Apparatus and method for synthesizing bio-signal

Abstract

The present invention relates to an apparatus and a method for, when given a single bio-signal, synthesizing (generating) the bio-signal into a signal which can be derived according to a change in frequency amplitude, and providing the signal. More specifically, the apparatus comprises: a characteristic representation unit which represents the characteristics of the bio-signal; a first mapping unit for mapping a morphological characteristic among the represented characteristics of the bio-signal to a normal distribution; a second mapping unit for mapping a frequency characteristic among the represented characteristics of the bio-signal to a uniform distribution; and a generation unit for generating a synthesized bio-signal on the basis of the mapping results of the first mapping unit and the second mapping unit, wherein the original state of a patient can be more accurately captured by minimizing the loss of important information that can occur in N-ary processing and purification processes, and uncertainty can be quantified and explained by presenting various possibilities derived from the original signal.

Description

Biosignal synthesis device and method

[0001] The present invention relates to a biosignal synthesis device and method, and more particularly, to a device and method for generating synthetic data for a biosignal using a pass filter and a Bayesian-deep learning-based signal synthesis framework.

[0002] Biosignals are utilized in various ways as they are one of the important indicators for measuring health in human life. In particular, biosignals such as arterial blood pressure (ABP) and electrocardiogram (ECG) play a key role in maintaining patient life by reflecting the physiological changes of patients in real time in medical fields such as intensive care units and operating rooms.

[0003] Typically, to measure biosignals, primary vendors, such as sensors, measure the biosignals first, refine them, and then provide them. Secondary vendors, such as data processors, perform secondary processing before providing biosignal information. Furthermore, library developers and tutorial developers related to biosignals reprocess the refined biosignals further, transforming them into a form that is easier to utilize. Finally, researchers further process biosignal data that has undergone three or more Nth-level processing to suit their research needs.

[0004] As the Nth processing is performed through a pass filter in the multi-stage purification process described above, loss of biosignal information inevitably occurs, which may have a negative impact on patient treatment and research.

[0005] (Patent Document 1) Republic of Korea Patent Publication No. 10-2022-0095009 (published on July 6, 2022)

[0006] The present invention aims to solve the above technical problem, and its purpose is to synthesize (generate) and provide a signal that can be derived by a pass filter effect when a pulse biosignal is given. In another aspect, the present invention aims to restore a biosignal that has been modified through N-th processing to a signal close to the original signal.

[0007] According to an embodiment of the present invention for solving the above technical problem, a biosignal synthesis device may include a feature representation unit representing the feature of the biosignal; a first mapping unit mapping morphological features among the features of the represented biosignal to a normal distribution; a second mapping unit mapping frequency amplitude features among the features of the represented biosignal to a uniform distribution; and a generation unit generating a synthesized biosignal based on the mapping results of the first mapping unit and the second mapping unit.

[0008] In addition, the first mapping unit and the second mapping unit can be performed by a self-developed mechanism derived from a model of the VAE (Variational Auto-encoder) series, which is a Bayesian deep learning.

[0009] In addition, the second mapping unit maps a characteristic to which a pass filter effect - an effect of adjusting the amplitude characteristics of the frequency - is applied according to a coefficient of the frequency characteristic, and the generation unit can convert the mapping result of the first mapping unit and the mapping result of the second mapping unit into the synthesized biosignal and provide it using a decoder.

[0010] In addition, the biosignal is a biosignal having a linear pulsating characteristic as a pulsating signal, and the synthesized biosignal can have a planar shape by generating a plurality of the synthesized biosignals so that they are enhanced only in frequency amplitude without changing the morphological characteristics such as phase change and time axis shift based on the biosignal.

[0011] A biosignal synthesis method according to another embodiment of the present invention may include a step of representing morphological characteristics and frequency amplitude characteristics of the biosignal input from a feature representation unit; a step of mapping the morphological characteristics to a normal distribution in a first mapping unit; a step of mapping the frequency amplitude characteristics to a uniform distribution in a second mapping unit; and a step of generating a synthesized biosignal based on a mapping result of the first mapping unit and a mapping result of the second mapping unit in a generation unit.

[0012] In addition, the step of mapping to the normal distribution and the step of mapping to the uniform distribution can be performed by a self-developed mechanism derived from a Bayesian deep learning-based VAE (Variational Auto-encoder) series model.

[0013] In addition, the step of mapping to the above uniform distribution can map a characteristic to which a pass filter effect - an effect of adjusting the amplitude characteristics of the frequency - is applied according to the coefficient of the above frequency characteristic.

[0014] In addition, the input biosignal is a biosignal having a pulse characteristic in the form of a line, and the synthesized biosignal can be synthesized to have a planar shape by generating a plurality of the synthesized biosignals so that only the frequency amplitude is enhanced without changing the morphological characteristics such as phase change and time axis shift based on the biosignal.

[0015] The biosignal synthesis device and method according to one embodiment of the present invention described above can more accurately capture the original state of a patient by minimizing the loss of important information that may occur during the Nth purification process, and can quantify and explain uncertainty by suggesting various possibilities derived from the original signal.

[0016] Furthermore, by presenting pulse biosignals that reflect uncertainty, it can enable decision-making based on more accurate information, and by synthesizing various signals through frequency-pass filter coefficient adjustment, it can provide additional information and insight, enabling analysis of frequency-specific characteristics of biosignals, thereby aiding in accurate data analysis and interpretation, and promoting new discoveries and insights in the medical and research fields.

[0017] Additionally, by serving as a base model that supports transfer learning for models that reflect the frequency characteristics of pulse signals, it can provide essential functions for reflecting various patient states and conditions.

[0018] Additionally, by synthesizing the frequency pass filter effect, it can be used to analyze how the frequency amplitude change of a pulse signal responds to external effects, thereby helping in the development of biosignal analysis algorithms.

[0019] FIG. 1 is a drawing schematically illustrating a biosignal synthesis device according to one embodiment of the present invention.

[0020] FIG. 2 is a diagram illustrating a learning process of a mapping unit of a biosignal synthesis device according to one embodiment of the present invention.

[0021] FIG. 3 is a drawing illustrating the operation of a biosignal synthesis device according to one embodiment of the present invention when there is only a first mapping unit.

[0022] FIG. 4 is a diagram illustrating the operation of a biosignal synthesis device according to one embodiment of the present invention.

[0023] FIG. 5 is a flowchart illustrating an operation method of a biosignal synthesis device according to one embodiment of the present invention.

[0024] The present invention can have various modifications and various embodiments, and specific embodiments are illustrated in the drawings and described in detail.

[0025] However, this is not intended to limit the present invention to a specific embodiment, but should be understood to include all modifications, equivalents, or substitutes included in the spirit and technical scope of the present invention.

[0026] The terminology used in this application is only used to describe specific embodiments and is not intended to limit the present invention. The singular expression includes the plural expression unless the context clearly indicates otherwise. In this application, it should be understood that the terms "comprise" or "have" indicate the presence of a characteristic, number, step, operation, component, part, or combination thereof described in the specification, but do not exclude in advance the possibility of the presence or addition of one or more other characteristics, numbers, steps, operations, components, parts, or combinations thereof.

[0027] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. Terms defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and will not be interpreted in an idealized or overly formal sense unless explicitly defined herein.

[0028] Hereinafter, with reference to the attached drawings, preferred embodiments of the present invention will be described in more detail. In order to facilitate an overall understanding in describing the present invention, identical reference numerals will be used for identical components in the drawings, and redundant descriptions of identical components will be omitted.

[0029] FIG. 1 is a drawing schematically illustrating a biosignal synthesis device according to one embodiment of the present invention.

[0030] In one embodiment of the present invention, the biosignal may include all biosignals having pulsatile characteristics, and synthesizing the biosignal may include various forms of sub-signals that may be derived based on the input biosignal and the original signal prior to Nth refinement.

[0031] A biosignal synthesis device according to one embodiment of the present invention may include a characteristic representation unit (100), a first mapping unit (200), a second mapping unit (300), and a generation unit (400).

[0032] The characteristic representation unit (100) can represent morphological characteristics and frequency characteristics for the input biosignal, and generally, the characteristics can be represented using an encoder, but is not limited thereto, and any method can be used as long as the morphological characteristics and frequency characteristics can be represented for a signal having a pulsating characteristic.

[0033] The first mapping unit (200) can support regularity and controllability in the data generation process by mapping the morphological characteristics represented in the characteristic representation unit (100) to an N-dimensional normal distribution, thereby greatly improving consistency and controllability in the data generation process of the biosignal synthesis device, thereby ensuring consistent quality data.

[0034] The first mapping unit (200) focuses on learning an efficient latent representation of input data and can map the morphological characteristics of input data to a normal distribution.

[0035] The second mapping unit (300) can map the frequency characteristics represented in the characteristic representation unit (100) to a uniform distribution.

[0036] The second mapping unit (300) can provide the effect of a frequency pass filter that can adjust the amplitude of the frequency, and can finely adjust the detailed characteristics according to the change in the frequency amplitude of the biosignal by adjusting the coefficient of the frequency filter to the input frequency amplitude.

[0037] The first mapping unit (200) and the second mapping unit (300) can operate simultaneously or sequentially to perform data synthesis, and can be configured with a self-developed mechanism derived from a model of the VAE (Variational Auto-encoder) series, which is a Bayesian deep learning.

[0038] The results of the first mapping unit (200) and the second mapping unit (300) can represent characteristics in the form of random variables of a normal distribution and a uniform distribution, respectively.

[0039] The generation unit (400) can be formed as a decoder, and can synthesize a biosignal waveform by realizing real number values ​​in various combinations from the probability variables of the mapping results of the first mapping unit (200) and the second mapping unit (300) and then inputting them into the decoder, and the synthesized biosignal waveform can be visualized through a display (not shown) and provided to the user.

[0040] The input biosignal is a biosignal having a pulsating characteristic, and can generally be expressed in the form of a line pulsating like a periodic function. However, according to one embodiment of the present invention, the biosignal synthesized in the generation unit (400) can generate multiple synthesized signals in the Y-axis direction with respect to the input biosignal.

[0041] Specifically, since the signals mapped in the first mapping unit (200) and the second mapping unit (300) are random variables, by realizing these random variables as real values, multiple synthesized biosignal waveforms can be generated centered on the input biosignals, and accordingly, when the synthesized signals are expressed on the same graph, they can be visualized in the form of a surface by overlapping each other. In addition, it goes without saying that the multiple synthesized biosignal waveforms can be represented and used in the form of independent lines, depending on the user's selection.

[0042] In addition, with respect to the meaning of generating a plurality of synthesized biosignal waveforms in the Y-axis direction, in expressing a biosignal that changes over time on a display (not shown), the X-axis is generally set as the time axis, and the Y-axis is set as the variable axis representing the biosignal value, and then provided as a graph. In such a graph, it is natural that the biosignal that changes over time is provided in the form of a linear graph, and the result of synthesizing the biosignal according to one embodiment of the present invention can be synthesized for the biosignal value of the Y-axis rather than the time variable of the X-axis, and accordingly, a plurality of synthesized biosignal waveforms can be generated and provided in the upper and lower directions of the Y-axis based on the input biosignal.

[0043] FIG. 2 is a diagram illustrating a learning process of a mapping unit (a first mapping unit (200) and a second mapping unit (300)) of a biosignal synthesis device according to one embodiment of the present invention.

[0044] As described in the above drawing 1, the first mapping unit (200) and the second mapping unit (300) can be configured with a self-developed mechanism derived from a model of the VAE (Variational Auto-encoder) series, which is a Bayesian deep learning.

[0045] When an Nth-processed biosignal (hereinafter, “original signal”) is input, a filtering operation using cascading filters can be performed to represent features of the original signal. The cascading filter performs mathematical morphological filtering and can be designed in various ways by combining high pass, low pass, and band pass. In one embodiment of the present invention, cascading filtering can be performed using a combination of high pass and low pass to represent features, and band pass can also be utilized as needed.

[0046] Referring to the 'Feature Representation' section, the first signal (X_H) and the second signal (X_L) can be represented by performing morphological filtering by applying the high-pass first coefficient (fc1) and the low-pass second coefficient (fc2) to the step filter for the original signal located in the central part.

[0047] Thereafter, the third signal (X_HH) and the fourth signal (X_HL) can be represented by applying the high-pass third coefficient (fc3) and the low-pass fourth coefficient (fc4) to the first signal (X_H), and the fifth signal (X_LH) and the sixth signal (X_LL) can be represented by applying the high-pass fifth coefficient (fc5) and the low-pass sixth coefficient (fc6) to the second signal (X_L).

[0048] In Fig. 2, the first signal (X_H) to the sixth signal (X_LL) are represented by performing a two-stage step filtering, but the seventh coefficient (fc7) to the Nth coefficient (fcn) can be applied thereafter to additionally represent the seventh signal to the Nth signal, and if frequency amplitude-based filtering that sequentially applies a high pass or a low pass can be performed, the signals can be represented without limitation and used as features for the original signal.

[0049] In addition, the original signal can be encoded after being applied to the encoder and mapped in the first mapping unit (200) and the second mapping unit (300) to generate features of a synthesized biosignal (hereinafter, synthesized features).

[0050] Synthetic features can be generated to have the same form as the second signal (X_HH) to the sixth signal (X_LL) among the first signal (X_H) to the sixth signal (X_LL) generated in the feature representation stage, thereby minimizing loss.

[0051] Real values ​​realized in the first mapping unit (200) and the second mapping unit (300) are applied to the decoder to synthesize the second signal (X_HH) to the sixth signal (X_LL), and the original signal is restored based on the second signal (X_HH) to the sixth signal (X_LL), the restored signal is compared with the original signal to calculate the loss, and learning can be performed repeatedly so that the loss is reduced.

[0052] FIG. 3 is a drawing illustrating the operation of a biosignal synthesis device according to one embodiment of the present invention when there is only a first mapping unit (200).

[0053] As described in the above-described Fig. 1, the first mapping unit (200) can map the morphological characteristics among the characteristics represented in the characteristic representation unit (100) for the input original signal to an N-dimensional normal distribution.

[0054] The mapped characteristics are decoded in the generation unit (400) to imitate the characteristics of the original signal and restore the replicated characteristics into a biosignal. Since the biosignal synthesized in the generation unit (400) imitates the characteristics of the original signal, a single line-shaped biosignal based on the original signal can be generated.

[0055] In the case of replicating a biosignal using only the first mapping unit (200) as in Fig. 3, since synthesis according to frequency amplitude is not performed, all information is mapped to a normal distribution, so changes such as phase shift and time axis shift may occur during the mapping process, and due to such phase shift and time axis shift, it is impossible to control synthesis according to frequency amplitude changes while fixing morphological characteristics (basic skeleton).

[0056] In order to solve the above problem, in one embodiment of the present invention, the first mapping unit (200) and the second mapping unit (300) can be used together as shown in FIG. 4.

[0057] In the first mapping unit (200), morphological characteristics can be mapped to a normal distribution for the characteristics represented in the characteristic representation unit (100), and then, as in FIG. 4, in the second mapping unit (300), a pass filter effect can be synthesized according to the coefficient of the frequency characteristic to additionally perform mapping to a uniform distribution.

[0058] The frequency filter can be used without limitation as long as it is a filter that can filter frequencies, such as a high-pass filter or a low-pass filter, and by adjusting the filtering coefficient of such a frequency filter, it is possible to generate multiple synthetic signals in the Y-axis direction of a linear graph based on the frequency skeleton while maintaining the basic skeleton of the original signal, i.e., according to the effect of the frequency pass filter.

[0059] That is, the biosignal synthesis device according to one embodiment of the present invention can synthesize signals of various shapes that appear to have a planar shape as shown in FIG. 4 by generating a plurality of synthesized biosignals in terms of frequency and amplitude through mapping of the first mapping unit (200) and the second mapping unit (300), while the original signal is a pulsating biosignal in a linear form.

[0060] FIG. 5 is a flowchart illustrating an operation method of a biosignal synthesis device according to one embodiment of the present invention.

[0061] A biosignal to be synthesized can be input into a biosignal synthesis device (S1100).

[0062] The input bio-signals can include all bio-signals with pulsatile characteristics, such as blood pressure, electrocardiogram (ECG), electroencephalogram, electromyogram, pulse, and electrocardiogram, and all methods for measuring bio-signals can be used without any particular restrictions.

[0063] The characteristics of the biosignal input from the feature representation section (100) can be represented (S1200).

[0064] By applying the represented characteristics to the first mapping unit (200), the morphological characteristics of the biosignal can be mapped to an N-dimensional normal distribution (S1300).

[0065] The biosignal synthesized in the above step S1300 is a biosignal synthesized in a random form, and in order to maintain the basic structure of the input biosignal, the pass filter effect according to the frequency amplitude characteristics can be mapped to a uniform distribution in the second mapping unit (300) (S1400).

[0066] In the generation unit, a biosignal can be synthesized based on the mapping result of the first mapping unit and the mapping result of the second mapping unit (S1500).

[0067] As described above, the biosignal synthesis device and method according to one embodiment of the present invention can more accurately capture the original state of a patient by minimizing the loss of important information that may occur during the Nth purification process, and can quantify and explain uncertainty by suggesting various possibilities derived from the original signal.

[0068] Furthermore, by collecting and presenting various signals reflecting uncertainty, it can enable decision-making based on more accurate information, and by synthesizing various signals through adjusting frequency-pass filter coefficients, it can provide additional information and insight, enabling analysis of frequency-specific characteristics of biosignals, thereby aiding in accurate data analysis and interpretation, and promoting new discoveries and insights in the medical and research fields.

[0069] Additionally, by serving as a base model that supports transfer learning for models that reflect frequency changes in pulse signals, it can provide essential functions for reflecting various patient states and conditions.

[0070] Additionally, by synthesizing the frequency pass filter effect, it can be used to analyze how the frequency change of a pulse signal responds to external effects, thereby helping in the development of biosignal analysis algorithms.

[0071] The characteristics, structures, effects, etc. described in the above-described embodiments are included in at least one embodiment of the present invention, and are not necessarily limited to only one embodiment. Furthermore, the characteristics, structures, effects, etc. exemplified in each embodiment can be combined or modified and implemented in other embodiments by a person having ordinary skill in the art to which the embodiments belong.

[0072] Accordingly, the contents related to such combinations and modifications should be interpreted as being included within the scope of the present invention. In addition, although the above description focuses on the embodiments, these are merely examples and do not limit the present invention. Those skilled in the art to which the present invention pertains will appreciate that various modifications and applications not illustrated above are possible without departing from the essential characteristics of the present embodiments. For example, each component specifically shown in the embodiments can be modified and implemented. In addition, the differences related to such modifications and applications should be interpreted as being included within the scope of the present invention defined in the appended claims.

[0073] By presenting pulse biosignals that reflect uncertainty, decision-making based on more accurate information can be enabled. By synthesizing various signals through frequency-pass filter coefficient adjustment, additional information and insight can be provided, enabling analysis of frequency-specific characteristics of biosignals. This helps in accurate data analysis and interpretation, and can promote new discoveries and insights in the medical and research fields.

[0074] 100: Top of the characteristic table

[0075] 200: First Mapping Department

[0076] 300: Second Mapping Department

[0077] 400: Generation Unit

Claims

1. In a biosignal synthesis device, A characteristic representation section representing the characteristics of the above biosignal; A first mapping unit that maps morphological characteristics among the characteristics of the represented biosignal to a normal distribution; A second mapping unit that maps the amplitude characteristics of the frequency among the characteristics of the represented biosignal to a uniform distribution; and A generation unit that generates a synthesized biosignal based on the mapping results of the first mapping unit and the second mapping unit; A biosignal synthesis device including:

2. In paragraph 1, A biosignal synthesis device characterized in that the first mapping unit and the second mapping unit are performed by a self-developed mechanism derived from a model of the VAE (Variational Auto-encoder) series, which is a Bayesian deep learning.

3. In paragraph 1, The second mapping section maps the characteristics to which the pass filter effect is applied according to the coefficients of the frequency characteristics, A biosignal synthesis device characterized in that the above generation unit converts the mapping result of the first mapping unit and the mapping result of the second mapping unit into the synthesized biosignal and provides the result using a decoder.

4. In paragraph 1, The biosignal is a biosignal having a pulse characteristic in the form of a line, and the synthesized biosignal is a biosignal synthesis device characterized in that it has a planar form by generating a plurality of the synthesized biosignals so that the synthesized biosignals are enhanced only in frequency amplitude without changing the morphological characteristics such as phase change and time axis shift based on the biosignal.

5. In a method for synthesizing biosignals, A step of representing the morphological characteristics and frequency amplitude characteristics of the biosignal input from the feature representation section; A step of mapping the morphological characteristics to a normal distribution in the first mapping section; A step of mapping the amplitude characteristics of the frequency to a uniform distribution in the second mapping section; and A step of generating a synthesized biosignal based on the mapping result of the first mapping unit and the mapping result of the second mapping unit in the generation unit; A method for synthesizing biosignals including:

6. In paragraph 5, The step of mapping to the above normal distribution and the step of mapping to the above uniform distribution are A biosignal synthesis method characterized by being performed using a self-developed mechanism derived from a Bayesian deep learning-based VAE (Variational Auto-encoder) series model.

7. In paragraph 5, The step of mapping to the above uniform distribution is A biosignal synthesis method characterized in that the characteristics are mapped by applying a pass filter effect - an effect of adjusting the amplitude characteristics of the frequency - according to the coefficients of the frequency characteristics.

8. In paragraph 5, A biosignal synthesis method characterized in that the input biosignal is a biosignal having a linear pulsating characteristic, and the synthesized biosignal is synthesized to have a planar shape by generating a plurality of the synthesized biosignals so that the frequency amplitude is only enhanced without changing the morphological characteristics such as phase change and time axis shift based on the biosignal.

Images