System for Predicting Paroxysmal Atrial Fibrillation by using Artificial Intelligence

Abstract

A paroxysmal atrial fibrillation prediction system using AI is disclosed. The present embodiment provides an AI-based paroxysmal atrial fibrillation prediction system that uses artificial intelligence technology and short-duration electrocardiograms to automatically predict the probability of atrial fibrillation occurrence and provides a screening service that classifies risk levels based on probability and provides this information to medical staff, thereby enabling preparation for early response to cardiac abnormalities and enhancing the efficiency and reliability of medical service provision.

Description

Technology Field

[0001] One embodiment of the present invention relates to a paroxysmal atrial fibrillation prediction system using AI. Background Technology

[0002] The following description merely provides background information related to the present embodiment and does not constitute prior art.

[0003] Technology exists for a remote medical service system for health checkups and disease prediction utilizing a conventional medical AI engine (KR 10-2495086 B1). It measures the examinee's biometric information, receives the data, and provides a service that performs screenings and predicts diseases through AI technology. However, the conventional technology lacks specific services for specialized diseases such as atrial fibrillation, and specific programs for service channels have not been developed. There is a need for specialized and specific screening services, such as a service that provides a system to calculate the probability of atrial fibrillation occurring from an electrocardiogram that appears normal during a screening.

[0004] As atrial fibrillation is found in one-third of ischemic strokes, screening methods and early diagnosis to detect it are necessary to prevent stroke.

[0005] Therefore, there is a need for an active electrocardiogram screening service that can predict the potential onset of atrial fibrillation early and provide the results to medical professionals to support their early response decisions. The problem to be solved

[0006] The present embodiment aims to provide an AI-based paroxysmal atrial fibrillation prediction system that uses artificial intelligence technology and short-term electrocardiograms to automatically predict the probability of atrial fibrillation occurrence and provides a screening service that classifies risk levels based on probability and provides this information to medical staff, thereby enabling preparation for early response to cardiac abnormalities and enhancing the efficiency and reliability of medical service provision. means of solving the problem

[0007] According to one aspect of the present embodiment, a paroxysmal atrial fibrillation prediction system is provided, comprising: a data acquisition device for acquiring single-lead electrocardiogram data of a plurality of people as identification data; a data analysis device for generating preprocessed identification data by preprocessing the single-lead electrocardiogram data, extracting an R peak from the preprocessed identification data, generating an R peak pattern by patterning the R peak, and outputting a potential atrial fibrillation occurrence probability value based on the preprocessed identification data, the R peak, the R peak pattern, and label data; and a result monitoring device for monitoring the patient's condition based on the paroxysmal atrial fibrillation occurrence probability value. Effects of the invention

[0008] As explained above, according to the present embodiment, by using artificial intelligence technology and a short-term electrocardiogram, a service is provided that automatically predicts the probability of atrial fibrillation occurrence, and a screening service is provided to medical staff that classifies risk levels based on the probability and provides this information, thereby enabling preparation for early response to cardiac abnormalities and increasing the efficiency and reliability of medical service provision. Brief explanation of the drawing

[0009] FIG. 1 is a diagram illustrating biosignal data processing in the field of biosignal processing according to the present embodiment. FIG. 2 is a diagram showing the P, Q, R, S, and T waves (P wave, QRS complex, T wave) and characteristic indicators of the electrocardiogram according to the present embodiment. Figure 3 is a diagram showing a paroxysmal atrial fibrillation prediction system using AI according to the present embodiment. FIG. 4 is a diagram showing a paroxysmal atrial fibrillation identification device according to the present embodiment. FIG. 5 is a diagram showing the learning process of a paroxysmal atrial fibrillation identification device according to the present embodiment. Specific details for implementing the invention

[0010] Hereinafter, the present embodiment will be described in detail with reference to the attached drawings.

[0011] FIG. 1 is a diagram illustrating biosignal data processing in the field of biosignal processing according to the present embodiment.

[0012] The data analysis device (320) performs an electrocardiogram to diagnose heart diseases such as arrhythmia, angina pectoris, myocardial infarction, and cardiac hypertrophy by recording electrical signals generated during a heartbeat as waveforms by current.

[0013] The data analysis device (320) according to the present embodiment can be applied to one-dimensional (1D) biosignal data processing in the field of biosignal processing. The data analysis device (320) divides the P waveform, QRS-complex (N,S,V), T waveform, and noise waveform included in the electrocardiogram waveform into waveform units.

[0014] The data analysis device (320) classifies the electrocardiogram waveform into one of N (Normal beat), S (Supraventricular ectopic beat), V (Ventricular ectopic beat), F (Fusion beat), and Q (Unknown beat).

[0015] FIG. 2 is a diagram showing the P, Q, R, S, and T waves (P wave, QRS complex, T wave) and characteristic indicators of the electrocardiogram according to the present embodiment.

[0016] An electrocardiogram (ECG) waveform appears as a sequence of beats, which can be broadly classified into the normal beat (N), supraventricular beat (S), and ventricular beat (V). A single beat in an ECG waveform fundamentally includes the P wave, QRS wave, and T wave.

[0017] The data analysis device (320) detects the P wave, Q wave, R wave, S wave, and T wave included in the electrocardiogram waveform and classifies the normal beat (N), supraventricular beat (S), and ventricular beat (V). The data analysis device (320) performs localization of the P wave, Q wave, R wave, S wave, and T wave and outputs various characteristic indicators of the electrocardiogram.

[0018] The data analysis device (320) distinguishes the PR interval, QRS interval, QT interval, ST segment, and RR interval for the input electrocardiogram waveform.

[0019] The data analysis device (320) can classify beats based on the characteristic information of P waveform, Q waveform, R waveform, S waveform, and T waveform based on PR interval, QRS interval, QT interval, ST segment, and RR interval.

[0020] The data analysis device (320) classifies heartbeats into normal beats (N), supraventricular beats (S), and ventricular beats (V). When reading an electrocardiogram waveform, the data analysis device (320) can detect abnormal conditions based on localization and classification information.

[0021] The data analysis device (320) detects abnormal conditions (arrhythmia, abnormal beats (S, V), ST, QTc, etc.) using values ​​obtained by performing localization on the electrocardiogram waveform.

[0022] The data analysis device (320) performs classification on the electrocardiogram waveform to classify the heart beat into a normal beat (N), a supraventricular beat (S), and a ventricular beat (V), and detects arrhythmia.

[0023] The data analysis device (320) applies a segmentation technique to the electrocardiogram waveform to identify each section of the P waveform, Q waveform, R waveform, S waveform, T waveform, and fibrillation waveform included in the electrocardiogram waveform. Based on each section, the data analysis device (320) classifies the heart beat into one of N (Normal beat), S (Supraventricular ectopic beat), V (Ventricular ectopic beat), F (Fusion beat), and Q (Unknown beat).

[0024] Figure 3 is a diagram showing a paroxysmal atrial fibrillation prediction system using AI according to the present embodiment.

[0025] The AI-based paroxysmal atrial fibrillation prediction system according to the present embodiment receives electrocardiogram data through an electrocardiogram device, performs artificial intelligence analysis for atrial fibrillation identification, and transmits and monitors the analysis results.

[0026] The AI-based paroxysmal atrial fibrillation prediction system according to the present embodiment can be applied to a service capable of identifying the occurrence of various arrhythmias in electrocardiogram data collected through a single-lead normal electrocardiogram device, as well as atrial fibrillation.

[0027] The AI-based paroxysmal atrial fibrillation prediction system according to the present embodiment provides a screening service that identifies the occurrence of paroxysmal atrial fibrillation early, classifies the patient's condition according to risk level, predicts signs of atrial fibrillation that are difficult for medical staff to visually confirm, and supports medical staff in making rapid early response decisions regarding atrial fibrillation.

[0028] The AI-based paroxysmal atrial fibrillation prediction system according to the present embodiment identifies the occurrence of paroxysmal atrial fibrillation in a normal patient using electrocardiogram data obtained through a single-lead electrocardiogram patch or a simple electrocardiogram test, thereby reducing patient discomfort associated with electrocardiogram measurement and rapidly predicting atrial fibrillation by lightweighting the prediction model.

[0029] The AI-based paroxysmal atrial fibrillation prediction system (300) according to the present embodiment includes a data collection device (310), a data analysis device (320), and a result monitoring device (330). The components included in the AI-based paroxysmal atrial fibrillation prediction system (300) are not necessarily limited thereto.

[0030] The data collection device (310) collects electrocardiogram data. The data collection device (310) collects electrocardiogram data output as a result of performing a single-lead electrocardiogram patch or a simple electrocardiogram test at a hospital. The data collection device (310) is linked with an atrial fibrillation measurement device installed at the hospital.

[0031] The data collection device (310) transmits electrocardiogram data to the data analysis device (320) in a cloud or on-premises format. The data collection device (310) acquires single-lead electrocardiogram data for multiple people as identification data.

[0032] The data analysis device (320) outputs the analysis results obtained by performing an analysis on the electrocardiogram data. The data analysis device (320) outputs the analysis results obtained by analyzing the electrocardiogram data received from the data collection device (310) based on artificial intelligence. The data analysis device (320) uses a model that identifies the probability of occurrence of paroxysmal atrial fibrillation by utilizing normal waveform electrocardiogram data that has identified paroxysmal atrial fibrillation.

[0033] The data analysis device (320) generates a result by calculating a probability value of paroxysmal atrial fibrillation occurrence as a percentage from 0 to 100. The data analysis device (320) generates preprocessed identification data by preprocessing single-lead electrocardiogram data received from the data collection device (310), extracts an R peak from the preprocessed identification data, generates an R peak pattern by patterning the R peak, and outputs a potential probability value of atrial fibrillation occurrence based on the preprocessed identification data, R peak, R peak pattern, and label data.

[0034] The result monitoring device (330) monitors the patient's condition based on the probability value of paroxysmal atrial fibrillation occurrence. The result monitoring device (330) is linked with a terminal installed in a hospital. The result monitoring device (330) notifies the patient's condition based on the probability value of paroxysmal atrial fibrillation occurrence received from the data analysis device (320). The result monitoring device (330) is preferably installed in a hospital, but is not necessarily limited thereto. The result monitoring device (330) can check the diagnosis classification according to the probability value of atrial fibrillation occurrence and risk level included in the analysis results using a monitoring program.

[0035] The result monitoring device (330) can assist in the diagnosis of medical staff. The result monitoring device (330) can output high risk (e.g., 100–71%), inter-risk (e.g., 70–41%), and low risk (e.g., 40–0%) based on the probability value of paroxysmal atrial fibrillation received from the data analysis device (320).

[0036] The result monitoring device (330) outputs the result in a report format by adding medical staff opinions based on the probability value of paroxysmal atrial fibrillation occurrence. The result monitoring device (330) outputs whether the patient has atrial fibrillation based on the probability value of paroxysmal atrial fibrillation occurrence, classifying it into High Risk, Interim Risk, and Low Risk.

[0037] The result monitoring device (330) converts the probability value of paroxysmal atrial fibrillation occurrence into a percentage and then classifies the percentage into a preset range. If the percentage falls within a preset first range (e.g., 100–71%), the result monitoring device (330) classifies the occurrence of atrial fibrillation in the patient as high risk and outputs it. If the percentage falls within a preset second range (e.g., 70–41%), the result monitoring device (330) classifies the occurrence of atrial fibrillation in the patient as medium risk and outputs it. If the percentage falls within a preset third range (e.g., 40–0%), the result monitoring device (330) classifies the occurrence of atrial fibrillation in the patient as low risk and outputs it.

[0038] The result monitoring device (330) outputs a report in the form of a report by adding a medical opinion to at least one of the probability value of paroxysmal atrial fibrillation occurrence, high risk, medium risk, and low risk.

[0039] FIG. 4 is a diagram showing a paroxysmal atrial fibrillation identification device according to the present embodiment.

[0040] The data analysis device (320) uses a potential atrial fibrillation identification algorithm to process data for predicting the occurrence of paroxysmal atrial fibrillation by using artificial intelligence technology to find signs of atrial fibrillation that are not currently visible to the naked eye using single-lead electrocardiogram data in the form of a normal sinus rhythm.

[0041] The data analysis device (320) extracts important features from the collected single-lead electrocardiogram data to generate a paroxysmal atrial fibrillation identification algorithm model.

[0042] The data analysis device (320) discovers a new feature, including an R peak pattern, that was not used in conventional atrial fibrillation prediction studies and uses it for model training. The data analysis device (320) uses electrocardiogram data, R peak patterns, and label data to identify whether paroxysmal atrial fibrillation occurs in a normal sinus rhythm.

[0043] The data analysis device (320) uses single-lead electrocardiogram data to generate a lightweight algorithm model with simplified collection and processing, and makes a rapid decision using 30 seconds of measurement data.

[0044] The data analysis device (320) can identify the probability of occurrence of various arrhythmias that can be classified as atrial fibrillation as well as normal electrocardiogram data of a single lead (Lead II).

[0045] The data analysis device (320) can identify potential occurrences of paroxysmal atrial fibrillation using single-lead electrocardiogram data and can be applied to a clinical decision-making system used by medical staff.

[0046] The data analysis device (320) can predict the occurrence of atrial fibrillation in the patient's electrocardiogram in advance using single-lead electrocardiogram data, thereby reducing the patient's discomfort due to ECG measurement and making the prediction model lighter so that atrial fibrillation can be predicted quickly.

[0047] The data analysis device (320) according to the present embodiment includes a training unit (410) for generating a single-lead electrocardiogram data processing and identification algorithm model, and an identification unit (430) for identifying atrial fibrillation.

[0048] The training unit (410) collects electrocardiogram data from an electrocardiogram data set collected using a data collection device (310) and preprocesses the data. The training unit (410) separately learns single-lead electrocardiogram data to generate a potential atrial fibrillation identification model (428).

[0049] The training unit (410) includes a training data collection unit (412), a training data preprocessing unit (414), a training data feature extraction unit (416), a training data set generation unit (418), a deep learning model training unit (422), a model evaluation unit (424), a model optimization unit (426), and a potential atrial fibrillation identification model (428). The components included in the training unit (410) are not necessarily limited thereto.

[0050] The learning data collection unit (412) collects single-lead electrocardiogram (ECG) data from the data collection device (310) as learning data. The single-lead electrocardiogram data is preferably data acquired from the data collection device (310) in the 'Lead II' method, but is not necessarily limited thereto.

[0051] The data collection device (310) generates electrocardiogram data including a picture of the electrical activity of the heart amplified and recorded due to the most commonly used test in the medical industry to diagnose and monitor heart condition.

[0052] Electrocardiogram data is used to evaluate the electrical activity of the heart, detect abnormalities, and identify underlying heart disease. The data collection device (310) checks for symptoms by examining the electrocardiogram graph included in the electrocardiogram data to diagnose atrial fibrillation. The data collection device (310) records single-lead electrocardiogram data collected from the patient.

[0053] The learning data collection unit (412) collects single-lead electrocardiogram data from the data collection device (310). The single-lead electrocardiogram data is data measured in a single-lead manner by the data collection device (310), and each data point represents the electrical activity of the heart.

[0054] The data acquisition device (310) is a technology for measuring the electrical activity of the heart, and measures an electrocardiogram (ECG) from a patient by detecting regular heartbeats and abnormalities. A single-lead electrocardiogram is one of the most basic forms of measuring an electrocardiogram.

[0055] The data acquisition device (310) measures the electrocardiogram using a single-lead method. Single-lead refers to a method of measuring the electrocardiogram using only one electrode. Single-lead is generally a method of measuring the electrocardiogram by attaching electrodes to the chest or arm.

[0056] There are three types of methods for measuring an electrocardiogram with a single lead (Lead I, Lead II, and Lead III).

[0057] 'Lead I' is a method of measuring an electrocardiogram (ECG) by measuring the voltage between the left arm (LA) and the right arm (RA). 'Lead I' measures the ECG horizontally, measuring the voltage difference along the body's horizontal axis. As a standard ECG measurement method, 'Lead I' measures the heart's electrical activity from the front.

[0058] 'Lead II' is a method of measuring an electrocardiogram by measuring the voltage between the right leg (RL) and the left arm (LA). 'Lead II' measures the electrocardiogram diagonally and indicates the vertical electrical activity of the heart. 'Lead II' is one of the most widely used single leads and better captures specific parts of the heart's electrical activity.

[0059] 'Lead III' is a method of measuring an electrocardiogram by measuring the voltage between the right leg (RL) and the left leg (LL). 'Lead III' measures the electrocardiogram diagonally and indicates the vertical electrical activity of the heart. Like 'Lead II', 'Lead III' is used to measure specific parts of the heart.

[0060] The single-lead electrocardiogram measurement method is commonly used in many portable electrocardiogram devices and is suitable for simple and rapid electrocardiogram measurement. In the single-lead electrocardiogram measurement method, electrodes are attached to the patient's chest or arm, and electrical signals are measured by connecting the electrodes to connected equipment. The data acquisition device (310) displays the electrical activity of the heart as a graph based on the electrical signals.

[0061] The training data preprocessing unit (414) generates preprocessed training data by preprocessing single-lead electrocardiogram data. The training data preprocessing unit (414) generates preprocessed training data by performing noise removal, filter normalization, and 30-second segmentation on the training data.

[0062] The training data preprocessing unit (414) removes noise from the training data. The training data preprocessing unit (414) removes unnecessary noise from the training data to obtain an accurate electrocardiogram. The training data preprocessing unit (414) performs filter normalization on the training data. The training data preprocessing unit (414) normalizes the training data to enable comparison of measurement values ​​in various environments.

[0063] The training data preprocessing unit (414) performs 30-second segmentation on the training data. The training data preprocessing unit (414) divides the training data into segments at 30-second intervals to facilitate processing and analysis of each segment.

[0064] The training data preprocessing unit (414) preprocesses single-lead electrocardiogram data and converts it into a format suitable for the model. The training data preprocessing unit (414) generates preprocessed training data by performing noise removal, data normalization, and outlier processing on the single-lead electrocardiogram data.

[0065] The training data preprocessing unit (414) may use techniques such as moving average and filtering to remove noise from single-lead electrocardiogram data. The training data preprocessing unit (414) generates preprocessed training data by standardizing the magnitude of the signal for data normalization of the single-lead electrocardiogram data.

[0066] The training data feature extraction unit (416) extracts R peaks from the preprocessed training data and generates an R peak pattern that patterns the R peaks. The training data feature extraction unit (416) sets 1 to the value corresponding to the R peak among the preprocessed identification data that performs segmentation on the R peak pattern at a preset time unit, and sets 0 to the remaining data.

[0067] The training data feature extraction unit (416) extracts R peak pattern data from the preprocessed training data. The R peak represents an important point indicating electrical activity in the ventricles of the heart. The pattern of the R peak may change when atrial fibrillation occurs. In atrial fibrillation, the P wave is not clearly observed while showing irregular R peak intervals (RR interval) in the electrocardiogram. To identify paroxysmal atrial fibrillation in a normal sinus rhythm, ECG raw data and RR interval data were utilized, as it is expected that patterns of atrial fibrillation that are not visible to the naked eye will be observed.

[0068] During atrial fibrillation, unlike a normal heart rhythm, the interval between R peaks may become irregular or the shape of the R peak may change. Therefore, the occurrence pattern of the R peak is digitized and extracted as a new feature capable of identifying paroxysmal atrial fibrillation. The new feature value capable of identifying paroxysmal atrial fibrillation consists of 0 and 1, and is measured as 1 when the R peak occurs and 0 when it does not occur.

[0069] The training data feature extraction unit (416) extracts meaningful features from the preprocessed training data. The training data feature extraction unit (416) can extract features in the frequency domain, time domain, etc. The training data feature extraction unit (416) can use signal processing techniques and transformation techniques to extract R peaks from the preprocessed training data and to extract R peak patterns.

[0070] The training data set generation unit (418) generates a training data set based on preprocessed training data, R peak patterns, and label data. The training data set generation unit (418) receives label data that is set to 1 if determined to be atrial fibrillation and 0 if not determined to be atrial fibrillation.

[0071] The training data set generation unit (418) generates a training data set by integrating ECG raw data included in the training data, R peak pattern data extracted from the preprocessed training data, and label data items indicating the presence or absence of atrial fibrillation. Each training data contains information on ECG and R peak patterns corresponding to a length of 30 seconds. The training data set generation unit (418) applies the electrocardiogram data and R peak pattern data in the form of a two-dimensional input.

[0072] The training data set generation unit (418) generates a training data set based on preprocessed training data and features. The training data set generation unit (418) can separate collected single-lead electrocardiogram data into training and identification data. For example, the training data set generation unit (418) can allocate 70% for training and 30% for identification. The training data set generation unit (418) mixes the training and identification data randomly to ensure there is no bias.

[0073] The deep learning model training unit (422) trains a potential atrial fibrillation identification model (428) based on a training dataset. The deep learning model training unit (422) defines data inputs and outputs. The deep learning model training unit (422) defines the inputs and outputs of the training dataset and trains the model.

[0074] The deep learning model training unit (422) trains a potential atrial fibrillation identification model (428) using a training data set. The deep learning model training unit (422) uses a selected model from among multiple models, and in the case of a deep learning model, constructs a neural network. The deep learning model training unit (422) applies a learning algorithm through model training and updates the potential atrial fibrillation identification model (428) to minimize the loss function and maximize accuracy.

[0075] The model evaluation unit (424) evaluates the performance of the artificial intelligence model based on performance measurement indicators. The model evaluation unit (424) evaluates the potential atrial fibrillation identification model (428) based on performance measurement indicators.

[0076] The model evaluation unit (424) evaluates the performance of the model based on performance measurement indicators such as accuracy, sensitivity, and specificity. The model evaluation unit (424) evaluates the performance of the model using accuracy, sensitivity, specificity, etc. The model evaluation unit (424) performs training-validation data splitting. The model evaluation unit (424) verifies generalization performance by appropriately splitting the training data and the validation data.

[0077] The model evaluation unit (424) evaluates the performance of the learned potential atrial fibrillation identification model (428) by applying it to test data. The model evaluation unit (424) measures the performance of the model using indicators such as accuracy, precision, recall, and score for the potential atrial fibrillation identification model (428). When measuring the performance of the potential atrial fibrillation identification model (428), the model evaluation unit (424) may use techniques such as cross-validation to prevent overfitting.

[0078] The model optimization unit (426) optimizes the potential atrial fibrillation identification model. The model optimization unit (426) evaluates the learned model and performs optimization for performance improvement and generalization.

[0079] The model optimization unit (426) performs optimization work to improve the performance of the potential atrial fibrillation identification model (428). The model optimization unit (426) may perform optimization on the potential atrial fibrillation identification model (428) using techniques such as hyperparameter tuning, model structure modification, and data augmentation. The model optimization unit (426) finally generates the optimized potential atrial fibrillation identification model (428).

[0080] The potential atrial fibrillation identification model (428) outputs a potential atrial fibrillation occurrence probability value based on preprocessed training data, R peaks, R peak patterns, and label data. The potential atrial fibrillation identification model (428) outputs a potential atrial fibrillation occurrence probability value based on the learned result of comparing the R peak interval of the data set to 1 in the case of atrial fibrillation among the label data with the interval between R peaks included in the R peak pattern.

[0081] The potential atrial fibrillation identification model (428) identifies the location of the value 1 corresponding to the R peak in the preprocessed training data, and calculates the probability of occurrence of paroxysmal atrial fibrillation in the preprocessed training data by comparing the R peak pattern, which is a patterned interval between the locations of the value 1 corresponding to the R peak, with the R peak pattern corresponding to the label data that is determined to be atrial fibrillation and set to 1.

[0082] When the potential atrial fibrillation identification model (428) calculates the probability of occurrence of paroxysmal atrial fibrillation as a percentage and calculates the potential atrial fibrillation occurrence probability value (430), it outputs a value closer to 1 as the probability of occurrence of atrial fibrillation is higher, and outputs a value closer to 0 as the probability of occurrence of atrial fibrillation is lower.

[0083] The identification unit (430) identifies paroxysmal atrial fibrillation using a potential atrial fibrillation identification model (428). The identification unit (430) preprocesses single-lead electrocardiogram data for paroxysmal atrial fibrillation identification, extracts features to generate input data that can be used in the model, and applies it to the potential atrial fibrillation identification model (428) to calculate the probability of occurrence of paroxysmal atrial fibrillation as a percentage value from 0 to 100 to identify atrial fibrillation.

[0084] The identification unit (430) includes an identification data input unit (432), an identification data preprocessing unit (434), an identification data feature extraction unit (436), a potential atrial fibrillation identification model (428), and a potential atrial fibrillation occurrence probability value (430). The components included in the identification unit (430) are not necessarily limited thereto.

[0085] The identification data input unit (432) acquires single-lead electrocardiogram data for a plurality of people from the data collection device (310) as identification data. The single-lead electrocardiogram data is preferably data acquired from the data collection device (310) in the 'Lead II' method, but is not necessarily limited thereto.

[0086] Single-lead electrocardiogram data is data measured in a single-lead manner by a data acquisition device (310), and each data point represents the electrical activity of the heart.

[0087] The identification data preprocessing unit (434) generates preprocessed identification data by preprocessing single-lead electrocardiogram data. The identification data preprocessing unit (434) generates preprocessed identification data by performing noise removal, filter normalization, and pre-set time unit segmentation on the single-lead electrocardiogram data.

[0088] The identification data preprocessing unit (434) generates preprocessed identification data by performing noise removal, filter normalization, and 30-second segmentation on the identification data. The identification data preprocessing unit (434) removes noise from the identification data. The identification data preprocessing unit (434) obtains an accurate electrocardiogram by removing unnecessary noise from the identification data.

[0089] The identification data preprocessing unit (434) performs filter normalization on the identification data. The identification data preprocessing unit (434) normalizes the identification data to enable comparison of measurements in various environments. The identification data preprocessing unit (434) performs 30-second segmentation on the identification data. The identification data preprocessing unit (434) divides the identification data into segments at 30-second intervals to facilitate processing and analysis of each segment.

[0090] The identification data preprocessing unit (434) preprocesses single-lead electrocardiogram data and converts it into a format suitable for the model. The identification data preprocessing unit (434) generates preprocessed identification data by performing noise removal, data normalization, and outlier processing on the single-lead electrocardiogram data. The identification data preprocessing unit (434) may use techniques such as moving average and filtering to remove noise from the single-lead electrocardiogram data. The identification data preprocessing unit (434) generates preprocessed identification data by standardizing the magnitude of the signal for data normalization of the single-lead electrocardiogram data.

[0091] The identification data feature extraction unit (436) extracts R peaks from the preprocessed identification data and generates an R peak pattern that patterns the R peaks. The identification data feature extraction unit (436) sets 1 to the value corresponding to the R peak among the preprocessed identification data that has been segmented at a preset time unit on the R peak pattern, and sets 0 to the remaining data.

[0092] The identification data feature extraction unit (436) extracts multiple R peaks from preprocessed identification data in which segmentation is performed on single-lead electrocardiogram data at preset time units, and generates an R peak pattern that patterns the intervals between the multiple R peaks.

[0093] The R peak represents a significant point indicating electrical activity in the heart's ventricles. The pattern of the R peak can change when atrial fibrillation occurs. In atrial fibrillation, the P wave is not clearly observed, while irregular R-peak intervals (RR intervals) are present on the electrocardiogram.

[0094] To identify paroxysmal atrial fibrillation in normal sinus rhythms, ECG raw data and RR interval data were utilized, as it is expected that patterns of atrial fibrillation invisible to the naked eye will be visible. During atrial fibrillation, unlike a normal heart rhythm, the R peak interval may become irregular or the shape of the R peak may change. Therefore, the occurrence pattern of the R peak is digitized and extracted as a new feature capable of identifying paroxysmal atrial fibrillation. The new feature value capable of identifying paroxysmal atrial fibrillation consists of 0 and 1, measured as 1 when the R peak occurs and 0 when it does not occur.

[0095] The identification data feature extraction unit (436) extracts meaningful features from the preprocessed identification data. The identification data feature extraction unit (436) can extract features in the frequency domain, time domain, etc. The identification data feature extraction unit (436) can use signal processing techniques and transformation techniques to extract R peaks from the preprocessed identification data and to extract R peak patterns.

[0096] The potential atrial fibrillation identification model (428) is applied to single-lead electrocardiogram data to calculate the probability of paroxysmal atrial fibrillation occurrence as a percentage, which is the potential atrial fibrillation occurrence probability value (430). The potential atrial fibrillation identification model (428) outputs the potential atrial fibrillation occurrence probability value based on preprocessed identification data, R peak, R peak pattern, and label data.

[0097] The potential atrial fibrillation identification model (428) outputs a potential atrial fibrillation occurrence probability value based on the learned result compared with the R peak interval of the data set to 1 in the case of atrial fibrillation among the label data, based on the interval between R peaks included in the R peak pattern.

[0098] The potential atrial fibrillation identification model (428) identifies the location of the value 1 corresponding to the R peak in the preprocessed identification data, and calculates the probability of occurrence of paroxysmal atrial fibrillation in the preprocessed identification data by comparing the R peak pattern, which is a patterned interval between the locations of the value 1 corresponding to the R peak, with the R peak pattern corresponding to the label data that is determined to be atrial fibrillation and set to 1.

[0099] When the potential atrial fibrillation identification model (428) calculates the probability of occurrence of paroxysmal atrial fibrillation as a percentage and calculates the potential atrial fibrillation occurrence probability value (430), it outputs a value closer to 1 as the probability of occurrence of atrial fibrillation is higher, and outputs a value closer to 0 as the probability of occurrence of atrial fibrillation is lower.

[0100] The data analysis device (320) identifies paroxysmal atrial fibrillation based on potential atrial fibrillation occurrence probability value (430) and provides the result to a medical professional.

[0101] FIG. 5 is a diagram showing the learning process of a paroxysmal atrial fibrillation identification device according to the present embodiment.

[0102] The data analysis device (320) uses ECG, R peak pattern, and label data to identify whether paroxysmal atrial fibrillation is present in a normal sinus rhythm. The data analysis device (320) newly extracts an R peak pattern that was not previously used and uses it to train a potential atrial fibrillation identification model (428).

[0103] The data analysis device (320) identifies whether paroxysmal atrial fibrillation is present in a normal electrocardiogram based on electrocardiogram data, R peak patterns, label data, and whether atrial fibrillation is present. The data analysis device (320) can generate a lightweight potential atrial fibrillation identification model (428) by using single-lead electrocardiogram data to simplify the collection and processing of electrocardiogram data. The data analysis device (320) can make a rapid determination using only 30-second length electrocardiogram data.

[0104] The data analysis device (320) identifies paroxysmal atrial fibrillation by using only signal data without combining images and signals. The data analysis device (320) extracts an R peak pattern from preprocessed training data.

[0105] The R peak is an important point indicating regular activity in the heart's ventricles. The R peak pattern is the pattern in which the R peak is detected, and this R peak pattern can change when atrial fibrillation occurs.

[0106] The data analysis device (320) converts the R peak occurrence pattern into data and extracts it as a new feature because, unlike a normal heart rhythm during atrial fibrillation, the R peak interval may become irregular or the subsequent shape may change.

[0107] The data analysis device (320) generates new data by setting the interval where no R peak occurs to 0 and the interval where an R peak occurs to 1. The data analysis device (320) outputs a regular R peak pattern as a normal pattern when the electrocardiogram is usually normal, but outputs an R peak pattern with irregular intervals when there is atrial fibrillation or an abnormality. Since the data analysis device (320) determines normality based on the interval between R peaks, it extracts and patterns the R peaks.

[0108] The ECG raw data includes potential difference values. The pattern data also has a pattern at equal intervals. The measurement intervals of the ECG raw data and the R peak pattern data are the same. The label data indicating the presence or absence of atrial fibrillation is output as output data with the value set to 1 if it is atrial fibrillation and set to 0 if it is not atrial fibrillation.

[0109] Label data is output after the training process in the learning model. Even if atrial fibrillation is not visible in the electrocardiogram data, individuals diagnosed with atrial fibrillation are set to 1, and those not diagnosed are set to 0. In the case of paroxysmal atrial fibrillation, which is the early stage of atrial fibrillation, the atrial electrocardiogram may appear normal even after measuring for one hour, but atrial fibrillation may appear after measuring for another hour.

[0110] The data analysis device (320) determines whether paroxysmal atrial fibrillation is present based only on normal electrocardiogram data. The data analysis device (320) analyzes the raw ECG data and R peak pattern data of a person with atrial fibrillation to generate a potential atrial fibrillation identification model (428) that predicts the likelihood of atrial fibrillation in people who have not yet been diagnosed with paroxysmal atrial fibrillation.

[0111] Paroxysmal atrial fibrillation refers to an abnormal electrocardiogram pattern occurring in the atria of the heart. Normally, the atria are the upper part of the heart, and the lower atria are responsible for receiving blood entering the upper part. Atrial fibrillation occurs when the heart's electrical system malfunctions, generally triggered by abnormal activity in the heart's nerve cells.

[0112] Atrial fibrillation can cause rapid and irregular heartbeats. Rapid heartbeats can prevent the heart from functioning efficiently and hinder proper blood circulation. Atrial fibrillation can cause symptoms by lowering blood pressure and preventing blood from being properly supplied to the entire body.

[0113] While the symptoms of atrial fibrillation can vary from person to person, they generally include a feeling of a fast and irregular heartbeat, chest pain or discomfort, dizziness or vertigo, shortness of breath, fatigue, and occasional fainting. Atrial fibrillation can manifest due to underlying heart problems such as heart disease, hypertension, myocardial infarction, or heart valve disease, or lifestyle habits like excessive alcohol consumption, caffeine intake, stress, and smoking can trigger paroxysmal atrial fibrillation.

[0114] The data analysis device (320) uses an electrocardiogram (ECG) to identify paroxysmal atrial fibrillation. Based on the ECG data, the data analysis device (320) extracts specific patterns showing paroxysmal atrial fibrillation or extracts characteristic features of rapid atrial beats or irregular atrial fibrillation.

[0115] The data analysis device (320) predicts intermittent attacks of atrial fibrillation based on normal electrocardiogram data, as paroxysmal atrial fibrillation is often asymptomatic. Normal sinus rhythm refers to the rhythm of a normal electrocardiogram.

[0116] Atrial fibrillation is the most common type of fibrillation that occurs in the atria of the heart. The atria are the upper part of the heart, responsible for receiving blood flowing to the lower part of the heart. Atrial fibrillation refers to a condition caused by the abnormal operation of the heart's electrical system.

[0117] Normal atria contract and relax at precise times to send blood to other parts of the heart, namely the atria. However, when atrial fibrillation occurs, the atria contract abnormally fast and irregularly. Due to atrial fibrillation, the atria may not relax properly, and blood may not flow sufficiently to other parts of the heart.

[0118] Rapid atrial beats reduce the reliability and efficiency of the heart, and can lead to a drop in blood pressure and problems with blood circulation. Rapid atrial beats can cause symptoms and put strain on the heart. Generally, the most common symptom of atrial fibrillation is that the heart beat feels fast and irregular.

[0119] The data analysis device (320) checks how linearly the R peak pattern rises and falls to measure paroxysmal atrial fibrillation. The data analysis device (320) predicts data in which paroxysmal atrial fibrillation has occurred based on normal electrocardiogram data. Since the R peak pattern may appear similar to normal in the case of intermittent atrial fibrillation, the data analysis device (320) measures the R peak over a long period to generate a potential atrial fibrillation identification model (428).

[0120] The data analysis device (320) uses a potential atrial fibrillation identification model (428) to output a potential atrial fibrillation occurrence probability value (430), and recognizes that the closer the probability value is to 1, the closer it is to a diagnosis of atrial fibrillation, and the closer it is to 0, the closer it is to a probability that there is no atrial fibrillation.

[0121] The data analysis device (320) can extract an R peak from single-lead electrocardiogram data, generate an R peak pattern, and additionally reflect the R peak pattern to enable accurate measurement of paroxysmal atrial fibrillation.

[0122] Since paroxysmal atrial fibrillation is a normal pulse in most cases and occurs intermittently, R peaks are patterned to confirm normal atrial function, and this is additionally incorporated into the electrocardiogram data to predict paroxysmal ECG.

[0123] When the data analysis device (320) collects and learns patterns over a long period, it learns the interval in which a minute change in the pattern of the R peak occurring in intermittent paroxysmal atrial fibrillation is detected.

[0124] The data analysis device (320) can determine the possibility that, within the R peak pattern, patterns that are not actually determined to be atrial fibrillation but have subtly patterned R peaks and changes in intervals may subsequently lead to paroxysmal atrial fibrillation. To find the R peak pattern, the data analysis device (320) uses data labeled as 1 and data labeled as 0 to find the R peak pattern through training data and predicts whether there is a high probability of future paroxysmal atrial fibrillation. The data analysis device (320) does not predict future paroxysmal atrial fibrillation based solely on the R peak pattern, but predicts the likelihood of future paroxysmal atrial fibrillation by additionally reflecting the R peak pattern in the ECG and label data.

[0125] The above description is merely an illustrative explanation of the technical concept of the present embodiment, and a person skilled in the art to which the present embodiment belongs would be able to make various modifications and variations within the scope of the essential characteristics of the present embodiment. Accordingly, the present embodiments are intended to explain, not limit, the technical concept of the present embodiment, and the scope of the technical concept of the present embodiment is not limited by these embodiments. The scope of protection of the present embodiment shall be interpreted by the claims below, and all technical concepts within an equivalent scope shall be interpreted as being included within the scope of rights of the present embodiment. Explanation of the symbols

[0126] 310: Data acquisition device 320: Data analysis device 330: Result monitoring device

Claims

Claim 1 A paroxysmal atrial fibrillation prediction system characterized by comprising: a data acquisition device for acquiring single-lead electrocardiogram data of multiple people as identification data; a data analysis device including a potential atrial fibrillation identification model that generates preprocessed identification data by preprocessing the single-lead electrocardiogram data, extracts an R peak from the preprocessed identification data, generates an R peak pattern in which the R peak is patterned into binary data by setting 1 to the value corresponding to the R peak and 0 to the remaining data, and outputs a potential atrial fibrillation occurrence probability value based on the preprocessed identification data in the form of a normal sinus rhythm, the R peak, the R peak pattern, and label data set to 1 if determined to be atrial fibrillation and 0 if not determined to be atrial fibrillation; and a result monitoring device for monitoring the patient's condition based on the potential atrial fibrillation occurrence probability value. Claim 2 A paroxysmal atrial fibrillation prediction system according to claim 1, wherein the result monitoring device outputs whether the patient has developed atrial fibrillation based on the probability value of the occurrence of paroxysmal atrial fibrillation, classifying it into High Risk, Interim Risk, and Low Risk. Claim 3 A paroxysmal atrial fibrillation prediction system according to claim 1, wherein the result monitoring device converts the probability value of paroxysmal atrial fibrillation occurrence into a percentage, divides the percentage into a preset range, outputs whether the patient has atrial fibrillation as having a high risk if the percentage falls within a preset first range, outputs whether the patient has atrial fibrillation as having a medium risk if the percentage falls within a preset second range, and outputs whether the patient has atrial fibrillation as having a low risk if the percentage falls within a preset third range. Claim 4 In paragraph 3, the result monitoring device is characterized by outputting in a report format by adding a medical staff opinion to at least one of the probability value of occurrence of paroxysmal atrial fibrillation, high risk, medium risk, and low risk. Claim 5 A paroxysmal atrial fibrillation prediction system according to claim 1, characterized in that the data collection device is linked with an atrial fibrillation measurement device installed in a hospital. Claim 6 A paroxysmal atrial fibrillation prediction system according to claim 1, wherein the result monitoring device is linked with a terminal installed in a hospital.

Images