Method, program and apparatus for generating long-term heart rate variability based on short-term measured cardiac signals

Abstract

According to an embodiment of the present disclosure, a method, program, and apparatus for generating long-term heart rate variability based on a cardiac signal measured short-term, performed by a computing device, are disclosed, the method including: acquiring cardiac data including cardiac signals measured within a first reference time; and generating heart rate variability for a second reference time, the second reference time being greater than the first reference time, based on the cardiac data using a pre-trained deep learning model.

Description

[Technical Field]

[0001] The present disclosure relates to artificial intelligence technology in the medical field, and more particularly to a method that can utilize artificial intelligence to generate heart rate variability itself from a cardiac signal measured short-term, similar to that extracted from a cardiac signal measured long-term, and to extract features from the generated heart rate variability. [Background technology]

[0002] The heart does not always beat uniformly. The intervals between heartbeats can get longer or shorter. When a person is healthy, the variations in the intervals between heartbeats are generally large and complex. On the other hand, when a person is ill or under stress, the complexity of the intervals between heartbeats is generally significantly reduced. Large variations in heartbeats indicate a good ability to adapt to changes inside or outside the body, while small variations in heartbeats indicate a reduced ability to adapt to the internal or external environment.

[0003] As mentioned above, the state of the human body can be understood from the fluctuations in heartbeats, and the concept that emerged from this is heart rate variability (HRV). Heart rate variability is a numerical representation of the changes that occur when the intervals between heartbeats are measured sequentially. In other words, heart rate variability is a set of values ​​that numerically represent the changes that occur when the intervals between heartbeats, such as the interval between the first and second beats, the interval between the second and third beats, etc., are measured.

[0004] Heart rate variability is usually measured over 24 hours or 5 minutes as the gold standard. Several studies have investigated the suitability of ultra-short-term (less than 5 minutes) heart rate variability data and found that the shorter the time, the greater the discrepancy from the gold standard. In particular, there was a significant difference in the frequency-related characteristics of heart rate variability. Therefore, it is generally recommended to measure cardiac signals for 5 minutes or more to analyze heart rate variability.

[0005] However, it is difficult for users to maintain a focused state continuously for five minutes to measure heart rate signals. Therefore, even during the five-minute heart rate signal measurement process, noise due to movement is likely to be introduced, and especially during 24-hour measurements, a large amount of noise is inevitable. This noise reduces the reliability of heart rate variability analysis results. Therefore, if performance equivalent to the optimal standard can be achieved in a short measurement period, this would be a great help in terms of user convenience and would increase the reliability of heart rate variability analysis by limiting noise. Summary of the Invention [Problem to be solved by the invention]

[0006] The present disclosure aims to provide a method that can utilize artificial intelligence to generate heart rate variability itself from short-term measured cardiac signals, similar to that extracted from long-term measured cardiac signals, and to extract features from the generated heart rate variability.

[0007] However, the problems to be solved by this disclosure are not limited to those mentioned above, and other problems not mentioned will be clearly understood from the description below. [Means for solving the problem]

[0008] According to one embodiment of the present disclosure, there is disclosed a method for generating long-term heart rate variability (HRV) based on a cardiac signal measured for a short period of time, the method including: acquiring cardiac data including a cardiac signal measured within a first reference time; and generating HRV for a second reference time, the second reference time being greater than the first reference time, based on the cardiac data, using a pre-trained deep learning model.

[0009] Alternatively, the cardiac data may further include at least one of biological information, disease information, or physical activity information of the person whose cardiac signal is measured.

[0010] Alternatively, the step of generating heart rate variability for a second reference time, which is a value greater than the first reference time, based on the cardiac data using a pre-trained deep learning model may include the steps of inputting the cardiac data into a pre-trained first model to calculate a mean and standard deviation for the heart rate variability for the second reference time, predicting a probability distribution for the heart rate variability for the second reference time based on the mean and standard deviation, and performing sampling based on the predicted probability distribution to calculate a value included in the heart rate variability for the second reference time.

[0011] Alternatively, the step of performing sampling based on the predicted probability distribution to calculate a value included in the heart rate variability at the second reference time may include, if the predicted probability distribution corresponds to a normal distribution, performing random sampling on the predicted probability distribution to calculate a value included in the heart rate variability at the second reference time.

[0012] Alternatively, the step of performing sampling based on the predicted probability distribution to calculate a value included in the heart rate variability at the second reference time may include, if the predicted probability distribution does not correspond to a normal distribution, performing modeling on the predicted probability distribution, and performing random sampling on the probability distribution generated through the modeling to calculate a value included in the heart rate variability at the second reference time.

[0013] Alternatively, the step of performing sampling based on the predicted probability distribution to calculate a value included in the heart rate variability at the second reference time may include, if the predicted probability distribution corresponds to a normal distribution, performing random sampling on the predicted probability distribution to calculate a value included in the heart rate variability at the second reference time.

[0014] Alternatively, the step of performing sampling based on the predicted probability distribution to calculate a value included in the heart rate variability at the second reference time may include, if the predicted probability distribution does not correspond to a normal distribution, performing modeling on the predicted probability distribution, and performing random sampling on the probability distribution generated through the modeling to calculate a value included in the heart rate variability at the second reference time.

[0015] Alternatively, the modeling for the predicted probability distribution may include a Gaussian mixture model or a normalizing flow.

[0016] Alternatively, the step of performing sampling based on the predicted probability distribution to calculate a value included in the heart rate variability at the second reference time may include the steps of determining whether the predicted probability distribution corresponds to a normal distribution, selecting one of a plurality of sampling techniques for extracting the heart rate variability at the second reference time based on the result of the determination, and performing sampling on the probability distribution using the selected sampling technique to calculate a value included in the heart rate variability at the second reference time.

[0017] Alternatively, the step of generating heart rate variability for a second reference time, which is greater than the first reference time, based on the cardiac data using a pre-trained deep learning model may further include rearranging the order of values ​​included in the calculated heart rate variability for the second reference time using a pre-trained second model based on self-attention.

[0018] Alternatively, the step of rearranging the order of values ​​included in the calculated heart rate variability at the second reference time using a second model based on pre-trained self-attention may include the steps of inputting the heart rate variability at the second reference time into the second model to generate an attention map indicating the correlation of values ​​included in the heart rate variability, and rearranging the order of values ​​included in the heart rate variability based on the attention map.

[0019] Alternatively, the step of generating heart rate variability for a second reference time, which is a value greater than the first reference time, based on the cardiac data using a pre-trained deep learning model may include inputting the cardiac data into a third model, which is a pre-trained generative model, to calculate a value contained in the heart rate variability for the second reference time.

[0020] Alternatively, the step of generating heart rate variability for a second reference time, which is a value greater than the first reference time, based on the cardiac data using a pre-trained deep learning model may include inputting the cardiac data into a fourth model, which is a pre-trained sequence-to-sequence model, to calculate a value included in the heart rate variability for the second reference time.

[0021] According to one embodiment of the present disclosure, there is provided a computer program stored on a computer-readable storage medium. When executed by one or more processors, the computer program performs operations for generating long-term heart rate variability (HRV) based on short-term measured cardiac signals. The operations include acquiring cardiac data including cardiac signals measured within a first reference time period, and generating HRV for a second reference time period, the second reference time period being greater than the first reference time period, based on the cardiac data using a pre-trained deep learning model.

[0022] According to one embodiment of the present disclosure, there is provided a computing device for generating long-term heart rate variability (HRV) based on cardiac signals measured over a short period of time. The device includes a processor including at least one core, a memory including program code executable by the processor, and a network unit for acquiring cardiac data including cardiac signals measured over a first reference time period. The processor may generate HRV for a second reference time period, the second reference time period being greater than the first reference time period, based on the cardiac data using a pre-trained deep learning model. [Effects of the Invention]

[0023] According to the present disclosure, in a situation where cardiac signals must be measured for a long period of time, long-term measurement performance can be achieved with only a short-term measurement, thereby greatly improving user convenience.

[0024] Furthermore, since the heart rate variability itself can be generated even with only a short measurement, the probability of noise occurring during measurement of the heart signal can be significantly reduced.

[0025] Furthermore, since heart rate variability itself, similar to that extracted from long-term measured cardiac signals, can be generated from short-term measured cardiac signals, a versatile system can be realized that can be readily applied to various diagnostic algorithms utilizing heart rate variability. [Brief explanation of the drawings]

[0026] [Figure 1] FIG. 1 is a block diagram of a computing device according to one embodiment of the present disclosure. [Figure 2] FIG. 2 is a conceptual diagram illustrating a cardiac signal measured within a first reference time period and heart rate variability within a second reference time period according to an embodiment of the present disclosure. [Figure 3]FIG. 10 is a block diagram illustrating a process for generating heart rate variability for a second reference time period based on cardiac data measured within a first reference time period, according to one embodiment of the present disclosure. [Figure 4] FIG. 10 is a block diagram illustrating a process for generating heart rate variability for a second reference time using a deep learning model according to an embodiment of the present disclosure. [Figure 5] FIG. 10 is a block diagram illustrating a process for extracting heart rate variability features for a second reference time using a deep learning model according to an embodiment of the present disclosure. [Figure 6] 1 is a flowchart illustrating a method for generating long-term heart rate variability based on short-term measured cardiac signals according to one embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0027] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily implement the present disclosure. The embodiments presented in this disclosure are provided to enable those skilled in the art to use or practice the contents of the present disclosure. Therefore, various modifications to the embodiments of the present disclosure will be apparent to those skilled in the art. That is, the present disclosure may be embodied in various different forms and is not limited to the following embodiments.

[0028] Throughout the specification of the present disclosure, the same or similar reference numerals refer to the same or similar components. In addition, in order to clearly explain the present disclosure, reference numerals of parts that are not relevant to the explanation of the present disclosure may be omitted from the drawings.

[0029] The term "or" as used in this disclosure is intended to mean an inclusive "or" rather than an exclusive "or." That is, unless otherwise specified or otherwise clear from the context in this disclosure, "X uses A or B" should be understood to mean one of the natural inclusive permutations. For example, unless otherwise specified or otherwise clear from the context in this disclosure, "X uses A or B" can be interpreted as either X uses A, X uses B, or X uses both A and B.

[0030] The term "and / or" as used in this disclosure must be understood to indicate and include all possible combinations of one or more of the associated listed concepts.

[0031] The terms "comprises" and / or "comprising" as used in this disclosure should be understood to mean that the specified features and / or components are present. However, the terms "comprises" and / or "comprising" should not be understood to exclude the presence or addition of one or more other features, other components and / or combinations thereof.

[0032] In this disclosure, unless otherwise specified or clear from the context as referring to the singular form, the singular should generally be construed as including "one or more."

[0033] The term "nth (n is a natural number)" used in this disclosure can be understood as an expression used to distinguish components of the present disclosure from one another based on a predetermined criterion, such as functional, structural, or convenience of description. For example, in this disclosure, components that perform different functional roles can be classified as a first component or a second component. However, components that are substantially identical within the technical concept of the present disclosure but must be distinguished for convenience of description can also be classified as a first component or a second component.

[0034] The term "acquire" as used in this disclosure may be understood to mean not only receiving data from an external device or system via a wired or wireless communication network, but also generating data in an on-device form.

[0035] Meanwhile, the terms "module" or "unit" used in this disclosure may be understood to refer to an independent functional unit that processes computing resources, such as a computer-related entity, firmware, software or a portion thereof, hardware or a portion thereof, or a combination of software and hardware. Here, a "module" or "unit" may refer to a unit composed of a single element or a unit expressed as a combination or collection of multiple elements. For example, as a concept of connotation, a "module" or "unit" may refer to a hardware element or a collection of hardware elements of a computing device, an application program that performs a specific software function, a processing procedure implemented by executing software, or a collection of instructions for executing a program. Furthermore, as a broad concept, a "module" or "unit" may refer to a computing device itself that constitutes a system, or an application executed on a computing device. However, the above concepts are merely examples, and the concepts of "module" and "unit" may be defined in various ways within the scope of understanding of those skilled in the art based on the contents of this disclosure.

[0036] The term "model" as used in this disclosure may be understood as a system implemented using mathematical concepts and language to solve a specific problem, a collection of software units for solving a specific problem, or an abstract model of a processing process for solving a specific problem. For example, a neural network "model" may refer to a system implemented as a neural network that has problem-solving capabilities through learning. Here, a neural network may have problem-solving capabilities by optimizing parameters connecting nodes or neurons through learning. A neural network "model" may include a single neural network or a neural network ensemble in which multiple neural networks are combined.

[0037] The explanations of the above terms are intended to aid in understanding the present disclosure. Therefore, unless the above terms are explicitly stated as matters that limit the contents of the present disclosure, care should be taken not to use them in a way that limits the technical ideas of the contents of the present disclosure.

[0038] FIG. 1 is a block diagram of a computing device according to one embodiment of the present disclosure.

[0039] The computing device 100 according to an embodiment of the present disclosure may be a hardware device or part of a hardware device that performs comprehensive data processing and calculations, or may be a software-based computing environment connected via a communication network. For example, the computing device 100 may be a server that performs intensive data processing functions and shares resources, or a client that shares resources by interacting with the server. The computing device 100 may also be a cloud system in which multiple servers and clients interact with each other to comprehensively process data. The above description is merely an example of a type of computing device 100, and various types of computing device 100 may be configured within the scope of what one skilled in the art would understand based on the contents of this disclosure.

[0040] 1, a computing device 100 according to an embodiment of the present disclosure may include a processor 110, a memory 120, and a network unit 130. However, since FIG. 1 is merely an example, the computing device 100 may include other components for implementing a computer environment. Also, the computing device 100 may include only some of the disclosed components.

[0041] The processor 110 according to an embodiment of the present disclosure may be understood as a component including hardware and / or software for performing computing operations. For example, the processor 110 may read a computer program to perform data processing for machine learning. The processor 110 may process operations such as input data processing for machine learning, feature extraction for machine learning, and error calculation based on backpropagation. The processor 110 for performing such data processing may include a central processing unit (CPU), a general-purpose graphics processing unit (GPGPU), a tensor processing unit (TPU), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), etc. The types of processor 110 described above are merely examples, and various types of processor 110 may be configured within the scope of what one skilled in the art would understand based on the present disclosure.

[0042] The processor 110 may use a pre-trained deep learning model to generate heart rate variability for a second reference time, which is greater than the first reference time, based on cardiac data including a cardiac signal measured within a first reference time. Here, the cardiac signal may be understood as a signal indicating cardiac activity or status. The reference time may be a value preset by a user taking into consideration various factors such as the measurement environment and purpose of use of the cardiac signal. For example, the cardiac signal may include a 12-lead electrocardiogram, a Holter electrocardiogram, or photoplethysmography (PPG). The first reference time may be 10 to 30 seconds, and the second reference time may be 5 to 24 hours, in accordance with the gold standard for heart rate variability analysis.

[0043] The processor 110 may calculate heart rate variability features from the heart rate variability at the second reference time generated through the deep learning model. For example, the processor 110 may calculate heart rate variability features such as standard deviation of RR interval (SDRR), mean RR, standard deviation of all NN intervals (SDNN), root mean square of the successive differences (RMSSD), power in low frequency range (LF), or power in high frequency range (HF) by mathematically calculating values ​​included in the heart rate variability at the second reference time. The processor 110 may then combine the calculated heart rate variability features to derive values ​​necessary for diagnosis, such as a stress index, stress resistance, autonomic nervous balance, autonomic nervous activity, fatigue, depression level, or anxiety disorder.

[0044] The memory 120 according to an embodiment of the present disclosure may be understood as a component including hardware and / or software for storing and managing data processed by the computing device 100. That is, the memory 120 may store any type of data generated or determined by the processor 110 and any type of data received by the network unit 130. For example, the memory 120 may include at least one type of storage medium selected from the group consisting of flash memory, hard disk, multimedia card micro, card-type memory, random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, and optical disk. The memory 120 may also include a database system that manages data in a predetermined manner. The types of memory 120 described above are merely examples, and various configurations of the memory 120 are possible within the scope of what would be understood by one skilled in the art based on the present disclosure.

[0045] The memory 120 may structure and organize and manage data, data combinations, and program code executable by the processor 110 required for the processor 110 to perform calculations. For example, the memory 120 may store cardiac data received via the network unit 130 (described below). The memory 120 may store program code for operating a neural network model to receive cardiac data and perform learning, program code for operating the neural network model to receive cardiac data and perform inference according to the intended use of the computing device 100, and processed data generated by executing the program code.

[0046] The network unit 130 according to an embodiment of the present disclosure may be understood as a component that transmits and receives data via any type of known wired or wireless communication system. For example, the network unit 130 may transmit and receive data using a wired or wireless communication system such as a local area network (LAN), wideband code division multiple access (WCDMA), long term evolution (LTE), wireless broadband internet (WiBro), 5th generation mobile communication (5G), ultra wide-band, ZigBee, radio frequency (RF) communication, wireless LAN, wireless fidelity (WiFi), near field communication (NFC), or Bluetooth. The above-described communication systems are merely examples, and various wired or wireless communication systems for transmitting and receiving data by the network unit 130 may be applied in addition to the above examples.

[0047] The network unit 130 may receive data necessary for the processor 110 to perform calculations via wired or wireless communication with any system or any client. The network unit 130 may also transmit data generated by calculations by the processor 110 via wired or wireless communication with any system or any client. For example, the network unit 130 may receive cardiac data by communicating with a database in a hospital environment, a cloud server that performs tasks such as medical data standardization, a client such as a smart watch, or a medical computing device. The network unit 130 may transmit output data of the neural network model, and intermediate data and processed data derived during the calculation process of the processor 110, via communication with the database, server, client, computing device, or the like.

[0048] FIG. 2 is a conceptual diagram illustrating a cardiac signal measured within a first reference time period and heart rate variability within a second reference time period according to one embodiment of the present disclosure.

[0049] In cardiac signal 10 measured within the first reference time, the numerical value indicating the interval between peaks indicates the value included in the heart rate variability within the first reference time. For example, as shown in Figure 2, the interval between the first peaks is 822 ms, and the numerical value 822 indicates a first value of heart rate variability, and the interval between the second peaks is 857 ms, and the numerical value 857 indicates a second value of heart rate variability. In this way, by listing the numerical values ​​indicating the interval between peaks of cardiac signal 10 measured within the first reference time, the heart rate variability within the first reference time can be calculated.

[0050] 2 , the heart rate variability (HRV) 20 for a second reference time generated from the cardiac signal 10 measured within a first reference time by the computing device 100 according to an embodiment of the present disclosure may include a value included in the HRV for the first reference time and a newly generated value 25. The newly generated value 25 may be understood as a value calculated based on a result of predicting what characteristics and aspects the cardiac signal 10 measured within the first reference time would exhibit if measured up to the second reference time. That is, the computing device 100 may input the cardiac signal 10 measured within the first reference time into a deep learning model to generate a HRV value predicted for a second reference time greater than or equal to the first reference time. The computing device 100 may then generate the HRV for the second reference time by combining the HRV value calculable from the cardiac signal 10 measured within the first reference time with the newly generated value.

[0051] FIG. 3 is a block diagram illustrating a process for generating heart rate variability for a second reference time period based on cardiac data measured within a first reference time period, according to one embodiment of the present disclosure.

[0052] 3, cardiac data 30 according to an embodiment of the present disclosure may include a cardiac signal, such as an electrocardiogram, measured within a first reference time. The cardiac data 30 may further include at least one of biological information, disease information, or physical activity information of the person whose cardiac signal was measured. The biological information may include the age, sex, height, and weight of the person whose cardiac signal was measured. The disease information may include the presence or absence of a disease at the time of measurement and a breakdown of the disease. The physical activity information may include the condition and activity state at the time of measurement.

[0053] A computing device 100 according to an embodiment of the present disclosure may input cardiac data 30 into a deep learning model 200 to calculate long-term heart rate variability 35 corresponding to heart rate variability at a second reference time. The deep learning model 200 may include a first model 210 that predicts a probability distribution of heart rate variability and samples the heart rate variability according to the predicted distribution, a second model 215 that refines the heart rate variability sampled through the first model 210, a third model 220 that corresponds to a generative model, and a fourth model 230 that corresponds to a sequence-to-sequence model. The computing device 100 may calculate the long-term heart rate variability 35 by selectively or in combination using the first model 210 to the fourth model 230 depending on the type of cardiac signal included in the cardiac data 30, the user's intention, etc.

[0054] For reference, the process of generating the long-term heart rate variability 35 using the first model 210 and the second model 215 will be described in detail later with reference to FIGS.

[0055] The third model 220 can learn the distribution of heart rate variability within a first reference time from the cardiac data 30 and generate long-term heart rate variability 35 corresponding to the heart rate variability at a second reference time from the learned distribution. That is, the third model 220 can grasp the distribution of heart rate variability calculated from the cardiac data 30 including signals actually measured within the first reference time and generate a new long-term heart rate variability 35 that is similar to the heart rate variability calculated from signals actually measured up to the second reference time. For example, the third model 220 can include a generative neural network such as a generative adversarial network (GAN) or a variational autoencoder (VAE). However, the types of neural networks of the third model 220 described above are merely examples, and various generative neural networks other than those described above can be used as the neural network of the third model 220.

[0056] The fourth model 230 can learn characteristics of the cardiac data 30 that change over time and generate long-term heart rate variability 35 corresponding to the heart rate variability at a second reference time based on the learning results. That is, the fourth model 230 can predict how the cardiac data 30 will change after the first reference time and generate long-term heart rate variability 35 at the second reference time. For example, the fourth model 230 can include a sequence-to-sequence model such as a recurrent neural network (RNN) or a long short-term memory (LSTM). However, the type of neural network of the fourth model 230 described above is merely an example, and various generative neural networks other than the above examples can be used as the neural network of the fourth model 230.

[0057] The computing device 100 can calculate heart rate variability features 39 from the long-term heart rate variability 35 corresponding to the heart rate variability at the second reference time. Here, the heart rate variability features 39 can include SDRR, mean RR, SDNN, RMSSD, LF, HF, etc. The computing device 100 does not directly extract heart rate variability features 39 from cardiac data 30 that does not meet the optimal standard, but instead generates long-term heart rate variability 35 that meets the optimal standard and then extracts heart rate variability features 39 from the long-term heart rate variability 35, thereby easily extracting a variety of features.

[0058] Fig. 4 is a block diagram illustrating a process for generating heart rate variability for a second reference time using a deep learning model according to an embodiment of the present disclosure, and Fig. 5 is a block diagram illustrating a process for extracting heart rate variability features for a second reference time using a deep learning model according to an embodiment of the present disclosure.

[0059] 4, computing device 100 may input cardiac data 40 including cardiac signals measured within a first reference time into first model 210 to calculate mean 43 and standard deviation 45 for long-term heart rate variability corresponding to heart rate variability within a second reference time. Here, first model 210 may be a model trained based on supervised learning, but first model 210 may also be trained by a learning method other than supervised learning that is applicable based on the contents disclosed in this disclosure.

[0060] The computing device 100 can predict a probability distribution 47 of long-term heart rate variability using the mean 43 and standard deviation 45 for the long-term heart rate variability. The computing device 100 can then sample a long-term heart rate variability value 49 from the predicted probability distribution 47. Here, sampling can be understood as a process of extracting a desired value from a probability distribution. For example, if the predicted probability distribution 47 corresponds to a normal distribution, the computing device 100 can perform random sampling on the predicted probability distribution 47 to calculate the long-term heart rate variability value 49. In other words, if the predicted probability distribution 47 corresponds to a normal distribution, the computing device 100 can generate the long-term heart rate variability value 49 by randomly sampling values ​​present in the predicted probability distribution 47 until a second reference time is met. On the other hand, if the predicted probability distribution 47 does not correspond to a normal distribution, the computing device 100 can perform modeling on the predicted probability distribution 47. Here, modeling the probability distribution 47 may include modeling such as a Gaussian mixture model or normalizing flow as a process of transforming the probability distribution 47. In other words, even if the predicted probability distribution 47 does not follow a normal distribution, the computing device 100 can process the probability distribution 47 to make it easier to extract a desired value from the probability distribution 47. Then, the computing device 100 can calculate the long-term heart rate variability value 49 by performing random sampling on the modeled probability distribution.

[0061] Meanwhile, the computing device 100 may use sampling techniques other than the random sampling described above. For example, the computing device 100 may first determine whether the predicted probability distribution 47 corresponds to a normal distribution. If the predicted probability distribution 47 corresponds to a normal distribution, the computing device 100 may perform random sampling to extract samples that exist on the normal distribution and generate a long-term heart rate variability value 49. If the predicted probability distribution 47 does not correspond to a normal distribution, there is a high probability that the results of performing random sampling will have a large error. Therefore, the computing device 100 may select another sampling technique, such as a clustering technique based on sample distance, to perform sampling on the predicted probability distribution 47 and extract a long-term heart rate variability value 49. That is, the computing device 100 may select one of a plurality of sampling techniques to perform sampling depending on whether the predicted probability distribution 47 corresponds to a normal distribution.

[0062] 5, the computing device 100 can rearrange the order of the long-term heart rate variability values ​​50 using a second model 220 based on self-attention. As shown in FIG. 4, the long-term heart rate variability values ​​50 are generated by sampling the probability distribution predicted by the first model 210. Therefore, the long-term heart rate variability values ​​50 cannot have the order information of the original cardiac data input to the first model 210. This is not a significant problem when calculating heart rate variability features (e.g., SDNN, etc.) for which order information is not important, but it can be a problem when calculating heart rate variability features (e.g., RMSSD, etc.) for which order information is important. Therefore, the computing device 100 can input the long-term heart rate variability values ​​50 to the second model 220 based on self-attention and rearrange the order of the long-term heart rate variability values ​​50 so that order information can be added to the long-term heart rate variability values ​​50 generated by sampling the probability distribution predicted by the first model 210.

[0063] Specifically, the computing device 100 may input the long-term heart rate variability values ​​50 into the second model to generate an attention map 55 indicating the correlation of the long-term heart rate variability values ​​50. The attention map may represent the degree of correlation between a particular long-term heart rate variability value and other long-term heart rate variability values ​​based on the particular long-term heart rate variability value. Here, a high correlation may mean a close order. For example, assuming that the long-term heart rate variability value 800, which corresponds to the first value, is most highly correlated with 890, the area in the attention map where 800 and 890 match may be displayed in the darkest color. This display on the attention map may be interpreted as indicating that 890 is preferably placed after 800 in the long-term heart rate variability. In other words, the computing device 100 may grasp and rearrange the order of the long-term heart rate variability values ​​50 through the attention map, thereby generating an ordered long-term heart rate variability value 59.

[0064] FIG. 6 is a flow chart illustrating a method for generating long-term heart rate variability based on short-term measured cardiac signals according to one embodiment of the present disclosure.

[0065] 6, a computing device 100 according to an embodiment of the present disclosure may acquire cardiac data including a cardiac signal measured within a first reference time period (S210). For example, if the computing device 100 does not include a separate measurement unit, the computing device 100 may acquire cardiac data including a cardiac signal measured within the first reference time period via wired or wireless communication with equipment that measured the cardiac signal. If the computing device 100 includes a separate measurement unit, the computing device 100 may generate cardiac data by measuring the cardiac signal within the first reference time period via the measurement unit. Meanwhile, the cardiac data may further include at least one of human biological information, disease information, or physical activity information stored at the time of measuring the cardiac signal, in addition to the cardiac signal measured within the first reference time period.

[0066] The computing device 100 may use a pre-trained deep learning model to generate heart rate variability for a second reference time, which is a value greater than that for the first reference time, based on cardiac data (S220). Specifically, the computing device 100 may input the cardiac data into a pre-trained first model to calculate a mean and standard deviation for the heart rate variability for the second reference time. The computing device 100 may predict a probability distribution for the heart rate variability for the second reference time based on the mean and standard deviation. The computing device 100 may perform sampling based on the predicted probability distribution to calculate values ​​included in the heart rate variability for the second reference time. The computing device 100 may then use a pre-trained second model based on self-attention to rearrange the order of the values ​​included in the calculated heart rate variability for the second reference time. The computing device 100 may also input the cardiac data into a pre-trained third model, which is a generative model, to calculate values ​​included in the heart rate variability for the second reference time. The computing device 100 may also input the cardiac data into a fourth model, which is a pre-trained sequence-to-sequence model, to calculate values ​​contained in the heart rate variability at the second reference time.

[0067] Although not shown in FIG. 6, the computing device 100 may extract characteristics of the heart rate variability at the second reference time based on the values ​​contained in the heart rate variability at the second reference time calculated by step S210.

[0068] The various embodiments of the present disclosure described above can be combined with additional embodiments and can be modified within the scope that can be understood by those skilled in the art from the above detailed description. The embodiments of the present disclosure are illustrative in all respects and should not be construed as limiting. For example, each component described as a single type can also be implemented in a distributed form, and similarly, components described as distributed can also be implemented in a combined form. Therefore, all modifications and variations derived from the meaning, scope, and equivalent concepts of the claims of the present disclosure should be construed as being within the scope of the present disclosure.

Claims

1. 1. A method for generating long-term heart rate variability based on short-term measured cardiac signals, performed by a computing device including at least one processor, the method comprising: acquiring cardiac data including cardiac signals measured within a first reference time period; generating a second reference time of heart rate variability based on the cardiac data using a pre-trained deep learning model, the second reference time being greater than the first reference time; A method comprising:

2. The method of claim 1 , wherein the cardiac data further comprises at least one of biological information, disease information, or physical activity information of the person measuring the cardiac signal.

3. The step of generating a heart rate variability of a second reference time, which is greater than the first reference time, based on the cardiac data using a pre-trained deep learning model includes: inputting the cardiac data into a pre-trained first model to calculate a mean and a standard deviation of the heart rate variability for the second reference time; predicting a probability distribution for the heart rate variability for the second reference time based on the mean and standard deviation; performing sampling based on the predicted probability distribution to calculate a value included in the heart rate variability for the second reference time; The method of claim 1 , comprising:

4. The step of calculating a value included in the heart rate variability for the second reference time by performing sampling based on the predicted probability distribution includes: If the predicted probability distribution corresponds to a normal distribution, The method of claim 3 , further comprising: performing random sampling on the predicted probability distribution to calculate values ​​contained in heart rate variability for the second reference time.

5. The step of calculating a value included in the heart rate variability for the second reference time by performing sampling based on the predicted probability distribution includes: If the predicted probability distribution does not correspond to a normal distribution, performing modeling on the predicted probability distribution; calculating a value included in the heart rate variability for the second reference time by performing random sampling on the probability distribution generated through the modeling; The method of claim 3, comprising:

6. The method of claim 5 , wherein the modeling for the predicted probability distribution includes a Gaussian mixture model or a normalizing flow.

7. The step of calculating a value included in the heart rate variability for the second reference time by performing sampling based on the predicted probability distribution includes: determining whether the predicted probability distribution corresponds to a normal distribution; selecting one of a plurality of sampling techniques for extracting the heart rate variability for the second reference time based on the result of the determination; performing sampling on the probability distribution via a selected sampling technique to calculate a value included in the heart rate variability for the second reference time; The method of claim 3, comprising:

8. The step of generating heart rate variability for a second reference time, which is greater than the first reference time, based on the cardiac data using a pre-trained deep learning model includes:

4. The method of claim 3, further comprising rearranging the order of values ​​included in the calculated heart rate variability for the second reference time using a second model based on pre-trained self-attention.

9. rearranging the order of values ​​included in the calculated heart rate variability for the second reference time using a second model based on pre-trained self-attention; inputting the heart rate variability at the second reference time into the second model to generate an attention map indicating correlations of values ​​included in the heart rate variability; rearranging the order of values ​​included in the HRV based on the attention map; The method of claim 8, comprising:

10. The step of generating heart rate variability for a second reference time, which is greater than the first reference time, based on the cardiac data using a pre-trained deep learning model includes: The method of claim 1 , further comprising inputting the cardiac data into a third model, which is a pre-trained generative model, to calculate a value included in heart rate variability at the second reference time.

11. The step of generating heart rate variability for a second reference time, which is greater than the first reference time, based on the cardiac data using a pre-trained deep learning model includes:

2. The method of claim 1, further comprising inputting the cardiac data into a fourth model, which is a pre-trained sequence-to-sequence model, to calculate a value included in the heart rate variability for the second reference time.

12. A computer program stored on a computer-readable storage medium, comprising: the computer program is configured to cause a computer to perform operations for generating long-term heart rate variability based on short-term measured cardiac signals; The operation is acquiring cardiac data including cardiac signals measured within a first reference time; generating heart rate variability for a second reference time, the second reference time being greater than the first reference time, based on the cardiac data using a pre-trained deep learning model; a computer program comprising:

13. 1. A computing device for generating long-term heart rate variability based on short-term measured cardiac signals, comprising: a processor including at least one core; a memory containing program code executable by the processor; a network unit for acquiring cardiac data including cardiac signals measured within a first reference time; Including, The processor uses a pre-trained deep learning model to generate heart rate variability for a second reference time based on the cardiac data, the second reference time being greater than the first reference time.

Images