery prediction model using the learned result, obtain the patient's second ECG data, predict the patient's revascularization surgery risk corresponding to the second ECG data using the revascularization surgery prediction model, and output the predicted result.

[0028] In addition, the processor performs learning of the blood vessel remodeling surgery prediction model by using a predetermined number of feature values ​​that are significant feature values ​​among all feature values ​​for each stage and additional blood pressure information, and the additional blood pressure information may include Systolic and Diastolic BP for each stage.

[0029] In addition, the processor confirms the performance of the vascular revascularization surgery prediction model as Balanced Accuracy, and the Balanced Accuracy can be configured by the following mathematical formula.

[0030]

[0031] In addition, the processor can classify an important part among the features of each Stage ECG based on a change in the electrocardiogram waveform, and if there is a latent change according to the change in the electrocardiogram waveform, it can determine it as the important part.

[0032] Additionally, the processor can perform visualization consistency by modifying the SHAP plot in the vascular revascularization surgery prediction model.

[0033] In addition, a method for controlling an electronic device based on a variational autoencoder using exercise stress electrocardiogram characteristics, which is performed by a processor of a device according to the present disclosure for achieving the above-described technical task, may include the steps of: acquiring first ECG data including exercise stress electrocardiogram information; preprocessing the first ECG data; learning the preprocessed first ECG data based on the variational autoencoder; and generating an artificial intelligence model using the learned result.

[0034] According to the present disclosure, when learning exercise stress electrocardiogram data using a variational autoencoder algorithm, various characteristics of exercise stress electrocardiogram can be learned by making maximum use of collected data, and easy-to-understand results that can be used in clinical practice can be provided, thereby improving user convenience.

[0035] In addition, according to the present disclosure, by using exercise electrocardiography as learning data, it is possible to identify electrical characteristics of the heart based on electrocardiograms that could not be identified from existing resting ECGs, thereby enabling more accurate prediction of the patient's condition.

[0036] In addition, according to the present disclosure, there is an advantage in that interpretable results can be derived because a variational autoencoder with a decoder is utilized.

[0037] In addition, according to the present disclosure, the necessity of revascularization surgery can be determined using an artificial intelligence model that utilizes only systolic and diastolic blood pressure during exercise stress electrocardiography and exercise stress electrocardiography tests, thereby improving user convenience.

[0038] In addition, according to the present disclosure, patients requiring revascularization surgery can be accurately distinguished based on the results of an exercise stress test, thereby reducing unnecessary tests, thereby reducing the economic burden and enabling rapid screening of patients with coronary artery disease.

[0039] In addition, according to the present disclosure, there is an advantage in that it is possible to quickly predict patients requiring revascularization surgery and to visually determine the need for surgery in a specific area on an exercise stress electrocardiogram.

[0040] In addition, according to the present disclosure, patients with coronary artery disease who are predicted to require revascularization surgery can be selected in a non-invasive and efficient manner using only exercise stress electrocardiogram and blood pressure data.

[0041] Additionally, according to the present disclosure, artificial intelligence is applied to visually identify specific patterns indicating the need for revascularization surgery, so that medical professionals can intuitively understand and evaluate them.

[0042] The effects of the present disclosure are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below.

[0043] Figure 1 is a configuration diagram of an electronic device based on a variational autoencoder using exercise stress electrocardiogram characteristics according to the present disclosure.

[0044] FIG. 2 is a diagram illustrating a flowchart of a method for developing an artificial intelligence model based on a variational autoencoder using exercise electrocardiogram characteristics according to the present disclosure.

[0045] FIG. 3 is a diagram illustrating a configuration of a processor and a variational autoencoder according to the present disclosure.

[0046] Figure 4 is a diagram illustrating the execution concept of a step-by-step variational autoencoder according to the present disclosure.

[0047] FIG. 5 is a diagram illustrating the learned exercise stress test performance according to the present disclosure.

[0048] FIG. 6 is a diagram illustrating a flowchart of a method for predicting vascular revascularization surgery based on a variational autoencoder using exercise stress electrocardiogram characteristics according to the present disclosure.

[0049] FIG. 7 is a diagram illustrating a flowchart of a method for predicting vascular remodeling surgery according to the present disclosure.

[0050] Figure 8 is a diagram illustrating the prediction results for whether or not revascularization surgery is necessary according to the present disclosure.

[0051] FIG. 9 is a diagram illustrating a SHAP Plot of a vascular revascularization surgery prediction model according to the present disclosure.

[0052] FIG. 10 is a diagram illustrating classification of important parts among 32 features of a stage ECG according to the present disclosure.

[0053] FIG. 11 is a diagram illustrating an example of classifying important parts of stage 1, stage 2, and stage 3 according to the present disclosure.

[0054] FIG. 12 is a diagram illustrating an example of classifying important parts of stage 4, stage recovery, according to the present disclosure.

[0055] FIG. 13 is a diagram illustrating a modified SHAP Plot of a vascular revascularization surgery prediction model according to the present disclosure.

[0056] Figure 14 is a diagram illustrating the results of a prediction of a vascular remodeling surgery according to the present disclosure.

[0057] FIG. 15 is a diagram illustrating a configuration of an electronic device according to the present disclosure.

[0058] Throughout this disclosure, the same reference numerals denote the same components. This disclosure does not describe all elements of the embodiments, and any content that is common in the technical field to which this disclosure pertains or that overlaps between embodiments is omitted. The terms "part, module, element, block" used in the specification may be implemented in software or hardware, and depending on the embodiments, multiple "parts, modules, elements, blocks" may be implemented as a single component, or a single "part, module, element, block" may include multiple components.

[0059] Throughout the specification, when a part is said to be "connected" to another part, this includes not only direct connection but also indirect connection, and indirect connection includes connection via a wireless communication network.

[0060] Additionally, when a part is said to "include" a component, this does not mean that it excludes other components, but rather that it may include other components, unless otherwise specifically stated.

[0061] Throughout the specification, when we say that an element is "on" another element, this includes not only cases where the element is in contact with the other element, but also cases where another element exists between the two elements.

[0062] The terms first, second, etc. are used to distinguish one component from another, and the components are not limited by the aforementioned terms.

[0063] Singular expressions include plural expressions unless the context clearly indicates otherwise.

[0064] The identification codes for each step are used for convenience of explanation and do not describe the order of each step. Each step may be performed in a different order than specified unless the context clearly indicates a specific order.

[0065] The operating principle and embodiments of the present disclosure are described below with reference to the attached drawings.

[0066] The present invention can be implemented not only in a server system but also in various devices capable of performing computational processing and providing results to a user. For example, the present invention can include a computer, a server device, and a mobile terminal, or can be implemented in any one of these forms.

[0067] Here, the computer may include, for example, a notebook, desktop, laptop, tablet PC, slate PC, etc. equipped with a web browser.

[0068] The above server device is a server that processes information by communicating with an external device, and may include an application server, a computing server, a database server, a file server, a game server, a mail server, a proxy server, and a web server.

[0069] The above portable terminal may include, for example, a wireless communication device that ensures portability and mobility, and may include all kinds of handheld-based wireless communication devices such as a PCS (Personal Communication System), GSM (Global System for Mobile communications), PDC (Personal Digital Cellular), PHS (Personal Handyphone System), PDA (Personal Digital Assistant), IMT (International Mobile Telecommunication)-2000, CDMA (Code Division Multiple Access)-2000, W-CDMA (W-Code Division Multiple Access), WiBro (Wireless Broadband Internet) terminal, a smart phone, and a wearable device such as a watch, a ring, a bracelet, an anklet, a necklace, glasses, contact lenses, or a head-mounted device (HMD).

[0070] The artificial intelligence-related functions according to the present disclosure are operated through a processor and memory. The processor may be composed of one or more processors. In this case, one or more processors may be a general-purpose processor such as a CPU, an AP, a DSP (Digital Signal Processor), a graphics-only processor such as a GPU or a VPU (Vision Processing Unit), or an artificial intelligence-only processor such as an NPU. One or more processors control the processing of input data according to predefined operation rules or artificial intelligence models stored in memory. Alternatively, if one or more processors are artificial intelligence-only processors, the artificial intelligence-only processors may be designed with a hardware structure specialized for processing a specific artificial intelligence model.

[0071] The predefined operation rules or artificial intelligence models are characterized by being created through learning. Here, being created through learning means that the basic artificial intelligence model is learned by a learning algorithm using a plurality of learning data, thereby creating a predefined operation rules or artificial intelligence model set to perform a desired characteristic (or purpose). This learning may be performed in the device itself on which the artificial intelligence according to the present disclosure is performed, or may be performed through a separate server and / or system. Examples of the learning algorithm include, but are not limited to, supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning.

[0072] An artificial intelligence model may be composed of multiple neural network layers. Each of the multiple neural network layers has multiple weight values, and performs neural network operations through operations between the operation results of the previous layer and the multiple weights. The multiple weights of the multiple neural network layers may be optimized based on the learning results of the artificial intelligence model. For example, the multiple weights may be updated so that the loss value or cost value obtained from the artificial intelligence model is reduced or minimized during the learning process. The artificial neural network may include a deep neural network (DNN), and examples thereof include, but are not limited to, a convolutional neural network (CNN), a deep neural network (DNN), a recurrent neural network (RNN), a restricted boltzmann machine (RBM), a deep belief network (DBN), a bidirectional recurrent deep neural network (BRDNN), or deep Q-networks.

[0073] The processor can create a neural network, train (or learn) a neural network, perform computations based on received input data, and generate information signals based on the results of the computations, or retrain the neural network.

[0074] Neural networks include CNN (Convolutional Neural Network), RNN (Recurrent Neural Network), perceptron, multilayer perceptron, FF (Feed Forward), RBF (Radial Basis Network), DFF (Deep Feed Forward), LSTM (Long Short Term Memory), GRU (Gated Recurrent Unit), AE (Auto Encoder), VAE (Variational Auto) Encoder), DAE (Denoising Auto Encoder), SAE (Sparse Auto Encoder), MC (Markov Chain), HN (Hopfield Network), BM (Boltzmann Machine), RBM (Restricted Boltzmann Machine), DBN (Deep Belief Network), DCN (Deep Convolutional Network), DN (Deconvolutional Network), DCIGN (Deep Convolutional Inverse Graphics Network), Generative Adversarial Network (GAN), Liquid State Machine (LSM), Extreme Learning Machine (ELM), It will be understood by those skilled in the art that any neural network may be included, including but not limited to ESN (Echo State Network), DRN (Deep Residual Network), DNC (Differentiable Neural Computer), NTM (Neural Turning Machine), CN (Capsule Network), KN (Kohonen Network), and AN (Attention Network).

[0075] According to an exemplary embodiment of the present disclosure, the processor may be configured to perform a process for processing a CNN (Convolution Neural Network) such as GoogleNet, AlexNet, VGG Network, etc., a R-CNN (Region with Convolution Neural Network), a RPN (Region Proposal Network), a RNN (Recurrent Neural Network), a S-DNN (Stacking-based deep Neural Network), a S-SDNN (State-Space Dynamic Neural Network), a Deconvolution Network, a DBN (Deep Belief Network), a RBM (Restricted Boltzman Machine), a Fully Convolutional Network, an LSTM (Long Short-Term Memory) Network, a Classification Network, Generative Modeling, eXplainable AI, Continual AI, Representation Learning, AI for Material Design, BERT, SP-BERT, MRC / QA for natural language processing, Text Analysis, Dialog System, GPT-3, GPT-4, Visual Analytics for vision processing, Visual Understanding, Video Synthesis, ResNet for data intelligence, Anomaly Detection, Prediction, Time-Series Forecasting, Optimization, Various artificial intelligence structures and algorithms, including Recommendation and Data Creation, can be utilized, but are not limited thereto. Hereinafter, embodiments of the present disclosure will be described in detail with reference to the attached drawings.

[0076] Figure 1 is a configuration diagram of an electronic device based on a variational autoencoder using exercise electrocardiogram characteristics according to the present disclosure.

[0077] Referring to FIG. 1, the electronic device (100) includes an input module (110), a sensor module (120), a processor (130), a display module (140), a memory (150), a communication module (160), and a camera module (170).

[0078] The input module (110) acquires data.

[0079] The sensor module (120) senses data.

[0080] The processor (130) performs an artificial intelligence model development method and a vascular remodeling surgery prediction method according to the process.

[0081] That is, the processor (130) may obtain first ECG data including exercise stress electrocardiogram information through the input module (110), preprocess the first ECG data, learn the preprocessed first ECG data based on a variational autoencoder, and create an artificial intelligence model using the learning result. In addition, the processor (130) may obtain first ECG data including exercise stress electrocardiogram information through the input module (110), preprocess the first ECG data, learn the preprocessed first ECG data based on a variational autoencoder, create a revascularization surgery prediction model using the learning result, and obtain second ECG data of the patient through the input module (110) and predict the risk of revascularization surgery of the patient corresponding to the second ECG data using the revascularization surgery prediction model.

[0082] The display module (140) displays a graphic image according to a control command from the processor (130).

[0083] The memory (150) stores at least one process for performing an operation and stores user input and data.

[0084] The communication module (160) transmits and receives data with an external device (200).

[0085] Here, the external device (200) includes an external device such as a smartphone, PC, laptop, tablet PC, etc.

[0086] The camera module (170) captures images of the front.

[0087] The camera module (170) photographs a subject in front according to a control command from the processor (130).

[0088] The processor (130) obtains data corresponding to the Median beat ECG Tag from XML, which is the original data of exercise electrocardiography, through the input module.

[0089] The processor (130) includes a preprocessing module (135) and a variational autoencoder (131).

[0090] The preprocessing module (135) adjusts the length of the median beat ECG to 300 and maps the stage information.

[0091] The preprocessing module (135) constructs a pre-learning dataset by dividing the Stage information into Stage 1, Stage 2, Stage 3, Stage 4, and Stage Recovery as Median beat ECG. A detailed description thereof is provided in Fig. 4.

[0092] The variational encoder (131) includes an encoder (132), a latent space (133), and a decoder (134).

[0093] The encoder (132) uses a convolutional neural network (CNN) to represent the features of the electrocardiogram as 32 continuous values ​​and compresses them into the latent space (133). A detailed description of this is provided in Fig. 3.

[0094] The decoder (134) reconstructs the electrocardiogram using 32 consecutive values.

[0095] The processor (130) performs learning for creating an artificial intelligence model by going through a step of reducing the difference between the input ECG and the reconstructed ECG. A detailed description of this is provided in FIG. 4.

[0096] The performance of a variational autoencoder refers to the difference between the input ECG and the reconstructed ECG.

[0097] The performance of a variational autoencoder is measured by the Pearson correlation coefficient, which is described in detail in Figure 5.

[0098] Additionally, the processor (130) divides the Stage information into Stage 1, Stage 2, Stage 3, Stage 4, Stage early recovery, and Stage mid-term recovery to construct a pre-learning dataset for the vascular revascularization surgery prediction model. A detailed description thereof is provided in FIG. 3.

[0099] The processor (130) performs the vascular revascularization surgery prediction model by using a predetermined number of feature values ​​that are significant feature values ​​among all feature values ​​for each stage and additional blood pressure information.

[0100] The above additional blood pressure information includes systolic and diastolic BP for each stage. A detailed description of this is provided in Figure 4.

[0101] The processor (130) verifies the performance of the above-described vascular revascularization surgery prediction model as Balanced Accuracy, and the Balanced Accuracy is configured by the following mathematical formula 1. A detailed description of mathematical formula 1 is provided in FIG. 4.

[0102] The processor (130) classifies important parts of the features of each Stage ECG based on changes in the electrocardiogram waveform.

[0103] The processor (130) determines whether a latent change exists based on a change in the electrocardiogram waveform, as an important part. A detailed description of this is provided in Fig. 10.

[0104] The processor (130) modifies the SHAP plot in the above-described vascular revascularization surgery prediction model to ensure visual consistency. A detailed description thereof is provided in FIG. 13.

[0105] The processor (130) determines that the patient's ECG is at a high risk of revascularization surgery if it approaches the first line (1411), and determines that the patient's ECG is at a low risk of revascularization surgery if it approaches the second line (1412). A detailed explanation of this is provided in FIG. 14.

[0106] However, the components illustrated in FIG. 1 are not essential for implementing the present invention according to the present disclosure, and thus the present invention described in this specification may have more or fewer components than the components listed above.

[0107] The communication module (160) may include one or more components that enable communication with an external device, and may include, for example, at least one of a broadcast reception module, a wired communication module, a wireless communication module, a short-range communication module, and a location information module.

[0108] The input module (110) is for inputting image information (or signal), audio information (or signal), data, or information input from a user, and may include at least one camera, at least one microphone, and at least one user input unit. Voice data or image data collected by the input module (110) may be analyzed and processed into a user control command.

[0109] The display module (140) displays (outputs) information processed in the present invention. For example, the present invention can display execution screen information of a running application program (e.g., an application), or UI (User Interface) or GUI (Graphical User Interface) information based on such execution screen information.

[0110] The memory (150) can store data supporting various functions of the present invention, programs for the operation of the control unit, input / output data (e.g., music files, still images, videos, etc.), and a plurality of application programs (or applications) driven by the artificial intelligence-based user behavior pattern analysis device (100), data for the operation of the device, and commands. At least some of these application programs can be downloaded from an external server via wireless communication.

[0111] The memory (150) may include at least one type of storage medium among a flash memory type, a hard disk type, an SSD (Solid State Disk type), an SDD (Silicon Disk Drive type), a multimedia card micro type, a card type memory (e.g., SD or XD memory, etc.), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, and an optical disk. In addition, the memory (150) may be a database connected by wire or wirelessly, although separate from the present invention, and may be implemented as a database system.

[0112] The processor (130) may be implemented as at least one core, a memory storing data for an algorithm for controlling the operation of components within the present invention or a program reproducing the algorithm, and at least one processor (not shown) that performs the aforementioned operations using the data stored in the memory. In this case, the memory and the processor may be implemented as separate chips. Alternatively, the memory and the processor may be implemented as a single chip.

[0113] In addition, the processor (130) can control any one or a combination of the components discussed above to implement various embodiments according to the present disclosure described in FIGS. 2 to 15 below.

[0114] At least one component may be added or deleted to correspond to the performance of the components illustrated in Figure 1. Furthermore, it will be readily apparent to those skilled in the art that the relative positions of the components may be altered to correspond to the performance or structure of the system.

[0115] Meanwhile, each component illustrated in FIG. 1 refers to software and / or hardware components such as a Field Programmable Gate Array (FPGA) and an Application Specific Integrated Circuit (ASIC).

[0116] FIG. 2 is a diagram illustrating a flowchart of a method for developing an artificial intelligence model based on a variational autoencoder using exercise electrocardiogram characteristics according to the present disclosure.

[0117] The present invention is performed by an artificial intelligence model development device (100) or a processor (130) of the artificial intelligence model development device (100).

[0118] Referring to FIG. 2, the processor (130) obtains first ECG data including exercise electrocardiogram information through the input module (110) (S210).

[0119] The processor (130) obtains data corresponding to the Median beat ECG Tag from XML, which is the original data of exercise electrocardiography, through the input module (110).

[0120] The processor (130) preprocesses the first ECG data (S220).

[0121] The processor (130) learns the preprocessed first ECG data based on a variational autoencoder (S230).

[0122] The processor (130) creates an artificial intelligence model using the learning results (S240).

[0123] The processor (130) obtains second ECG data including the patient's exercise electrocardiogram information through the input unit (110).

[0124] The processor (130) uses the artificial intelligence model to predict the condition of the patient from the second ECG data.

[0125] The processor (130) controls the display module (140) to output the prediction result.

[0126] FIG. 3 is a diagram illustrating a configuration of a processor and a variational autoencoder according to the present disclosure.

[0127] FIG. 3(a) is a diagram illustrating a processor (130) including a preprocessing module (135) and a variational autoencoder (131).

[0128] Figure 3(b) is a diagram illustrating a variational autoencoder (131).

[0129] Referring to FIG. 3(a), the processor (130) includes a preprocessing module (135) and a variational autoencoder (131).

[0130] The preprocessing module (135) adjusts the length of the median beat ECG to a preset length (e.g., 300) and maps the stage information.

[0131] The preprocessing module (135) divides the above Stage information into Stage 1, Stage 2, Stage 3, Stage 4, and Stage Recovery, and constructs a pre-learning dataset for an artificial intelligence model based on a variational autoencoder using the exercise load electrocardiogram characteristics according to the present disclosure.

[0132] In addition, the preprocessing module (135) divides the Stage information into Stage 1, Stage 2, Stage 3, Stage 4, Stage early recovery, and Stage mid-term recovery, thereby constructing a pre-learning dataset for a variational autoencoder-based vascular revascularization surgery prediction model using exercise stress electrocardiogram characteristics according to the present disclosure.

[0133] Referring to FIG. 3(b), the processor (130) includes a variational autoencoder (131), and the variational autoencoder (131) includes an encoder (132), a latent space (133), and a decoder (134).

[0134] The encoder (132) uses a convolutional neural network (CNN) to represent the features of the electrocardiogram as 32 continuous values ​​and compresses them into the latent space (133).

[0135] The decoder (134) reconstructs the electrocardiogram using 32 consecutive values.

[0136] According to the present invention, there is an advantage in that interpretable results can be derived because a variational autoencoder with a decoder is utilized.

[0137] Figure 4 is a diagram illustrating the execution concept of a step-by-step variational autoencoder according to the present disclosure.

[0138] Referring to Fig. 4, as an execution concept (410) of a step-by-step variational autoencoder, the data preprocessing module (135) performs stage mapping.

[0139] The preprocessing module (135) adjusts the length of the median beat ECG to 300 and maps the stage information.

[0140] Median beat ECG (median electrocardiogram) is a method of condensing multiple ECG cycles into a single representative cycle. This reduces noise and allows for a clearer identification of characteristics.

[0141] The following describes the main steps and concepts of a median beat ECG.

[0142] First, data collection. Multiple electrocardiogram (ECG) cycles are collected. These are typically extracted from ECG signals recorded over a specific period of time.

[0143] Second, QRS detection. The QRS complex (a key feature of the electrocardiogram) is detected from the collected electrocardiogram signal. The QRS complex represents ventricular depolarization and is the central part of the electrocardiogram cycle.

[0144] Third, there's period alignment. All periods are aligned based on the detected QRS complex. This ensures that specific features of each period are in the same location.

[0145] Fourth, the median is calculated. Data from all cycles at the same location are collected and the median is calculated. This process reduces noise and creates a representative cycle by calculating the median at each point in each cycle.

[0146] Fifth, median beat ECG generation. Using the calculated median, a single representative ECG cycle is generated. This cycle represents the original multiple cycles and can be used for more precise analysis or diagnosis.

[0147] Explain the advantages of median beat ECG.

[0148] First, noise reduction. By using the median of multiple cycles, noise can be effectively reduced.

[0149] Second, feature preservation. This simplifies the signal while preserving important features more clearly.

[0150] Explain the disadvantages.

[0151] First, it's complex. Analyzing multiple cycles and calculating the median requires additional computation.

[0152] Second, data dependency. A meaningful median can only be calculated if there are sufficient periods.

[0153] Median beat ECGs are commonly used to improve the analysis and interpretation of electrocardiogram signals. They can be particularly useful in noisy environments.

[0154] The preprocessing module (135) divides the above Stage information into Stage 1, Stage 2, Stage 3, Stage 4, and Stage Recovery to build a pre-learning dataset by dividing the Median beat ECG.

[0155] For Stage 1, the variational autoencoder (131) receives Stage 1 Median Beat ECG as input.

[0156] Stage 1 encoder (132) uses a convolutional neural network (CNN) to represent the features of the electrocardiogram as 32 continuous values ​​and compresses them into the latent space (133).

[0157] Stage 1 decoder (134) reconstructs the electrocardiogram using 32 consecutive values.

[0158] The processor (130) performs learning by going through a step of reducing the difference between the input ECG and the reconstructed ECG.

[0159] Describes risk prediction for revascularization surgery.

[0160] The processor (130) includes a revascularization surgery risk prediction module.

[0161] The input data is the last median beat ECG for each stage, which best represents the status of each stage.

[0162] The processor (130) uses six median beat ECGs from each of stage 1, stage 2, stage 3, stage 4, stage recovery early recovery, and stage mid-term recovery.

[0163] The processor (130) extracts the characteristics of input data by utilizing the encoder module of the previously developed exercise load test variational autoencoder.

[0164] The processor (130) performs model learning and evaluation using 90 significant feature values ​​among a total of 196 feature values ​​for 6 stages and 12 additional blood pressure information (Systolic and Diastolic BP by stage).

[0165] The processor (130) learns a model for predicting the risk of revascularization surgery using the extracted features using XGBoost.

[0166] Here, XGBoost (Extreme Gradient Boosting) is a high-performance machine learning library based on the gradient boosting algorithm. It boasts outstanding accuracy and efficiency. XGBoost is particularly effective on datasets with large amounts of data and complex features.

[0167] The model's performance was confirmed by Balanced Accuracy and AUROC, which were 0.758 and 0.839 (0.799 ~ 0.839), respectively.

[0168] In the case of the exercise stress test interpretation criteria, the Balanced Accuracy and AUROC are 0.756 and 0.804 (0.762~0.804), respectively.

[0169] Balanced Accuracy is composed of the following mathematical formula 1.

[0170]

[0171] It corresponds to the area under the receiver operating characteristic curve (AUROC).

[0172] The processor (130) is a revascularization surgery prediction probability output data between 0 and 1 in the revascularization surgery prediction model, and identifies features with high importance.

[0173] The technical features of the present invention are further described.

[0174] The present invention relates to the development of an exercise stress test (Electrocardiogram)-based model using the variational autoencoder technique. Utilizing this technology, it is possible to develop interpretable artificial intelligence algorithms for various purposes based on EST electrocardiograms.

[0175] Exercise stress electrocardiography (EECG) measures the ECG under stress, allowing it to detect abnormal ECG patterns not visible on a standard 12-electrode ECG. It is used for patients suspected of having coronary artery disease or requiring follow-up. However, the test's accuracy in predicting coronary artery disease is limited, with a sensitivity of 67% and a specificity of 77%. Therefore, utilizing AI technology to identify ECG patterns that humans cannot discern could improve predictive accuracy.

[0176] However, exercise electrocardiography has the disadvantage that it is difficult to successfully train an artificial intelligence model.

[0177] Because exercise stress electrocardiograms have a larger dimension than standard 12-electrode electrocardiograms, more data is needed for successful artificial intelligence learning. However, exercise stress tests are mainly performed on symptomatic patients, so the amount of data collected is limited. Labeling the data for a specific purpose further reduces the amount of data, making it difficult to achieve sufficient learning when developing deep learning algorithms.

[0178] Many artificial intelligence studies are developing fine-tuned artificial intelligence models tailored to a desired purpose by applying a base model that has learned characteristics from a large dataset in advance to a limited dataset.

[0179] In the present invention, we aim to create a basic model for developing an artificial intelligence model utilizing exercise stress electrocardiogram.

[0180] The present invention uses a variational autoencoder algorithm to develop a base model.

[0181] The variational autoencoder algorithm is an unsupervised learning algorithm that learns the characteristics of model data without label-related data loss. It assumes a normal distribution for data characteristics and adds noise to learn the distribution, giving it the advantage of being flexible enough to handle data not used in training.

[0182] According to the present invention, by using exercise electrocardiography as learning data, it is possible to identify electrical characteristics of the heart based on electrocardiography that could not be identified from existing resting ECGs, thereby enabling more accurate prediction of a patient's condition.

[0183] FIG. 5 is a diagram illustrating the learned exercise stress test performance according to the present disclosure.

[0184] Referring to Figure 5, the performance of the variational autoencoder, as the learned exercise stress test performance (510), refers to the difference between the input ECG and the reconstructed ECG. The performance of the variational autoencoder is measured by the Pearson correlation coefficient.

[0185] Describes the performance of exercise stress testing.

[0186] For Stage 1, it is 0.935.

[0187] For Stage 2, it is 0.933.

[0188] For Stage 3, it is 0.934.

[0189] For Stage 4, it is 0.934.

[0190] For Stage Recovery, it is 0.94.

[0191] According to the present invention, the finally learned exercise load test performance is statistically significantly greater than 0.9 for each STAGE, so it can be confirmed that the reconstruction performance is good.

[0192] FIG. 6 is a diagram illustrating a flowchart of a method for predicting vascular revascularization surgery based on a variational autoencoder using exercise stress electrocardiogram characteristics according to the present disclosure.

[0193] The present invention is performed by a vascular remodeling surgery prediction device (100) or a processor (130) of the vascular remodeling surgery prediction device (100).

[0194] Referring to FIG. 6, the processor (130) obtains first ECG data including exercise electrocardiogram information through the input module (110) (S610).

[0195] The processor (130) obtains data corresponding to the Median beat ECG Tag from XML, which is the original data of exercise electrocardiography, through the input module (110).

[0196] The processor (130) preprocesses the first ECG data (S620).

[0197] The processor (130) learns the preprocessed first ECG data based on a variational autoencoder (S630).

[0198] The processor (130) uses the learning results to create a vascular revascularization surgery prediction model (S640).

[0199] The processor (130) obtains second ECG data including the patient's exercise stress electrocardiogram information through the input module (110) (S650).

[0200] The processor (130) uses the above-mentioned revascularization surgery prediction model to predict the risk of revascularization surgery of the patient corresponding to the second ECG data (S660).

[0201] The processor (130) controls the display module (140) to output the prediction result (S670).

[0202] FIG. 7 is a diagram illustrating a flowchart of a method for predicting vascular revascularization surgery according to the present disclosure.

[0203] Referring to FIG. 7, the present invention is performed by a vascular remodeling surgery prediction device (100) or a processor (130) of the vascular remodeling surgery prediction device (100).

[0204] The processor (130) obtains the original exercise electrocardiogram (S710).

[0205] The processor (130) preprocesses data corresponding to the original exercise electrocardiogram (S720).

[0206] The processor (130) predicts the risk of revascularization surgery based on the preprocessed data (S730).

[0207] The processor (130) outputs the prediction result (S740).

[0208] Figure 8 is a diagram illustrating the prediction results for whether or not revascularization surgery is necessary according to the present disclosure.

[0209] Referring to FIG. 8, the processor (130) can display an electrocardiogram (810) by stage to indicate whether revascularization surgery is necessary as a % and a predicted reading result and use the basis for the judgment.

[0210] FIG. 9 is a diagram illustrating a SHAP Plot of a vascular revascularization surgery prediction model according to the present disclosure.

[0211] Referring to FIG. 9, the processor (130) controls the display module (140) to display a SHAP Plot (910) of a vascular revascularization surgery prediction model.

[0212] FIG. 10 is a diagram illustrating classification of important parts among 32 features of a stage ECG according to the present disclosure.

[0213] Figure 10(a) is a drawing illustrating Stage Latent 1 (910).

[0214] Figure 10(b) is a drawing illustrating Stage Latent 13 (920).

[0215] Referring to FIG. 10(a) and FIG. 10(b), the processor (130) classifies important parts among the features of each Stage ECG based on changes in the electrocardiogram waveform.

[0216] The processor (130) determines that a latent change exists according to a change in the electrocardiogram waveform, as an important part.

[0217] For example, the processor (130) checks for a change in the electrocardiogram waveform according to a value between -5 and 5 for one latent, and if there is a change according to the value, the latent is determined to be an important part, and if not, the latent is determined to be an unimportant part.

[0218] Here, negative values ​​indicate a blue ECG waveform, and positive values ​​indicate a red ECG waveform.

[0219] Here, the important part means the part with high priority, and the unimportant part means the part with low priority.

[0220] FIG. 11 is a diagram illustrating an example of classifying important parts of stage 1, stage 2, and stage 3 according to the present disclosure.

[0221] Figure 11(a) is a drawing showing an example of classifying an important part (1110) of Stage 1.

[0222] Fig. 11(b) is a drawing showing an example of classifying an important part (1120) of Stage 2.

[0223] Figure 11(c) is a drawing showing an example of classifying an important part (1130) of Stage 3.

[0224] The processor (130) classifies important and unimportant parts among 32 features of an individual Stage ECG.

[0225] Referring to FIG. 11(a), the processor (130) determines 16 important parts (1110).

[0226] Referring to FIG. 11(b), the processor (130) determines 14 important parts (1120).

[0227] Referring to FIG. 11(c), the processor (130) determines 14 important parts (1130).

[0228] FIG. 12 is a diagram illustrating an example of classifying important parts of stage 4, stage recovery, according to the present disclosure.

[0229] Figure 12(a) is a drawing showing an example of classifying an important part (1210) of Stage 4.

[0230] Figure 12(b) is a drawing showing an example of classifying an important part (1220) of stage recovery.

[0231] The processor (130) classifies important and unimportant parts among 32 features of an individual Stage ECG.

[0232] Referring to FIG. 11(a), the processor (130) determines 12 important parts (1210).

[0233] Referring to FIG. 11(b), the processor (130) determines 17 important parts (1220).

[0234] Therefore, for the final revascularization surgery prediction model, the processor (130) uses 90 latent values ​​out of 196 latents of 6 STAGE VAE models (stage 1, stage 2, stage 3, stage 4, stage recovery early recovery, stage mid-term recovery) and additional blood pressure information (Systolic, Diastolic BP by Stage: 12).

[0235] FIG. 13 is a diagram illustrating a modified SHAP Plot of a vascular revascularization surgery prediction model according to the present disclosure.

[0236] Referring to FIG. 13, the processor (130) performs visualization consistency by modifying the SHAP plot (1310) in the vascular revascularization surgery prediction model.

[0237] For example, the SHAP Plot is modified to enhance the visual effect in the results of the revascularization surgery prediction model.

[0238] The processor (130) performs visualization consistency by converting the SHAP value to a positive number if it is negative.

[0239] Figure 14 is a diagram illustrating the results of a prediction of a vascular remodeling surgery according to the present disclosure.

[0240] Referring to FIG. 14, as the prediction result (1410) of the revascularization surgery, the processor (130) determines that the risk of revascularization surgery is high when the patient's ECG is close to the first line (1411), and determines that the risk of revascularization surgery is low when the patient's ECG is close to the second line (1412). Here, the third line (1413) is used as the reference line, and the fourth line (1414) is the patient's ECG line.

[0241] FIG. 15 is a diagram illustrating a configuration of a vascular revascularization surgery prediction device according to the present disclosure.

[0242] Referring to FIG. 15, the present invention includes a device (1600). The device (1500) may include a memory (1502), a processor (1503), a transceiver (1504), and a peripheral device (1501). In addition, as an example, the device (1500) may further include other configurations and is not limited to the above-described embodiment.

[0243] More specifically, the device (1500) of FIG. 15 may be an exemplary hardware / software architecture, such as an NDN device, an NDN server, a content router, etc. In this case, as an example, the memory (1502) may be a non-removable memory or a removable memory. In addition, as an example, the peripheral device (1501) may include a display, GPS, or other peripheral devices, and is not limited to the above-described embodiment.

[0244] In addition, as an example, the above-described device (1500) may include a communication circuit such as the transceiver (1504), and may perform communication with an external device based thereon.

[0245] Additionally, as an example, the processor (1503) may be at least one of a general-purpose processor, a digital signal processor (DSP), a DSP core, a controller, a microcontroller, ASICs (Application Specific Integrated Circuits), FPGA (Field Programmable Gate Array) circuits, any other type of integrated circuit (IC), and one or more microprocessors associated with a state machine. In other words, it may be a hardware / software configuration that performs a control role for controlling the above-described device (1500).

[0246] At this time, the processor (1503) may execute computer-executable instructions stored in the memory (1502) to perform various essential functions of the present invention. For example, the processor (1503) may control at least one of signal coding, data processing, power control, input / output processing, and communication operations. In addition, the processor (1503) may control the physical layer, the MAC layer, and the application layers. In addition, for example, the processor (1503) may perform authentication and security procedures in the access layer and / or the application layer, and is not limited to the above-described embodiment.

[0247] For example, the processor (1503) can communicate with other devices via the transceiver (1504). For example, the processor (1503) can control a node to communicate with other nodes via a network through the execution of computer-executable instructions. That is, the communication performed in the present invention can be controlled. For example, the other nodes can be an NDN server, a content router, and other devices. For example, the transceiver (1504) can transmit an RF signal via an antenna and transmit the signal based on various communication networks.

[0248] In addition, as an example, MIMO technology, beamforming, etc. can be applied as antenna technology, and are not limited to the above-described embodiment. In addition, the signal transmitted and received through the transceiver (1504) can be modulated and demodulated and controlled by the processor (1503), and are not limited to the above-described embodiment.

[0249] The various embodiments of the present disclosure are not intended to list all possible combinations but rather to illustrate representative aspects of the present disclosure, and the matters described in the various embodiments may be applied independently or in combinations of two or more.

Claims

1. Memory storing the process for developing an artificial intelligence model based on a variational autoencoder using the characteristics of exercise stress electrocardiogram; and Includes a processor that performs operations according to the above process, The above processor, Acquire first ECG data including exercise stress electrocardiogram information, Preprocessing the above first ECG data, Learning the above preprocessed first ECG data based on the variational autoencoder, Using the above learned results to create an artificial intelligence model, An electronic device based on variational autoencoder using exercise stress electrocardiogram characteristics.

2. In paragraph 1, The above processor, Obtaining data corresponding to the Median beat ECG Tag from XML, which is the original data of exercise electrocardiography, through the above input module. An electronic device based on variational autoencoder using exercise stress electrocardiogram characteristics.

3. In paragraph 1, The above processor includes a preprocessing module, The above preprocessing module, Set the length of the median beat ECG to 300 and map the stage information, The above Stage information is divided into Stage 1, Stage 2, Stage 3, Stage 4, and Stage Recovery to build a pre-learning dataset by dividing the Median beat ECG. An electronic device based on variational autoencoder using exercise stress electrocardiogram characteristics.

4. In paragraph 1, The above processor comprises a variational autoencoder, The above variational autoencoder includes an encoder, a latent space, and a decoder. An electronic device based on variational autoencoder using exercise stress electrocardiogram characteristics.

5. In paragraph 4, The above encoder, Using a convolutional neural network (CNN), the features of the electrocardiogram are expressed as 32 continuous values ​​and compressed into the latent space. An electronic device based on variational autoencoder using exercise stress electrocardiogram characteristics.

6. In paragraph 5, The above decoder, Reconstructing an electrocardiogram using 32 consecutive values, Artificial intelligence model development device.

7. In paragraph 1, The above processor, Learning is performed by going through a step of reducing the difference between the input ECG and the reconstructed ECG. An electronic device based on variational autoencoder using exercise stress electrocardiogram characteristics.

8. In paragraph 1, The performance of the above variational autoencoder is Meaning the difference between the input ECG and the reconstructed ECG, An electronic device based on variational autoencoder using exercise stress electrocardiogram characteristics.

9. In paragraph 1, The performance of the above variational autoencoder is As measured by the Pearson correlation coefficient, An electronic device based on variational autoencoder using exercise stress electrocardiogram characteristics.

10. In paragraph 1, The above memory further stores a process for predicting vascular revascularization surgery based on the variational autoencoder using the exercise stress electrocardiogram characteristics. The above processor, Using the above learned results, a prediction model for revascularization surgery is created, Obtain the patient's second ECG data, Using the above revascularization surgery prediction model, the risk of revascularization surgery of the patient corresponding to the second ECG data is predicted, Outputting the predicted result above, An electronic device based on variational autoencoder using exercise stress electrocardiogram characteristics.

11. In paragraph 10, The above processor performs learning of the blood vessel remodeling surgery prediction model by using a predetermined number of feature values ​​that are significant feature values ​​among all feature values ​​for each stage and additional blood pressure information, The above additional blood pressure information includes Systolic and Diastolic BP for each individual stage. An electronic device based on variational autoencoder using exercise stress electrocardiogram characteristics.

12. In the first paragraph, the processor, The performance of the above vascular revascularization surgery prediction model was confirmed with Balanced Accuracy, The above Balanced Accuracy is composed of the following mathematical formula: An electronic device based on variational autoencoder using exercise stress electrocardiogram characteristics.

13. In the 10th paragraph, the processor, Among the features of each Stage ECG, the important parts are classified based on the changes in the electrocardiogram waveform. If there is a latent change according to the change in the above electrocardiogram waveform, it is determined as the important part. An electronic device based on variational autoencoder using exercise stress electrocardiogram characteristics.

14. In the 10th paragraph, the processor In the above vascular revascularization surgery prediction model, SHAP plot is modified to perform visualization consistency. An electronic device based on variational autoencoder using exercise stress electrocardiogram characteristics.

15. A method for controlling an electronic device based on a variational autoencoder using exercise load electrocardiogram characteristics, performed by a processor of the device, A step of acquiring first ECG data including exercise stress electrocardiogram information; A step of preprocessing the first ECG data; A step of learning the preprocessed first ECG data based on the variational autoencoder; and A step of creating an artificial intelligence model using the learned results is included. Control method of electronic device based on variational autoencoder using exercise stress electrocardiogram characteristics.

Images