Apparatus and method for refining bio-signal artifacts

Abstract

The present invention relates to an apparatus and a method for refining bio-signal artifacts and, more specifically, comprises: a database in which bio-signals are stored; an extraction unit for extracting only clean signals from among the bio-signals; an augmentation unit for generating augmented artifacts by augmenting the clean signals; a storage unit for storing the clean signals and the augmented artifacts; a training unit for training a refining model by using a dataset generated on the basis of the augmented artifacts; and a refining unit for refining artifacts of the bio-signals by using the refining model, wherein previously unusable data is refined and converted into useful information, and thus the range of use of data can be widened, and both the quality and amount of clinical data can be improved by reducing data loss.

Description

Biosignal artifact purification device and method

[0001] The present invention relates to a device and method for refining biosignal artifacts, and more particularly, to a device and method for enhancing biosignal artifacts to learn a refinement model and refining biosignal artifacts using the learned refinement model.

[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), photoplethysmography (PPG), and electrocardiogram (ECG) play a key role in maintaining the lives of patients by reflecting their physiological changes in real time in medical fields such as intensive care units and operating rooms.

[0003] These biosignals are used as an important means of confirming the survival of patients who cannot speak in environments such as intensive care units and operating rooms, and medical measures can be taken based on rapidly changing biosignals. However, monitoring biosignals can lead to incorrect medical judgments as inaccurate biosignals are measured for various reasons.

[0004] Inaccurate bio-signals can be broadly categorized into noise and artifacts. Noise refers to a reduction in image quality or clarity due to unwanted random fluctuations or errors in images or data, while artifacts are abnormal signals that occur due to external factors such as equipment limitations, patient movement, interference from specific substances, or mechanical defects.

[0005] While noise and artifacts share the common characteristic of being abnormal signals, noise refers to "noise" resulting from random errors, whereas artifacts arise from specific causes. While noise affects the quality and clarity of a signal, artifacts exhibit abrupt changes or changes in the signal's structure. Because noise and artifacts differ in their causes and forms, they must be distinguished and refined separately.

[0006] However, conventional technologies often consider noise and artifacts to be the same, and thus artifacts are often input into noise filtering techniques. As a result, artifacts are not refined and irregular or abnormal signals are output, resulting in signals that are lost, transformed, or distorted, being mistaken for normal signals, which can have a negative impact on patient diagnosis and treatment.

[0007] To solve the above problems, methods using filters or artificial intelligence (AI) models have been proposed as a method to refine only artifacts. However, simple filtering methods have limitations in that, although fixed-frequency band filters are useful for removing simple noise, there is a risk of losing useful information when biosignals and noise overlap, and they cannot properly reflect the time-frequency characteristics of fluctuating biosignals, making it difficult to effectively process abnormal signal changes.

[0008] Furthermore, while learning artificial intelligence requires a large amount of labeled data, securing such data is difficult in the case of clinical data. In addition, most artificial intelligence models are primarily focused on estimating missing values, which limits their ability to respond to complex artifacts in actual clinical settings. Consequently, their use in actual clinical settings is inevitably limited.

[0009] (Patent Document 1) Republic of Korea Patent Publication No. 10-2022-0165111 (published on December 14, 2022)

[0010] The present invention is intended to solve the above technical problem, and its purpose is to provide a refinement model capable of refining artifacts in real time.

[0011] In another aspect, the present invention aims to provide a training dataset for an artificial intelligence model by generating augmented artifacts by classifying and modeling artifacts based on clinical prior knowledge.

[0012] According to one embodiment of the present invention for solving the above technical problem, a biosignal artifact refinement device may include: a database in which biosignals are stored; an extraction unit for extracting only clean signals from the biosignals; an enhancement unit for enhancing the clean signals to generate enhancement artifacts; a storage unit for storing the clean signals and the enhancement artifacts; a learning unit for learning a refinement model using a dataset generated based on the enhancement artifacts; and a refinement unit for refining artifacts of the biosignals using the refinement model.

[0013] In addition, the augmentation unit can generate the augmented artifact by converting a portion of the clean signal into an artifact using an augmentation model that mathematically models the type of the artifact classified based on clinical prior knowledge.

[0014] In addition, the storage unit includes a clean signal storage unit that stores the clean signal; an augmented artifact storage unit that stores the augmented artifact; and a dataset storage unit that stores the training dataset generated based on the clean signal and the augmented artifact, wherein the dataset can group the augmented artifact, the clean signal, the difference of the clean signal, and the amplification conversion of the clean signal.

[0015] In addition, the learning unit further includes an objective function for calculating the loss of the refined model, and the objective function may be a function that compares an output value output by applying the augmented artifact to the refined model, a difference of the output value, and an amplification conversion of the output value with the clean signal, a difference of the clean signal, and an amplification conversion of the clean signal, respectively.

[0016] In addition, the purification unit performs a first purification by dividing the front and rear areas of a preset unit based on the start part of the artifact in the biosignal, and when the purification is completed, moves for a preset amount of time and then divides the front and rear areas again to perform a second purification, and accumulates the first purification result and the second purification result to purify the biosignal in real time.

[0017] A method for refining biosignal artifacts according to another embodiment of the present invention may include a step of extracting only clean signals from biosignals stored in a database in an extraction unit; a step of generating an augmented artifact by augmenting the clean signal in an augmentation unit; a step of storing the clean signal and the augmented artifact in a storage unit; a step of learning a refined model using a dataset generated based on the augmented artifact in a learning unit; and a step of refining an artifact of the biosignal using the refined model in a refinement unit.

[0018] In addition, the step of generating the augmented artifact may generate the augmented artifact by converting a portion of the clean signal into an artifact using an augmented model that mathematically models the type of the artifact classified based on clinical prior knowledge.

[0019] In addition, the storing step includes a step of storing the clean signal in a clean signal storage unit; a step of storing the augmented artifact in an augmented artifact storage unit; and a step of storing a learning dataset generated based on the clean signal and the augmented artifact in a dataset storage unit, wherein the dataset can group the augmented artifact, the clean signal, the difference of the clean signal, and the amplification conversion of the clean signal.

[0020] In addition, the learning step further includes an objective function for calculating the loss of the refined model, and the objective function may be a function that compares an output value output by applying the augmented artifact to the refined model, a difference of the output value, and an amplification conversion of the output value with the clean signal, a difference of the clean signal, and an amplification conversion of the clean signal, respectively.

[0021] In addition, the refining step may include a step of performing a first refinement by dividing a front and back area of ​​a preset unit based on the start of the artifact in the biosignal, a step of performing a second refinement by moving for a preset amount of time after the refinement is completed and then dividing the front and back area again, and a step of accumulating the first refinement result and the second refinement result.

[0022] The biosignal artifact purification device and method according to one embodiment of the present invention described above can expand the scope of data utilization by purifying previously unusable data and converting it into useful information, and can improve both the quality and quantity of clinical data by reducing data loss.

[0023] Additionally, it can reduce clinical errors and increase the accuracy of clinical diagnosis, thereby improving the quality of diagnostic and treatment decisions made by medical staff, enabling the establishment of personalized treatment plans for each patient.

[0024] Additionally, real-time processing of refined data can simplify medical data management and save the cost and effort of storing and managing large amounts of raw data.

[0025] FIG. 1 is a schematic diagram illustrating a biosignal artifact purification device according to one embodiment of the present invention.

[0026] Figure 2 is a drawing illustrating the types of artifacts.

[0027] Figure 3 is a diagram of the mathematical modeling of the artifact.

[0028] FIG. 4 and FIG. 5 are diagrams illustrating the generation of a dataset of a biosignal artifact refinement device according to one embodiment of the present invention.

[0029] FIG. 6 is a diagram illustrating a learning process of a biosignal artifact refinement device according to one embodiment of the present invention.

[0030] FIG. 7 is a diagram illustrating a process of refining a biosignal artifact of a biosignal artifact refining device according to one embodiment of the present invention.

[0031] FIG. 8 is a flowchart illustrating a biosignal artifact refinement method according to one embodiment of the present invention.

[0032] The present invention can have various modifications and various embodiments, and specific embodiments are illustrated in the drawings and described in detail.

[0033] 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.

[0034] 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 feature, number, step, operation, component, part, or combination thereof described in the specification, but do not preclude the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0035] 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.

[0036] 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.

[0037] One embodiment of the present invention relates to a device and method for refining an artifact of a biosignal using a refinement model. In the following description, an artifact may be defined as an abnormal structure or phenomenon unintentionally generated by external factors or mechanical interference, such as flushing of the arterial line, transient constriction in the arterial line, damping caused by thrombus in the arterial line, movement of a patient, disturbance of the transducer, etc., and may mean a signal different from noise generated due to random fluctuations or errors.

[0038] FIG. 1 is a schematic diagram illustrating a biosignal artifact purification device according to one embodiment of the present invention.

[0039] A biosignal artifact purification device according to one embodiment of the present invention may include a database (100), an extraction unit (200), an augmentation unit (300), a storage unit (400), a learning unit (500), a measurement unit (600), and a purification unit (700).

[0040] The database (100) may be a database that stores and manages information such as bio-signals, patient names, disease names, treatment histories, prescriptions, notes, demographic information, and death status for multiple patients. For example, the database (100) may be an open source data developed and organized by MIT using data generated in the intensive care unit of Beth Israel Deaconess Center, such as MIMIC-III (Medical Information Mart for Intensive Care). However, the database (100) is not limited thereto and any data that measures, stores, and manages bio-signals may be utilized.

[0041] The extraction unit (200) can extract only clean signals, which are pure bio-signals without noise or artifacts, from among the data stored in the database (100), and in particular, can extract signals without artifacts (hereinafter, “clean signals”) from among the atrial invasive blood pressure (ABP) and photoplethysmography (PPG) signals, but is not limited thereto and can use all bio-signals having pulsatile characteristics.

[0042] The augmentation unit (300) can generate an augmentation artifact by adding an artifact to the clean signal extracted from the extraction unit (200), and the augmentation artifact can be generated by simulation based on an augmentation model generated by mathematically modeling the type of artifact classified based on clinical prior knowledge, which will be described in detail in FIGS. 2 to 4 below.

[0043] The storage unit (400) may include a clean signal storage unit (410), an augmented artifact storage unit (420), and a dataset storage unit (430).

[0044] The clean signal storage unit (410) can store the clean signal extracted from the extraction unit (200), and the augmentation artifact storage unit (420) can store the augmentation artifact generated from the augmentation unit (300).

[0045] The dataset storage unit (430) can create a learning dataset for learning a refined model based on clean signals and augmented artifacts.

[0046] A learning dataset (hereinafter, “dataset”) can be generated based on a clean signal, and is described in detail in Figure 5 below.

[0047] In one embodiment of the present invention, the storage unit (400) is described as being divided into a clean signal storage unit (410), an augmented artifact storage unit (420), and a dataset storage unit (430). However, it may also be implemented in a form in which only the dataset is stored in the storage unit (400) after the clean signal and the augmented artifact are input to the dataset generation unit (not shown), or it may be implemented in a form without the dataset storage unit (430) by mapping and storing the data of the clean signal storage unit (410) and the data of the augmented artifact storage unit (420). Any method capable of performing learning of a refined model may be used without limitation.

[0048] The learning unit (500) can learn a refined model based on data stored in the dataset storage unit (430), and the detailed learning method of the learning unit (500) is described in detail in FIG. 6 below.

[0049] The measuring unit (600) can perform biosignal measurement without limitation using one or more of the conventional techniques depending on the biosignal to be measured by the user (clinician or engineer), and since the biosignal is measured using the conventional technique, a detailed description thereof will be omitted.

[0050] The purification unit (700) can refine artifacts of the biosignal in real time using the purification model learned in the learning unit (500) for the biosignal measured in the measurement unit (600).

[0051] The purification unit (700) divides the biosignal measured by the measurement unit (600) into preset units, and can perform a purification operation when an input signal is detected that includes a clean signal of a certain length in front of the point in time when an artifact occurs.

[0052] The specific purification method of the purification unit (700) is described in Fig. 7 below.

[0053] Figures 2 to 4 are drawings that explain the types of artifacts, mathematical modeling, and augmentation methods of artifacts in order to specifically explain augmentation of artifacts.

[0054] FIG. 2 is a drawing illustrating artifact classification for an ABP signal based on data (clinical prior knowledge) of a database (100). In one embodiment of the present invention, five artifacts are classified, but the present invention is not limited thereto and more diverse classifications can be made as needed. It goes without saying that the same classification can be applied to all biosignals having a pulsatile frequency, such as a PPG signal.

[0055] Figure 2 (a) is an example of an artifact caused by saturation to ABP maximum, which can generally occur during the process of cleaning a blocked arterial catheter as a thrombus or coagulation continues to form inside the arterial catheter.

[0056] Figure 2 (b) is an example of an artifact caused by saturation to ABP minimum, which can generally be caused by temporary contraction or compression of the arterial catheter due to arm movement or other external factors.

[0057] Figure 2 (c) is an example of an artifact caused by reduced pulse pressure, which can generally be caused by a thrombus partially blocking an arterial catheter.

[0058] Figure 2 (d) is an example of an artifact caused by amplifying pulse pressure, which may occur momentarily during the process of returning to a normal signal after another artifact occurs.

[0059] Figure 2 (e) is an example of an artifact caused by an impulse, which can generally be caused by patient movement or mechanical problems (cracking of tubing or mechanical damage).

[0060] Figure 3 is a drawing illustrating mathematical modeling of artifacts that may occur as in Figure 2.

[0061] Referring to FIG. 2, artifacts are abnormal signals unintentionally generated by external factors or mechanical interference, as described above, and can be recognized as increasing or decreasing in a certain pattern. In one embodiment of the present invention, considering this point, mathematical modeling can be performed for each artifact.

[0062] Artifacts caused by ABP maximum saturation or ABP minimum saturation can be implemented based on a combination of hyperbolic tangent functions, artifacts caused by pulse pressure decrease and increase can be implemented using a linear function, and artifacts caused by shock can be implemented using a sinc function.

[0063] However, this mathematical modeling is only one example, and if a biosignal can be mathematically modeled and converted into an artificial artifact form, it can be implemented in various ways without limitation, and a detailed description of the composition of the mathematical model is omitted.

[0064] FIG. 4 is a diagram illustrating the generation of a dataset of a biosignal artifact refinement device according to one embodiment of the present invention.

[0065] When the clean signal extracted from the extraction unit (200) is applied to the enhancement unit (300), the enhancement unit (300) can generate an enhancement artifact in which artifacts are combined using a mathematical model as shown in FIG. 3. However, in order to improve the learning efficiency of the refinement model and to match the refining operation in the refinement unit (700), only the last part of the entire clean signal can be converted into an artifact to generate an enhancement artifact.

[0066] For example, referring to FIG. 4, the clean signal extracted from the extraction unit (200) can generate segments of 30 seconds each, copy the last 5 seconds of each segment and set it as a target, and apply a mathematical model to the copied region to generate an augmented artifact converted into an artifact. The target can be used as an actual value (Ground Truth, GT) in the learning process of the learning unit (500) thereafter. In the embodiment, it has been described that segments are generated in 30-second units and the last 5 seconds are converted into artifacts, but this is not limited thereto, and the generation of segments and conversion into artifacts can be performed with a shorter or longer time, and it can be used regardless if the clean signal is cut into segments at a certain time unit and the rear region of the segment of the preset time is converted into an artifact.

[0067] The augmented artifacts generated in this way and the targets of the copied areas can be utilized as a single dataset, and although Fig. 4 shows a form in which only one augmented artifact is generated for each clean signal, it is obvious that a dataset can be generated by generating augmented artifacts of various forms through various mathematical models.

[0068] FIG. 5 is a drawing specifically explaining the generation of a dataset of a biosignal artifact refinement device according to one embodiment of the present invention of FIG. 4.

[0069] Using the example of Fig. 4, a plurality of first batches (batch 1) can be generated using the clean signal extracted from the extraction unit (200), and a second batch (batch 2) can be generated by copying the last part of the generated batch, and the second batch can be utilized as a target.

[0070] In the augmentation unit (300), the first batch can be divided at a preset ratio to generate augmented data (B1) and non-augmented data (B2), which can then be concatenated to generate a sixth batch (batch 6), which is an input signal vector. That is, the sixth batch can include data augmented with multiple types of artifacts and clean data.

[0071] In addition, as will be described below, the learning unit (500) in one embodiment of the present invention may include a plurality of loss functions, and the second batch may perform transformations in three forms of the first to third transformations in order to apply the plurality of loss functions. The first transformation may be the third batch (batch3) generated without transformation of the clean signal as is, the second transformation may be the fourth batch (batch4) generated by performing a difference transformation through difference calculation, and the third transformation may be the fifth batch (batch5) generated by performing an amplification transformation through a fast Fourier transform (FFT). Since the third to fifth batches are generated by transforming the same batch, they may be grouped together and set as a target vector, and may be matched with the sixth batch to form a learning dataset.

[0072] FIG. 6 is a drawing illustrating a learning process of a biosignal artifact refinement device according to one embodiment of the present invention. The left side is a part illustrating a refinement model, and the right side may include a process of optimizing the refinement model using objective functions.

[0073] In one embodiment of the present invention, the refined model may be a deep learning model using an attention function-based encoder-decoder, but is not limited thereto and may be utilized without limitation if it is an artificial intelligence (AI)-based algorithm.

[0074] When the sixth batch is applied to the refinement model, the refinement model divides each batch into frames, generates a frame function, and the encoder can encode the generated frame function into a dense vector with a small dimension. The dense vector can then be stored in the encoder's Long Short Term Memory (LSTM) to generate an attention value. In one embodiment of the present invention, the LSTM may utilize Bi-LSTM, but is not limited thereto.

[0075] The attention value can be a context vector value, and the context vector is applied to the decoder and stored in the decoder's LSTM, and then the attention score is calculated and the value stored in the LSTM is scaled to output the output signal vector, which is the artifact-refined output value, the second batch (batch 2), and in theory, the same batch as the second batch (batch 2), which is the target vector, should be output, so the loss can be calculated using the objective function.

[0076] As described in FIG. 5, in one embodiment of the present invention, the objective function may include first to third loss functions, and in order to apply the first to third loss functions, the second batch (batch 2), which is the target vector, may be converted into the form of the third batch (batch 3) to the fifth batch (batch 5), thereby generating GT values. In addition, in order to apply the first to third loss functions for loss calculation, the second' batch (batch 2'), which is the output value of the refinement model, may be converted into the form of the third' batch (batch 3') to the fifth' batch (batch 5') in the same manner as the third batch (batch 3) to the fifth batch (batch 5).

[0077] The third batch (batch 3) and the third' batch (batch 3') that did not convert the clean signal can calculate the loss using the first loss function, and the first loss function can be configured based on the mean squared error (MSE).

[0078] The fourth batch (batch 4) and the fourth' batch (batch 4') that have undergone differential transformation of the clean signal can calculate the loss using the second loss function, and the second loss function can be configured based on the root mean squared error (RMSE).

[0079] The fifth batch (batch 5) and the fifth' batch (batch 5') of clean signals subjected to fast Fourier transform can calculate losses using the third loss function, and the third loss function can be configured based on the mean squared error (MSE).

[0080] In summary, the objective function for optimizing refined model learning in one embodiment of the present invention may include multiple loss functions, and the multiple loss functions generate a transformed transformed signal for a single clean signal, and apply the clean signal and the transformed signal to multiple loss functions to calculate multiple loss values, thereby learning the refined model from various viewpoints, and thereby significantly improving the robustness of the refined model and the quality of the refined result.

[0081] In addition, it is possible to perform repetitive learning to reduce the loss value, and in the case where learning is performed using all batches but the refined model does not reach the required performance, it is natural that the extraction unit (200) can further improve the performance of the refined model by repeatedly performing the extraction of additional clean signals from the database (100), the generation of augmented artifacts in the augmentation unit (300), and the learning of the learning unit (500).

[0082] In addition, although not disclosed in the drawing, the refined model in one embodiment of the present invention may further include components such as one or more hidden layers, a softmax function, and an embedder.

[0083] FIG. 7 is a diagram illustrating a process of refining a biosignal artifact of a biosignal artifact refining device according to one embodiment of the present invention.

[0084] The process of measuring biosignals must be performed in real time, and even when artifacts occur, the results must be refined in real time to respond to emergencies. However, performing direct artifact refinement over a long period of time requires a lot of hardware resources or results must be output with a certain amount of time delay, which inevitably leads to limitations in use in actual hospital environments.

[0085] In one embodiment of the present invention, in order to perform purification in real time with low resource consumption and at the same time provide accurate artifact purification results, the purification unit (700) can divide and purify the biosignal measured by the measurement unit (600).

[0086] When an artifact is detected by the detection unit (not shown), the purification unit (700) can set a certain range of time before and after the artifact occurrence timing as a purification time area (hereinafter, purification area) that requires purification, and can set an input time area (hereinafter, input area) that is set to input a previous clean signal for a preset time based on the purification area into the purification model. Once the input area is set, the input area can be applied to the purification model to perform a purification operation for the purification area.

[0087] Specifically, the input region can be set to the same length as the segment used for learning in the above-described FIG. 5 (e.g., 30 seconds), and a clean signal of a certain length (e.g., 25 seconds) can be included in the front part based on the time point of occurrence of the artifact, and an artifact can be included in the back part for a length of a preset time range (e.g., 5 seconds), but is not limited thereto, and a clean signal of a preset length must be included in the front part based on the time point of occurrence of the artifact.

[0088] The input region can be set to the same time length as the segment used in learning the refined model, and this is to enable refining artifacts while maintaining the performance of the refined model by setting the input region in the same form as the 6th batch (batch 6) that generated augmented artifacts for learning the refined model.

[0089] Once the input region is configured, artifacts can be refined by applying biosignals from the input region to the refinement model. Once artifact refinement is complete, the refinement model can move the input region for a preset amount of time and then perform refinement operations again on the moved input region.

[0090] In addition, when the movement time of the preset input area is less than 5 seconds, a case may occur where the refined area overlaps with the previous refined area and the subsequent refined area, as shown in FIG. 7. In one embodiment of the present invention, when an overlapping area occurs, the refined results of the overlapping previous refined areas may be removed and the overlapping refinement results may be generated in an overlapping manner, but the present invention is not limited thereto, and various methods, such as a method of setting the average of the overlapping refinement results as the accumulated refinement result, may be utilized.

[0091] FIG. 8 is a flowchart illustrating a biosignal artifact refinement method according to one embodiment of the present invention.

[0092] In order to learn a refinement model that refines biosignal artifacts, the extraction unit (200) can extract only clean signal data from among the data stored in the database (100) (S1100).

[0093] The database (100) used in one embodiment of the present invention may be a database that stores and manages all user-related information, such as bio-signals, patient names, disease names, treatment histories, prescriptions, notes, demographic information, and death status for multiple patients, and the extraction unit (200) may extract only the bio-signals, particularly, clean signals that are pure bio-signals without artifacts and noise, and use them for learning a refinement model.

[0094] The augmentation unit (300) can receive the clean signal extracted from the extraction unit (200) and generate an augmented artifact by augmenting the artifact (S1200).

[0095] Augmented artifacts can be generated by simulation based on an augmented model that mathematically models the types of artifacts classified based on clinical prior knowledge, and can be generated by converting only a preset amount of time from the latter part of the entire input clean signal into an artifact.

[0096] A training dataset for learning a refined model based on clean signals and augmented artifacts can be created (S1300).

[0097] The detailed method of creating a learning dataset is described in detail in Figure 5 above, so it is omitted here.

[0098] Clean signals, augmented artifacts, and datasets can be stored in respective storage locations in the storage unit (400).

[0099] The learning unit (500) can learn a refined model using a dataset (S1400).

[0100] The learning method of the refined model is described in detail in Figure 6 above, so it is omitted here.

[0101] Once the training of the refined model is complete, artifacts in the biosignal can be refined using the trained refined model.

[0102] A user (clinician or engineer) can measure biosignals using one or more of the conventional methods (S1500).

[0103] The refinement model detects the measured biosignal in real time, and when it detects that an artifact has occurred, it sets an input area of ​​a preset range for refinement based on the point of occurrence of the artifact, and applies the input area to the refinement model to refine the artifact (S1600).

[0104] The purification method of the refined model is described in detail in Fig. 7 above, so it is omitted here.

[0105] As described above, the biosignal artifact purification device and method according to one embodiment of the present invention can expand the scope of data utilization by purifying previously unusable data and converting it into useful information, and can improve both the quality and quantity of clinical data by reducing data loss.

[0106] Additionally, it can reduce clinical errors and increase the accuracy of clinical diagnosis, thereby improving the quality of diagnostic and treatment decisions made by medical staff, enabling the establishment of personalized treatment plans for each patient.

[0107] Additionally, real-time processing of refined data can simplify medical data management and save the cost and effort of storing and managing large amounts of raw data.

[0108] The features, 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 just one embodiment. Furthermore, the features, structures, effects, etc. exemplified in each embodiment can be combined or modified in other embodiments by a person with ordinary skill in the art to which the embodiments pertain.

[0109] 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.

[0110] The present invention can be used in a biosignal measuring device in a medical facility such as an emergency room or an operating room, or in a research institute that conducts research using biosignals.

[0111] 100: Database

[0112] 200: Extraction Unit

[0113] 300: Augmentation

[0114] 400: Storage

[0115] 410: Clean signal storage unit

[0116] 420: Augmented Artifact Storage

[0117] 430: Dataset storage

[0118] 500: Learning Department

[0119] 600: Measurement section

[0120] 700: Refining Department

Claims

1. An extraction unit that extracts only clean signals from the above biosignals; An augmentation unit that augments the above clean signal to generate an augmentation artifact; A storage unit for storing the above clean signal and the above augmented artifact; A learning unit that learns a refined model using a data set generated based on the above augmented artifact; and A purification unit that purifies artifacts of the biosignal using the above purification model; A biosignal artifact purification device including:

2. In paragraph 1, The above augmentation part, A biosignal artifact refinement device characterized in that it generates the augmented artifact by converting a portion of the clean signal into an artifact using an augmented model that mathematically models the type of the artifact classified based on clinical prior knowledge.

3. In paragraph 1, The above storage unit A clean signal storage unit that stores the above clean signal; An augmentation artifact storage unit that stores the augmentation artifact; and A dataset storage unit that stores the learning data set generated based on the above clean signal and the above augmented artifact; Includes, A biosignal artifact refinement device characterized in that the above data set groups the augmented artifact, the clean signal, the difference of the clean signal, and the amplification conversion of the clean signal.

4. In paragraph 3, The above learning department It further includes an objective function for calculating the loss of the above refined model, A biosignal artifact refinement device characterized in that the objective function is a function that compares the output value output by applying the augmented artifact to the refined model, the difference of the output value, and the amplification conversion of the output value with the clean signal, the difference of the clean signal, and the amplification conversion of the clean signal, respectively.

5. In paragraph 1, The above purification unit, The first refinement is performed by dividing the pre- and post-region of the preset unit based on the start part of the artifact in the above biosignal, Once the purification is complete, it moves for a preset amount of time and then divides the front and rear areas again to perform a second purification. A biosignal artifact purification device characterized in that the biosignal is purified in real time by accumulating the first purification result and the second purification result.

6. A step of extracting only clean signals from the biosignals stored in the database in the extraction unit; A step of generating an augmentation artifact by augmenting the clean signal in the augmentation unit; A step of storing the clean signal and the augmented artifact in a storage unit; A step of learning a refined model using a data set generated based on the augmented artifact in the learning unit; and A step of refining artifacts of the biosignal using the refining model in the refining unit; A method for refining biosignal artifacts including:

7. In paragraph 6, The step of creating the above augmented artifact is: A biosignal artifact refinement method characterized in that the augmented artifact is generated by converting a portion of the clean signal into an artifact using an augmented model that mathematically models the type of the artifact classified based on clinical prior knowledge.

8. In paragraph 6, The above saving step is, A step of storing the clean signal in the clean signal storage unit; A step of storing the augmented artifact in the augmented artifact storage unit; and A step of storing a learning data set generated based on the clean signal and the augmented artifact in a data set storage unit; Includes, A biosignal artifact refinement method characterized in that the above data set groups the augmented artifact, the clean signal, the difference of the clean signal, and the amplification transformation of the clean signal.

9. In paragraph 8, The above learning steps are: It further includes an objective function for calculating the loss of the above refined model, A biosignal artifact refinement method, characterized in that the objective function is a function that compares an output value output by applying the augmented artifact to the refined model, a difference of the output value, and an amplification conversion of the output value with the clean signal, a difference of the clean signal, and an amplification conversion of the clean signal, respectively.

10. In paragraph 6, The above purifying step is, A step of performing a first refinement by dividing the front and back areas of a preset unit based on the start part of the artifact in the biosignal; A step of performing a second purification by moving for a preset amount of time and then dividing the front and back areas again after the purification is completed; and A step of accumulating the first purification result and the second purification result; A method for refining biosignal artifacts, characterized by including:

Images