Device and method for estimating bio-age using biomarker

Abstract

Disclosed are a device and method for estimating bio-age according to an embodiment. The device for estimating bio-age according to an embodiment comprises: one or more processors; and one or more memories storing instructions executed by the one or more processors. The one or more processors are configured to: extract one or more biomarkers by inputting input data including bio-signals of a user to a first model; and estimate the bio-age of the user on the basis of a second model modeled by using the one or more biomarkers.

Description

Device and method for estimating biological age using biomarkers

[0001] The disclosed embodiments relate to a technique for estimating biological age, and more specifically, to a technique for estimating the biological age of the heart using digital biomarkers extracted from an electrocardiogram.

[0002] [Cross-reference to related applications]

[0003] This application claims priority to Republic of Korea Provisional Patent Application No. 10-2023-0132027, filed October 4, 2023, and Republic of Korea Patent Application No. 10-2024-0126660, filed September 19, 2024, the entire contents of which are incorporated herein by reference.

[0004] Techniques for directly estimating age from health data, including existing electrocardiogram data, focus on estimating and learning a patient's true age. However, these methods run the risk of failing to detect health abnormalities, particularly cardiac abnormalities, that manifest in various ways.

[0005] For example, if the number of diseases a person has increases or their severity increases, the biological age should increase consistently. However, existing methods focus on learning age-related characteristics rather than disease-related characteristics because their main training goal is to estimate age.

[0006] Therefore, for example, if we compare the heart age of a 40-year-old myocardial infarction patient with that of a healthy 50-year-old, the heart age of the 40-year-old will be lower. In other words, existing age estimation methods inherently have the inherent problem of being able to measure age, but failing to provide health-related age indicators, or even producing distorted results.

[0007] Exemplary embodiments of the present invention provide an innovative estimation technique capable of estimating biological age using biomarkers as intermediaries. Specifically, exemplary embodiments of the present invention provide a technique for estimating the biological age of the heart using biomarkers extracted from electrocardiogram signals as intermediaries.

[0008] A biological age estimation device using a biomarker according to one embodiment is a biological age estimation device using a biomarker, comprising: one or more processors; and one or more memories storing instructions executed by the one or more processors, wherein the one or more processors are configured to: input input data including a biological signal of a user into a first model that has been previously learned to extract one or more biomarkers, and input the one or more biomarkers into a second model that has been previously learned to estimate the biological age of the user.

[0009] The above biosignal may include an electrocardiogram signal, and the above bioage may include a bioage of the heart.

[0010] The first model may include an artificial intelligence model that is trained to output one or more corresponding biomarkers when inputting learning data related to an electrocardiogram signal, and the second model may include a constraint regression model that is trained to model some of the output data in the learning process of the first model as independent variables, and select regression coefficients and independent variables that minimize an error from the correct value, and is trained to output a corresponding biological age when inputting a value of the output data.

[0011] The training data of the first model may be configured such that input values ​​including electrocardiogram signals and at least one of anatomical characteristics, physiological characteristics, and pathological characteristics corresponding to the electrocardiogram signals are labeled as pairs of correct values.

[0012] The above anatomical characteristics are characteristics related to cardiac hypertrophy, including at least one of left atrial enlargement (LAE), left ventricular enlargement (LVH), right atrial enlargement (Right Atrial Enlargement), right ventricular hypertrophy (Right Ventricular Hypertrophy), left ventricular mass index, and left atrial volume index, and the above physiological characteristics are characteristics related to systolic function, including left ventricular or right ventricular systolic or diastolic dysfunction, left ventricular ejection fraction, myocardial strain, brain natriuretic peptide (BNP) and pro-brain natriuretic peptide (ProBNP), and characteristics related to pulmonary arterial hypertension, including pulmonary arterial wedge pressure and pulmonary hypertension; And, as characteristics related to cardiac ischemia, at least one of coronary artery calcium score, presence of coronary artery stenosis, presence of cardiac ischemia, and Troponin test results; and, probability of death within a specific period of time; and, as characteristics related to valvular dysfunction, at least one of presence and score of stenosis or regurgitation of the aorta, mitral valve, pulmonary valve, and tricuspid valve.

[0013] The one or more processors may be configured to: extract one or more candidate biomarkers by inputting the user's electrocardiogram signal into a first model, and selecting some of the one or more candidate biomarkers as the one or more biomarkers, thereby extracting the one or more biomarkers.

[0014] The one or more processors may be configured to select some of the one or more candidate biomarkers as the one or more biomarkers based on at least one of a correlation between the one or more candidate biomarkers and the actual age of the user, an effect size of the correlation, and a biological plausibility of the one or more candidate biomarkers.

[0015] The one or more candidate biomarkers may be plural, and the one or more processors may be configured to: cluster the one or more candidate biomarkers into each group based on classification criteria, select one or more representative biomarkers for each group, and input the one or more representative biomarkers into the second model to estimate the biological age.

[0016] The classification criteria may include at least one of similarity and biological or medical relevance between the one or more candidate biomarkers.

[0017] The one or more processors may be configured to visually display on a screen of a device communicating with the processor at least one of: the contribution of the one or more biomarkers in determining the biological age together with the biological age, a description of the contribution, a difference between the user's cardiac biological age and their actual age, a grade corresponding to the difference, a health status according to the grade, and instructions for improving the health status.

[0018] The one or more processors may: perform a linear transformation on the one or more biomarkers so that the one or more biomarkers have a linear correlation with the actual age of the user with a fixed sign, and calculate a correlation between the one or more biomarkers for which the linear transformation has been completed and the actual age.

[0019] The second model may include a constraint regression model, and a first constraint of the constraint regression model may be characterized in that the one or more representative biomarkers exist for each group, and a second constraint of the constraint regression model may be characterized in that the magnitude of the correlation is within a preset range.

[0020] The input data includes at least one of a one-dimensional signal representing the user's electrocardiogram in time series and a two-dimensional signal representing the electrocardiogram as an image, and the first model includes at least one of a vision transformer, a convolutional neural network, and a recurrent neural network, and the one or more processors may be configured to: input the input data to a first-first model configured with at least one of the convolutional neural network and the recurrent neural network when the type of the input data is a one-dimensional signal, and input the input data to a first-second model configured with at least one of the vision transformer and the convolutional neural network when the type of the input data is a two-dimensional signal.

[0021] A method for estimating biological age using a biomarker according to one embodiment is a method performed by a biological age estimation device using a biomarker having one or more processors; and one or more memories storing instructions executed by the one or more processors, the method comprising: a step of inputting input data including a biological signal of a user into a first model that has been previously learned to extract one or more biomarkers; and a step of inputting the one or more biomarkers into a second model that has been previously learned to estimate the biological age of the user.

[0022] The above biosignal may include an electrocardiogram signal, and the above bioage may include a bioage of the heart.

[0023] The first model may include an artificial intelligence model that is trained to output one or more corresponding biomarkers when inputting learning data related to an electrocardiogram signal, and the second model may include a constraint regression model that is trained to model some of the output data in the learning process of the first model as independent variables, and select regression coefficients and independent variables that minimize an error from the correct value, and is trained to output a corresponding biological age when inputting a value of the output data.

[0024] The training data of the first model may be configured such that input values ​​including electrocardiogram signals and at least one of anatomical characteristics, physiological characteristics, and pathological characteristics corresponding to the electrocardiogram signals are labeled as pairs of correct values.

[0025] The above anatomical characteristics are characteristics related to cardiac hypertrophy, including at least one of left atrial enlargement (LAE), left ventricular enlargement (LVH), right atrial enlargement (Right Atrial Enlargement), right ventricular hypertrophy (Right Ventricular Hypertrophy), left ventricular mass index, and left atrial volume index, and the above physiological characteristics are characteristics related to systolic function, including left ventricular or right ventricular systolic or diastolic dysfunction, left ventricular ejection fraction, myocardial strain, brain natriuretic peptide (BNP) and pro-brain natriuretic peptide (ProBNP), and characteristics related to pulmonary arterial hypertension, including pulmonary arterial wedge pressure and pulmonary hypertension; And, as characteristics related to cardiac ischemia, at least one of coronary artery calcium score, presence of coronary artery stenosis, presence of cardiac ischemia, and Troponin test results; and, probability of death within a specific period of time; and, as characteristics related to valvular dysfunction, at least one of presence and score of stenosis or regurgitation of the aorta, mitral valve, pulmonary valve, and tricuspid valve.

[0026] The step of extracting the one or more biomarkers may include the step of inputting the user's electrocardiogram signal into a first model to extract one or more candidate biomarkers; and the step of extracting the one or more biomarkers by selecting some of the one or more candidate biomarkers as the one or more biomarkers.

[0027] The step of extracting the one or more biomarkers may include a step of selecting some of the one or more candidate biomarkers as the one or more biomarkers based on at least one of a correlation between the one or more candidate biomarkers and the actual age of the user, an effect size of the correlation, and a biological plausibility of the one or more candidate biomarkers.

[0028] The one or more candidate biomarkers may be plural, and the method may include: a step of clustering the one or more candidate biomarkers into each group based on classification criteria; a step of selecting one or more representative biomarkers for each group; and a step of inputting the one or more representative biomarkers into the second model to estimate the biological age.

[0029] The classification criteria may include at least one of similarity and biological or medical relevance between the one or more biomarkers.

[0030] The method may include: a step of visually displaying on a screen of the device at least one of the contribution of the one or more biomarkers in determining the biological age together with the biological age, a description of the contribution, a difference between the user's cardiac biological age and actual age, a grade corresponding to the difference, a health status according to the grade, and a guideline for improving the health status.

[0031] The method may further include: performing a linear transformation on the one or more biomarkers so that the one or more biomarkers have a linear correlation with the actual age of the user with a fixed sign; and calculating a correlation between the one or more biomarkers for which the linear transformation has been completed and the actual age.

[0032] The second model may include a constraint regression model, wherein a first constraint of the constraint regression model is that one or more representative biomarkers exist, and a second constraint of the constraint regression model may be that the magnitude of the correlation is within a preset range.

[0033] The input data includes at least one of a one-dimensional signal representing the user's electrocardiogram in time series and a two-dimensional signal representing the electrocardiogram as an image, and the first model includes at least one of a vision transformer, a convolutional neural network, and a recurrent neural network, and the method may include: when the type of the input data is a one-dimensional signal, inputting the input data to a first-first model configured with at least one of the convolutional neural network and the recurrent neural network; and when the type of the input data is a two-dimensional signal, inputting the input data to a first-second model configured with at least one of the vision transformer and the convolutional neural network.

[0034] The disclosed embodiments can extract digital biomarkers from electrocardiograms as an intermediate medium and utilize them to estimate biological age that meaningfully reflects the heart health status, such as the type and severity of heart disease burden, of a specific individual. Specifically, digital biomarkers that can estimate specific anatomical, functional, and disease characteristics of each individual are extracted from electrocardiograms, the extracted digital biomarkers are transformed to have a linear and consistent positive correlation with the biological age of the heart of the individual, and further, the regression coefficients of each of these are grouped into a linear regression equation that limits a specific range, thereby estimating biological age that reflects the health status.

[0035] The disclosed embodiments estimate biological age using a regression model in which extracted biomarkers are modeled as variables, thereby improving the ease and transparency of interpretation of which biomarkers influence biological age.

[0036] The disclosed embodiments visualize the contribution of each digital biomarker in estimating an individual's biological age, thereby providing individualized insights when interpreting the estimated biological age. This provides information on specific improvements and actions that can be taken to reduce an individual's estimated biological age. This enables the provision of personalized medical solutions.

[0037] The disclosed embodiments can provide medical professionals with immediate and actionable clinical information for treatment planning by relating the difference between estimated biological age and chronological age (defined as delta age) to health risks, such as risk of death and risk of comorbidities.

[0038] FIG. 1 is a block diagram illustrating a device for estimating biological age according to one embodiment.

[0039] FIG. 2 is a block diagram of a processor included in a device for estimating biological age according to one embodiment.

[0040] FIG. 3 is an exemplary diagram illustrating the operation of a device for estimating biological age according to one embodiment.

[0041] Figure 4 is a flowchart illustrating a method for estimating biological age according to one embodiment.

[0042] The foregoing description is provided to facilitate a comprehensive understanding of the methods, devices, and / or systems described herein. However, these are merely examples and the present invention is not limited thereto.

[0043] In describing embodiments, if it is determined that a detailed description of a known technology related to the present invention may unnecessarily obscure the gist of an embodiment, the detailed description will be omitted.

[0044] The terms described below are defined based on their functions in the present invention, and may vary depending on the intent or custom of the user or operator. Therefore, their definitions should be based on the overall content of this specification. The terminology used in this detailed description is intended solely to describe specific embodiments and should not be construed as limiting.

[0045] Unless explicitly stated otherwise, singular expressions include the plural. It should be noted that terms such as "first," "second," etc., are used only to distinguish between various components and are not limited by the terminology.

[0046] Meanwhile, the device of the present invention may be entirely hardware, or may have aspects that are partially hardware and partially software. For example, the system for estimating the severity of cognitive impairment in the elderly and each unit included therein in the present specification may collectively refer to a device for transmitting and receiving data of a specific format and content via electronic communication, and software related thereto. In the present specification, terms such as "unit," "module," "server," "system," "device," or "terminal" are intended to refer to a combination of hardware and software driven by the hardware. For example, the hardware herein may be a data processing device including a CPU or other processor. In addition, the software driven by the hardware may refer to a running process, object, executable, thread of execution, program, etc.

[0047] As used herein, the term "biological age" refers to an age that reflects an individual's heart health, including the type and severity of their disease burden. It is distinct from chronological age. For example, comparing the cardiac biological age of a 40-year-old patient with a heart attack and a healthy 50-year-old may result in the 40-year-old's age being lower. This age reflects a state related to heart health, which is different from the chronological age of a 50-year-old.

[0048] FIG. 1 is a block diagram illustrating a device (100) for estimating biological age using a biomarker according to one embodiment.

[0049] Referring to FIG. 1, a device (100) for estimating biological age using a biomarker according to one embodiment includes a sensor (110), a display (120), a memory (130), and a processor (140).

[0050] The sensor (110) can detect the health status of a user coming into contact with a device (100) for estimating biological age using a biomarker, and generate electrical signals and / or data values ​​corresponding to the detected status. In particular, the sensor (110) can measure the electrical and physical status of the heart and generate electrical signals and / or data that can be processed by the processor (140).

[0051] For example, the sensor (110) may include a biosensor (110) capable of measuring an electrocardiogram. At this time, the biosensor (110) may include a bioelectrode that can be attached to the skin to integrally measure an electrocardiogram, and a photoplethysmography sensor (110) that can indirectly measure heart rate and an electrocardiogram through changes in blood flow.

[0052] The display (120) can output image data under the control of the processor (140). The display (120) can visually display various data processed by one or more processors (140). The display (120) can output images previously stored in the memory (130) under the control of the processor (140).

[0053] For example, the display (120) can output text, images, videos, etc. related to the biological age estimated by the processor (140). As a specific example, the display (120) can provide the biological age of the heart estimated by the processor (140), the difference between the biological age and the actual age, and medical guidelines corresponding to the age difference in the form of text, images, and videos.

[0054] The display (120) can be implemented as an LCD (Liquid Crystal Display Panel), an OLED (Organic Light Emitting Diodes), etc., and in some cases, the display (120) can also be implemented as a flexible display (120), a transparent display (120), etc.

[0055] Meanwhile, the type of display (120) is exemplary and is not necessarily limited to a specific type as the above-described example, and any medium capable of visually expressing image data may be used.

[0056] The memory (130) stores various data required by components of the device (100) for estimating biological age using a biomarker. For example, the data may include software (programs (e.g., middleware, applications), etc.) executed by the processor (140), a set of instructions, and input data and / or output data for instructions related thereto.

[0057] At this time, the input data and output data can store the user's basic information (e.g., gender, actual age, weight, height), medical records (e.g., past medical records indicating underlying diseases, existing diseases, etc.), electrocardiogram signals, preprocessing algorithms, artificial intelligence models, methods, etc.

[0058] The processor (140) controls the overall operation of the device (100) for estimating biological age using a biomarker. Specifically, the processor (140) is connected to a sensor (110), a display (120), and / or a memory (130), and can control the overall operation by executing at least one command stored in the memory (130).

[0059] Specifically, the processor (140) can control the device (100) for estimating biological age using a biomarker through a command to perform the following operations: receiving an electrocardiogram signal collected through a sensor (110) or an electrocardiogram signal input from an external device (100), preprocessing the electrocardiogram signal, selecting an artificial intelligence model to be used, extracting a biomarker from the electrocardiogram signal, applying a mathematical transformation to the biomarker, estimating biological age, and providing a health status corresponding to the predicted age.

[0060] The processor (140) may represent any type of processor (140) capable of executing firmware, middleware, software, etc., which are comprised of instructions that cause the processor (140) to perform the described tasks.

[0061] Examples of processors (140) include central processing units, graphics processing units, digital signal processors, display processors, field programmable gate arrays, microprocessors, application-specific integrated circuits (e.g., application-specific integrated circuits (ASICs)), etc.

[0062] FIG. 2 is a block diagram of a processor (140) included in a device (100) for estimating biological age using a biomarker according to one embodiment.

[0063] Referring to FIG. 2, the processor (140) includes a receiving unit (141), a preprocessing unit (142), a first selection unit (143), an extraction unit (144), a verification unit (145), a second selection unit (146), a production unit (147), a modeling unit (148), and an output unit (149).

[0064] In the following embodiments, the receiving unit (141), the preprocessing unit (142), the first selection unit (143), the extraction unit (144), the verification unit (145), the second selection unit (146), the calculation unit (147), the modeling unit (148), and the output unit (149) may be implemented using one or more physically separated devices, or may be implemented by one or more processors (140) or a combination of one or more processors (140) and software, and may not be clearly separated in specific operations, unlike the illustrated example.

[0065] According to one embodiment, the receiver (141) can receive an electrocardiogram signal stored in the memory (130) or an electrocardiogram measured from the sensor (110). At this time, the electrocardiogram signal can be a printed electrocardiogram image or a one-dimensional electrocardiogram signal of single or multiple channels.

[0066] Here, the electrocardiogram image is an image that includes a graph recording the electrical activity of the heart, and can be a two-dimensional image printed on paper or output on a digital screen of the signal output from the electrocardiogram device.

[0067] A one-dimensional electrocardiogram signal of single or multiple channels is a signal that continuously records voltage changes over time corresponding to the electrical activity of the heart, and each channel can be understood as being measured from an independent electrode.

[0068] At this time, it is desirable that the ECG signal be a high-quality ECG signal that satisfies preset conditions. For example, it may be desirable for the ECG signal to satisfy preset sampling rate, duration, quality, source, and noise filtering processing methods.

[0069] For example, an electrocardiogram signal may have a preset sampling rate, for example, 500 Hz. The electrocardiogram signal may have a duration of at least 2.5 seconds, preferably at least 10 seconds, to ensure sufficient measurements. The electrocardiogram signal may be accompanied by an acknowledgment of the source of the data, such as a wearable device, an electrocardiogram monitor, a 12-lead electrocardiogram device, etc. The electrocardiogram signal may be filtered to limit a specific preset frequency or may be data that has been filtered in a specific order.

[0070] According to one embodiment, the preprocessing unit (142) can preprocess the received electrocardiogram signal.

[0071] The preprocessing unit (142) can perform noise removal processing on the received electrocardiogram signal. Specifically, if the input data is a printed image, the preprocessing unit (142) can remove logos, source marks, etc. that are unnecessary for feature extraction in addition to the electrocardiogram waveform. If the input data is a digital signal, the preprocessing unit (142) can apply a bandpass filter or a notch filter to remove noise.

[0072] The preprocessing unit (142) can refine the received electrocardiogram signal. In one embodiment, the preprocessing unit (142) can resize the data size of the received electrocardiogram signal or perform format conversion. For example, the preprocessing unit (142) can resize the size of the image so that it satisfies the size specifications that the first model can process. In this case, the image may be a printed image or data obtained by converting a digital electrocardiogram signal into an image.

[0073] The preprocessing unit (142) can convert the received electrocardiogram signal. According to one embodiment, the preprocessing unit (142) can perform a standardization task. Specifically, the preprocessing unit (142) can convert the electrocardiogram signal into voltage units, or perform resampling and normalization to convert the electrocardiogram signal to have various types of characteristics. According to one embodiment, the preprocessing unit (142) can divide the received electrocardiogram signal into multiple segments. At this time, the preprocessing unit (142) can divide the electrocardiogram signal composed of long time series data into segments continuously and non-overlappingly or overlappingly. According to one embodiment, the preprocessing unit (142) can divide the electrocardiogram signal of multiple channels into segments with fixed intervals using a synchronized sampling method. At this time, the electrocardiogram signal can be divided into segments for the same time interval for each channel. According to one embodiment, the preprocessing unit (142) can generate segments using an asynchronous sampling method. For example, the preprocessing unit (142) can apply a random time offset to each channel segment. The preprocessing unit (142) can apply a random time shift to each channel segment. The preprocessing unit (142) can cycle the time series data of each channel segment.

[0074] The first selection unit (143) can select the type of first model to be used based on the signal expression method of the electrocardiogram signal.

[0075] For example, the first selection unit (143) may select a model composed of at least one of a vision transformer and a CNN as the first model when the electrocardiogram signal is in an analog format, for example, when the input data is an image file in which a record of the electrocardiogram is printed.

[0076] As another example, the first selection unit (143) may select a model composed of at least one of RNN and CNN as the first model when the electrocardiogram signal is in digital format, for example, when the input data is a one-dimensional electrocardiogram signal of single or multiple channels.

[0077] The extraction unit (144) can extract one or more biomarkers from the input electrocardiogram signal using the first model. The extraction unit (144) can extract one or more biomarkers from the input electrocardiogram signal using the first model.

[0078] Here, a biomarker can refer to a characteristic or indicator that can be extracted from an electrocardiogram (ECG) signal and that can identify characteristics related to cardiac health. In other words, a biomarker can refer to a digital marker derived from an ECG signal (an ECG-derived digital marker). Specifically, a biomarker can include anatomical, physiological, and pathological characteristics related to the heart that can be estimated from an ECG signal.

[0079] The first model is a supervised learning model using training data to extract one or more biomarkers in response to the input electrocardiogram signal.

[0080] The training data may be configured such that input values ​​including electrocardiogram signals and at least one feature among anatomical features, physiological features, and pathological features corresponding to the electrocardiogram signals are labeled as pairs of correct values.

[0081] For example, the anatomical characteristic may be a characteristic related to cardiac hypertrophy, and may include at least one of Left Atrial Enlargement (LAE), Left Ventricular Enlargement (LVH), Right Atrial Enlargement, Right Ventricular Hypertrophy, Left Ventricular Mass Index, and Left Atrial Volume Index.

[0082] For example, the physiological characteristics may include characteristics related to systole, including at least one of left or right ventricular systolic or diastolic dysfunction, left ventricular ejection fraction, myocardial strain, brain natriuretic peptide (BNP), and pro-brain natriuretic peptide (ProBNP).

[0083] As another example, the physiological characteristics may include characteristics associated with pulmonary hypertension, including at least one of pulmonary wedge pressure and the presence of pulmonary hypertension. As another example, the physiological characteristics may include characteristics associated with cardiac ischemia, including at least one of coronary artery calcium score, the presence of coronary artery stenosis, the presence of cardiac ischemia, and troponin test results. As another example, the physiological characteristics may include the probability of death within a specific time period.

[0084] For example, the pathological characteristics may include characteristics associated with valvular dysfunction, including at least one of the presence and score of aortic, mitral, pulmonary, or tricuspid stenosis or regurgitation.

[0085] The first model can be trained using augmented training data to enhance model robustness. The training data can be augmented by adding noise (e.g., Gaussian noise, random noise). The training data can also be augmented by adding temporal distortion. For example, the training data can be randomly stretched or compressed along the time axis to introduce temporal distortion. The training data can also be augmented through random scaling. For example, the training data can be augmented by multiplying the magnitude of the electrocardiogram signal by a random factor within a random range.

[0086] Meanwhile, it is understood that in the second model, not only can all of the one or more biomarkers extracted by the extraction unit (144) be used as independent variables in the second model, but also the remaining biomarkers, which are judged to be inappropriate by the verification unit (145) described later among the one or more biomarkers extracted by the extraction unit (144), or some of the biomarkers selected by the second selection unit (146) can be used as independent variables in the second model.

[0087] For convenience of explanation, in an embodiment where the second selection unit (146) is used, the biomarkers extracted by the extraction unit (144) are referred to as candidate biomarkers, and the biomarkers selected by the second selection unit (146) are referred to as one or more biomarkers, distinguishing them by that term. That is, it is intended that the markers ultimately used in the second model are used as one or more biomarkers defined herein.

[0088] Below, the verification unit (145) and the second selection unit (146) are described as follows.

[0089] The verification unit (145) can verify the quality of one or more biomarkers extracted by the extraction unit (144). Specifically, the verification unit (145) can verify the usefulness of one or more biomarkers extracted by the extraction unit (144) using an exploratory data analysis technique. The verification unit (145) can verify whether the value of one or more biomarkers extracted by the extraction unit (144) has a value within a predefined range. The verification unit (145) can also verify the legitimacy of the format by indicating whether the actual age and / or heart-related health status of the patient is indicated for each of the one or more biomarkers extracted by the extraction unit (144).

[0090] At this time, it is preferable that low-quality biomarkers, among one or more biomarkers extracted by the extraction unit (144), which fall outside the preset level of usefulness or the preset range of values ​​or do not satisfy the data format, be discarded. On the other hand, high-quality biomarkers may be stored in a structured format in the memory (130).

[0091] The second selection unit (146) can select some of the one or more candidate biomarkers extracted through the extraction unit (144) as one or more biomarkers.

[0092] The second selection unit (146) may select some of the extracted candidate biomarkers as one or more biomarkers based on the correlation between the extracted candidate biomarkers and the user's actual age. In this case, it is assumed that the candidate biomarkers are multiple individuals.

[0093] Specifically, the second selection unit (146) may select some of the biomarkers that show a statistically significant relationship in a correlation test between multiple candidate biomarkers and actual age as one or more biomarkers to be used in the second model. For example, the second selection unit (146) may perform a correlation test, such as Pearson or Spearman, on the relationship between multiple candidate biomarkers and actual age, and select some of the biomarkers whose P-values ​​are lower than a predefined significance level as one or more biomarkers to be used in the second model.

[0094] The second selection unit (146) performs a correlation test on the relationship between multiple candidate biomarkers and actual age, calculates the effect size of the correlation, and selects some of the multiple candidate biomarkers in order of clinical importance as one or more biomarkers to be used in the second model.

[0095] The second selection unit (146) may select some of the candidate biomarkers as one or more biomarkers to be used in the second model based on the results of performing multiple comparison corrections on the multiple candidate biomarkers. For example, the second selection unit (146) may apply the Bonferroni Correction and FDR (False Discovery Rate) to the statistical test results between the multiple candidate biomarkers and actual age, and may select some of the candidate biomarkers as one or more biomarkers to be used in the second model based on whether the corrected P-Value satisfies the significance level.

[0096] The second selection unit (146) may select one or more biomarkers from among the plurality of candidate biomarkers to be used in the second model based on the results of a visual analysis of the relationship between the plurality of candidate biomarkers and actual age. For example, the second selection unit (146) may select one or more biomarkers from among the plurality of candidate biomarkers to be used in the second model using a scatter plot.

[0097] The second selection unit (146) can select one or more biomarkers to be used in the second model, excluding candidate biomarkers with low biological validity when considering the relationship between multiple candidate biomarkers and actual age.

[0098] The second selection unit (146) can divide the plurality of candidate biomarkers into groups and select a representative biomarker for each group as one or more biomarkers to be used in the second model. Specifically, the second selection unit (146) can cluster the plurality of candidate biomarkers into groups based on similarities, differences, biological / medical relevance, or a combination thereof. At this time, the second selection unit (146) can select the optimal biomarker for each group as a representative biomarker and select the representative biomarker for each group as one or more biomarkers to be used in the second model. Here, the meaning of optimality may mean minimizing information redundancy among the remaining candidate biomarkers in each group, best reflecting the typical characteristics of the group, or maximizing the predictive power of the second model.

[0099] The output unit (147) can apply a mathematical transformation to the values ​​of one or more biomarkers extracted by the extraction unit (144) or selected by the second selection unit (146). For example, the output unit (147) can perform a linear transformation on the values ​​of one or more biomarkers so that the correlation with actual age can be linear and consistent. Accordingly, the output unit (147) can linearly calculate the relationship between one or more biomarkers and actual age, while improving the simplicity, ease of interpretation, and computational efficiency of the model.

[0100] The calculation unit (147) can calculate a linear correlation between a plurality of biomarkers and the user's actual age while keeping the sign fixed. For example, if the correlation between a plurality of biomarkers and the actual age is positive, the calculation unit (147) can calculate a correlation with a fixed sign while maintaining the direction of the relationship so that when a linearly transformed independent variable increases, the dependent variable also increases. For example, the calculation unit (147) can transform a plurality of biomarkers through a monotonic transformation (monotonic power transformation) so that the order according to the size of the values ​​does not change, and in this process, can induce a linear relationship with age. Specifically, the monotonic transformation can be implemented as a transformation function including a square root function, a cube root function, a logarithmic function, a reciprocal function, and / or a box-cox function. The transformation function can be optimized using parameters.

[0101] From this, the output unit (147) can reflect the biological characteristic that heart health consistently worsens with age through fixation of the sign, and can simplify the interpretation of the correlation.

[0102] Here, optimization can mean maximizing the correlation coefficient or minimizing the error (e.g., mean squared error, residual sum of squares) between multiple transformed biomarkers and actual age. Alternatively, optimization can refer to the process of optimizing model quality assessment metrics (e.g., Akaike Information Criterion; AIC, Bayesian Information Criterion; BIC). It is desirable that the optimization process be validated through statistical testing and visual plots to demonstrate its validity.

[0103] The modeling unit (148) can model a second model using one or more biomarkers. In this case, the second model is a pre-trained regression model from which optimal regression coefficients are derived. The modeling unit (148) can model a regression model that estimates a user's biological age using one or more extracted or selected biomarkers as independent variables.

[0104] The modeling unit (148) may also model a second model by adding non-ECG biomarkers as independent variables. For example, non-ECG biomarkers may include non-ECG biomarkers extracted from vital signs, specimen test results, and the presence of concomitant diseases, in addition to information obtainable from ECGs.

[0105] For example, the modeling unit (148) may model a second model composed of a constrained regression model. At this time, the second model may have the following constraints. The second model may have a condition that representative biomarkers exist for each group as a first constraint. The second model may have a condition that the magnitude of the correlation between one or more biomarkers and actual age must be within a preset sign or range as a second constraint. At this time, the second model may have at least one of the first constraint and the second constraint as a constraint. Meanwhile, the first constraint presupposes that a clustering process will be performed on a plurality of candidate biomarkers by the second selection unit (146). That is, the second model may have both the first constraint and the second constraint, or in some cases, only the second constraint.

[0106] The modeling unit (148) can model the second model by using one or more candidate biomarkers that satisfy the first constraint and / or the second constraint (hereinafter, constraint) as biomarkers or by limiting the regression coefficient of a biomarker that does not satisfy the constraint to a certain range.

[0107] The output unit (149) can output a biological age corresponding to an electrocardiogram signal using the first model and the second model.

[0108] The output unit (149) may output the delta age, which is the difference between the biological age and the actual age. At this time, the output unit (149) may also output the grade corresponding to the delta age and the risk level of the corresponding disease. For example, if the delta age exists in the difference range of 0 or more and 10 or more, it may be output as grade 1, if it is 11 or more and 20 or less, it may be output as grade 2, and if it is over 20, it may be output as grade 3.

[0109] The output unit (149) can sequentially provide biomarkers that have contributed the most to estimating biological age. The output unit (149) can also output the clinical significance of the contribution of the biomarkers.

[0110] The output unit (149) may also output the contribution of one or more biomarkers in calculating biological age as visual data. For example, the output unit (149) may visually output the contribution of one or more biomarkers in calculating biological age using a coefficient plot, a waterfall plot, or a radar chart.

[0111] FIG. 3 is an exemplary diagram illustrating the operation of a device (100) for estimating biological age using a biomarker according to one embodiment.

[0112] Referring to FIG. 3, a device (100) for estimating biological age using a biomarker according to one embodiment receives an electrocardiogram signal. At this time, the electrocardiogram signal may be a one-dimensional electrocardiogram signal in digital format or two-dimensional image data in which a one-dimensional electrocardiogram signal is recorded. Here, there is no restriction on the format, as the one-dimensional electrocardiogram signal can be extracted from a two-dimensional image or the two-dimensional image can be an image depicting a one-dimensional electrocardiogram signal and are mutually compatible.

[0113] Thereafter, the device (100) for estimating biological age using a biomarker according to one embodiment selects a type of first model suitable for processing based on the format of the electrocardiogram signal. For example, if the input electrocardiogram signal is a one-dimensional signal, a first model composed of an RNN that is strong in extracting time-series features and a CNN that has learned one-dimensional signals may be selected. As another example, if the input electrocardiogram signal is a two-dimensional signal, a first model composed of a vision transformer that is strong in image processing and a CNN may be selected. Meanwhile, it is preferable that the input electrocardiogram signal be preprocessed so that the first model can process it.

[0114] Thereafter, a device (100) for estimating biological age using a biomarker according to one embodiment extracts one or more candidate biomarkers from an electrocardiogram signal using a first model.

[0115] Here, "candidate biomarker" refers to a candidate biomarker that will later be used as a modeling variable in the second model. In other words, the biomarker selected from among the candidate biomarkers is ultimately used, and biomarkers with potential for use are referred to as candidate biomarkers. Candidate biomarkers are indicators related to heart health, independent of the user's age. It is preferable to exclude indicators related to actual age.

[0116] Thereafter, the device (100) for estimating biological age using a biomarker according to one embodiment may select some of one or more candidate biomarkers. At this time, the selected biomarker is intended to be the optimal biomarker as it is most helpful in determining biological age, such as having a high correlation with actual age or being free of errors.

[0117] Thereafter, the device (100) for estimating biological age using a biomarker according to one embodiment can construct a second model using the selected biomarker. At this time, the second model may be a constrained regression model in which biomarkers satisfying constraints are used as independent variables. At this time, the biomarkers used as independent variables may be representative biomarkers selected from each group based on classification criteria. Alternatively, they may refer to all biomarkers previously selected without a classification process. At this time, the biomarkers may refer to variables selected in a minimum number while maximizing the predictive performance of the second model.

[0118] Thereafter, the device (100) for estimating biological age using a biomarker according to one embodiment can estimate biological age, i.e., biological age of the heart, using the second model, and estimate delta age, which is the difference between biological age and actual age. The device (100) for estimating biological age using a biomarker according to one embodiment can not only provide biological age and delta age on the display (120), but also visually provide the corresponding health status (risk, disease severity) and the contribution of the biomarker according to biological age estimation.

[0119] Furthermore, the device (100) for estimating biological age using a biomarker according to one embodiment may provide personalized medical advice with clinical diagnosis based on at least one of the estimated biological age, delta age, contribution of the biomarker, and health status.

[0120] Figure 4 is a flowchart illustrating a method for estimating biological age according to one embodiment.

[0121] A method for estimating biological age according to an embodiment of FIG. 4 can be performed by the device (100) of FIG. 1.

[0122] First, a method for estimating biological age according to one embodiment inputs input data including a user's electrocardiogram signal into a first model to extract one or more biomarkers (410).

[0123] Thereafter, a method for estimating biological age according to one embodiment estimates (420) the biological age of the user based on a second model modeled using one or more biomarkers.

[0124] Meanwhile, although FIG. 4 describes the method by dividing it into multiple steps, at least some of the steps may be performed in a different order, combined with other steps and performed together, omitted, divided into detailed steps and performed, or one or more steps not shown may be added and performed.

[0125] The disclosed embodiments are a biological age estimation device and method using a biomarker, which can be used in the digital medical industry in that a biomarker is extracted from an electrocardiogram signal using an artificial intelligence model and a biological age is estimated through a regression model built based on the extracted biomarker.

Claims

1. One or more processors; and A biological age estimation device using a biomarker having one or more memories storing instructions executed by one or more processors, One or more of the above processors: Input data including the user's biosignal is input into a pre-trained first model to extract one or more biomarkers, A device for estimating biological age using a biomarker, configured to estimate the biological age of the user by inputting one or more biomarkers into a second model that has been previously learned.

2. In paragraph 1, The above biosignals include electrocardiogram signals, A device for estimating biological age using a biomarker, wherein the biological age includes the biological age of the heart.

3. In paragraph 1, The first model includes an artificial intelligence model that is trained to output one or more corresponding biomarkers when inputting learning data related to an electrocardiogram signal, A biological age estimation device using a biomarker, wherein the second model is trained to select regression coefficients and independent variables that minimize the error from the correct value by modeling some of the output data in the learning process of the first model as independent variables, and includes a constraint regression model trained to output the corresponding biological age when the value of the output data is input.

4. In paragraph 3, The training data of the first model includes input values ​​including an electrocardiogram signal; and A biological age estimation device using a biomarker, wherein at least one characteristic among anatomical characteristics, physiological characteristics, and pathological characteristics corresponding to the electrocardiogram signal is configured to be labeled as a pair of correct values.

5. In paragraph 4, The above anatomical characteristics are characteristics related to cardiac hypertrophy, and include at least one of left atrial enlargement (LAE), left ventricular enlargement (LVH), right atrial enlargement (Right Atrial Enlargement), right ventricular hypertrophy (Right Ventricular Hypertrophy), left ventricular mass index (Left Ventricular Mass Index), and left atrial volume index (Left Atrial Volume Index). The above physiological characteristics are characteristics related to systolic function, such as whether the left or right ventricle has systolic or diastolic dysfunction, left ventricular ejection fraction, myocardial strain, brain natriuretic peptide (BNP) and pro-brain natriuretic peptide (ProBNP), and Characteristics associated with pulmonary hypertension, including pulmonary wedge pressure and the presence of pulmonary hypertension; and Characteristics related to myocardial ischemia, including coronary artery calcium score, presence of coronary artery stenosis, presence of myocardial ischemia, and troponin test results; and Includes at least one probability of death within a specific period of time, A device for estimating biological age using a biomarker, wherein the above pathological characteristics are characteristics related to valve dysfunction, and include at least one of the presence or absence of stenosis or regurgitation of the aortic, mitral, pulmonary, and tricuspid valves and a score.

6. In paragraph 1, One or more of the above processors: Inputting the user's electrocardiogram signal into the first model to extract one or more candidate biomarkers, A biological age estimation device using a biomarker, configured to extract one or more biomarkers by selecting some of the one or more candidate biomarkers as the one or more biomarkers.

7. In paragraph 6, One or more of the above processors: A biological age estimation device using a biomarker, configured to select some of the one or more candidate biomarkers as the one or more biomarkers based on at least one of a correlation between the one or more candidate biomarkers and the actual age of the user, an effect size of the correlation, and a biological validity of the one or more candidate biomarkers.

8. In paragraph 6, The above one or more candidate biomarkers are plural, One or more of the above processors: Clustering the above one or more candidate biomarkers into each group based on classification criteria, Select one or more representative biomarkers for each of the above groups, A device for estimating biological age using a biomarker, configured to estimate the biological age by inputting the one or more representative biomarkers into the second model.

9. In paragraph 8, A device for estimating biological age using a biomarker, wherein the classification criteria include at least one of similarity and biological or medical relevance between the one or more candidate biomarkers.

10. In paragraph 1, One or more of the above processors: A device for estimating biological age using a biomarker, configured to visually display on a screen of a device communicating with the processor at least one of the contribution of the one or more biomarkers in determining the biological age together with the biological age, a description of the contribution, a difference between the user's cardiac biological age and actual age, a grade corresponding to the difference, a health status according to the grade, and instructions for improving the health status.

11. In paragraph 1, One or more of the above processors: Performing a linear transformation on said one or more biomarkers so that said one or more biomarkers have a linear correlation with the actual age of said user with a fixed sign, A biological age estimation device using a biomarker, which calculates a correlation between one or more biomarkers for which the linear transformation has been completed and the actual age.

12. In paragraph 8 or paragraph 11, The second model includes a constraint regression model, A device for estimating biological age using a biomarker, characterized in that the first constraint of the constraint regression model is that one or more representative biomarkers exist for each group, and the second constraint of the constraint regression model is that the magnitude of the correlation is within a preset range.

13. In paragraph 1, The above input data includes at least one of a one-dimensional signal representing the user's electrocardiogram in time series and a two-dimensional signal representing the electrocardiogram as an image, The first model includes at least one of a vision transformer, a convolutional neural network, and a recurrent neural network, One or more of the above processors: When the type of the above input data is a one-dimensional signal, the input data is input into the 1-1 model consisting of at least one of the convolutional neural network and the recurrent neural network, A biological age estimation device using a biomarker, configured to input the input data into a first or second model comprising at least one of the vision transformer and the convolutional neural network when the type of the input data is a two-dimensional signal.

14. One or more processors; and A method performed by a biological age estimation device using a biomarker having one or more memories storing instructions executed by one or more processors, The above method is: A step of inputting input data including a user's biosignal into a pre-trained first model to extract one or more biomarkers; and A method for estimating biological age using a biomarker, comprising a step of estimating the biological age of the user by inputting the one or more biomarkers into a pre-learned second model.

15. In paragraph 14, The above biosignals include electrocardiogram signals, A method for estimating biological age using a biomarker, wherein the biological age includes the biological age of the heart.

16. In paragraph 14, The first model includes an artificial intelligence model that is trained to output one or more corresponding biomarkers when inputting learning data related to an electrocardiogram signal, A method for estimating biological age using a biomarker, wherein the second model is trained to select regression coefficients and independent variables that minimize the error from the correct value while modeling some of the output data in the learning process of the first model as independent variables, and includes a constraint regression model trained to output the corresponding biological age when the value of the output data is input.

17. In paragraph 16, The training data of the first model includes input values ​​including an electrocardiogram signal; and A method for estimating biological age using a biomarker, wherein at least one of the anatomical characteristics, physiological characteristics, and pathological characteristics corresponding to the electrocardiogram signal is configured to be labeled as a pair of correct values.

18. In paragraph 17, The above anatomical characteristics are characteristics related to cardiac hypertrophy, and include at least one of left atrial enlargement (LAE), left ventricular enlargement (LVH), right atrial enlargement (Right Atrial Enlargement), right ventricular hypertrophy (Right Ventricular Hypertrophy), left ventricular mass index (Left Ventricular Mass Index), and left atrial volume index (Left Atrial Volume Index). The above physiological characteristics are characteristics related to systolic function, such as whether the left or right ventricle has systolic or diastolic dysfunction, left ventricular ejection fraction, myocardial strain, brain natriuretic peptide (BNP) and pro-brain natriuretic peptide (ProBNP), and Characteristics associated with pulmonary hypertension, including pulmonary wedge pressure and the presence of pulmonary hypertension; and Characteristics related to myocardial ischemia, including coronary artery calcium score, presence of coronary artery stenosis, presence of myocardial ischemia, and troponin test results; and Includes at least one probability of death within a specific period of time, A method for estimating biological age using a biomarker, wherein the above pathological characteristics are characteristics related to valve dysfunction, and include at least one of the presence or absence of stenosis or regurgitation of the aortic, mitral, pulmonary, and tricuspid valves and a score.

19. In paragraph 14, The step of extracting one or more biomarkers is: A step of inputting the user's electrocardiogram signal into a first model to extract one or more candidate biomarkers; and A method for estimating biological age using a biomarker, comprising a step of extracting one or more biomarkers by selecting some of the one or more candidate biomarkers as the one or more biomarkers.

20. In paragraph 19, The step of extracting one or more biomarkers is: A method for estimating biological age using a biomarker, comprising a step of selecting some of the one or more candidate biomarkers as the one or more biomarkers based on at least one of a correlation between the one or more candidate biomarkers and the actual age of the user, an effect size of the correlation, and a biological validity of the one or more candidate biomarkers.

21. In paragraph 19, The above one or more candidate biomarkers are plural, The above method is: A step of clustering the one or more candidate biomarkers into each group based on classification criteria; A step of selecting one or more representative biomarkers for each of the above groups; A method for estimating biological age using a biomarker, comprising a step of estimating the biological age by inputting the one or more representative biomarkers into the second model.

22. In paragraph 21, A method for estimating biological age using biomarkers, wherein the classification criteria include at least one of similarity and biological or medical relevance between the one or more biomarkers.

23. In paragraph 14, The above method is: A method for estimating biological age using a biomarker, comprising the step of visually displaying on a screen of the device at least one of the contribution of the one or more biomarkers in determining the biological age together with the biological age, a description of the contribution, a difference between the user's cardiac biological age and actual age, a grade corresponding to the difference, a health status according to the grade, and a guideline for improving the health status.

24. In paragraph 14, The above method is: A step of performing a linear transformation on the one or more biomarkers so that the one or more biomarkers have a linear correlation with the actual age of the user with a fixed sign; A method for estimating biological age using a biomarker, further comprising a step of calculating a correlation between one or more biomarkers for which the linear transformation has been completed and the actual age.

25. In paragraph 21 or 24, The second model includes a constraint regression model, A method for estimating biological age using a biomarker, wherein the first constraint of the above constraint regression model is that one or more representative biomarkers exist, and the second constraint of the above constraint regression model is that the magnitude of the correlation is within a preset range.

26. In paragraph 14, The above input data includes at least one of a one-dimensional signal representing the user's electrocardiogram in time series and a two-dimensional signal representing the electrocardiogram as an image, The first model includes at least one of a vision transformer, a convolutional neural network, and a recurrent neural network, The above method is: When the type of the input data is a one-dimensional signal, a step of inputting the input data into a first-1 model composed of at least one of the convolutional neural network and the recurrent neural network; and A method for estimating biological age using a biomarker, comprising the step of inputting the input data into a first-second model comprising at least one of the vision transformer and the convolutional neural network, when the type of the input data is a two-dimensional signal.

Images