Method, apparatus, and program for removing noise from bio-signals

Abstract

The present disclosure provides an apparatus, a method, and a program for removing noise from bio-signals. The method according to an embodiment of the present disclosure comprises the steps of: obtaining a first electrocardiogram dataset, including a plurality of pieces of first electrocardiogram data including noise, and a second electrocardiogram dataset, including a plurality of pieces of clean second electrocardiogram data respectively corresponding to the plurality of pieces of first electrocardiogram data; training a classifier to classify the first electrocardiogram dataset with noise removed therefrom, which is obtained by inputting the first electrocardiogram dataset into a noise removal unit, and the second electrocardiogram dataset; and evaluating the performance of the classifier to obtain a noise removal unit having the optimal performance.

Description

Method, device and program for removing noise in biosignals

[0001] The present disclosure relates to artificial intelligence technology in the medical field, and more particularly, to a method, device, and program for obtaining an optimized method for removing noise in a biosignal by evaluating noise removal performance during the process of learning a deep learning model.

[0002] Meanwhile, recent advancements in information and communication technology have led to the utilization of artificial intelligence (AI) in various fields. In particular, the medical field, which previously relied on specialized and limited resources such as doctors and researchers to diagnose patients' heart disease types, is undergoing various attempts to integrate AI to improve efficiency. For example, research is currently underway on methods for analyzing electrocardiogram data and identifying potential heart disease in patients based on AI technology.

[0003] Meanwhile, biosignal data, such as electrocardiograms, contain various forms of noise due to factors such as the measurement environment, device limitations, and the patient's physiological condition. Noise degrades the quality of training data for deep learning models, becoming a major factor in reducing the accuracy and reliability of results. Therefore, removing noise from biosignals is recognized as an essential process for ensuring the accuracy and reliability of results.

[0004] The present disclosure is conceived in response to the aforementioned background technology, and aims to provide a method, device, and program for generating learning data including electrocardiogram data for contrastive learning.

[0005] However, the problems to be solved in this disclosure are not limited to the problems mentioned above, and other problems not mentioned can be clearly understood based on the description below.

[0006] A method for removing noise in a biosignal, performed by a computing device including at least one processor for realizing a task as described above, comprises the steps of: obtaining a first electrocardiogram data set including a plurality of first electrocardiogram data including noise and a second electrocardiogram data set including a plurality of clean second electrocardiogram data corresponding to the plurality of first electrocardiogram data; inputting the first electrocardiogram data set to a noise removal unit, and training a classifier to classify the first electrocardiogram data set from which noise has been removed and the second electrocardiogram data set; and evaluating the performance of the classifier to obtain a noise removal unit having optimal performance.

[0007] Alternatively, the noise removal unit includes a denoising module in which a plurality of parameters necessary for noise removal are set, and the step of obtaining a noise removal unit having optimal performance includes the steps of calculating a performance index of the classifier, identifying an optimal value for the plurality of parameters based on the obtained performance index, and setting the parameters of the denoising module to the identified optimal value, thereby obtaining a noise removal unit having optimal performance.

[0008] Alternatively, the performance indicator is the AUROC (Area Under the ROC Curve) of the classifier, and when the AUROC is within a preset range, multiple parameters set in the denoising module can be identified as the optimal values.

[0009] Alternatively, the plurality of parameters may be set for at least one of powerline noise removal, Gaussian smoothing, and bandpass filtering.

[0010] Alternatively, the step of obtaining a noise removal unit having optimal performance may include the steps of generating a combination of a plurality of parameters applicable to the denoising module, calculating the performance indicators corresponding to the combinations of the plurality of parameters, and identifying a parameter combination having an optimal value among the plurality of parameter combinations based on the calculated performance indicators.

[0011] Alternatively, the noise removal unit may include a neural network model, and may include a step of training the neural network model to remove noise from the first electrocardiogram data set including noise and the second electrocardiogram data set, based on the first electrocardiogram data set including noise and the second electrocardiogram data set, and training the classifier to classify the first electrocardiogram data set from which the noise has been removed and the second electrocardiogram data set from which the noise has been removed.

[0012] Alternatively, the step of training the neural network model to remove the noise may train the neural network model and the classifier based on an adversarial learning method.

[0013] Alternatively, the step of obtaining a noise removal unit having optimal performance may include the steps of calculating a performance index of the classifier, completing learning of the neural network model based on the calculated performance index, and obtaining the neural network model for which learning has been completed as the noise removal unit having optimal performance.

[0014] Alternatively, the performance indicator is the ACC (Accuracy) of the classifier, and if the ACC is less than or equal to a preset value, the learning of the neural network model can be completed.

[0015] A device for removing noise in a biosignal for realizing the task described above includes a processor including at least one core and a memory including program codes executable by the processor, wherein the processor obtains a first electrocardiogram data set including a plurality of first electrocardiogram data including noise and a second electrocardiogram data set including a plurality of clean second electrocardiogram data corresponding to the plurality of first electrocardiogram data, inputs the first electrocardiogram data set to a noise removal unit, trains a classifier to classify the first electrocardiogram data set with noise removed and the second electrocardiogram data set, and evaluates the performance of the classifier, thereby obtaining a noise removal unit having optimal performance.

[0016] A computer program stored in a computer-readable storage medium for realizing the task described above, wherein the computer program, when executed on one or more processors, performs operations for quantifying the quality of a biosignal, the operations including: obtaining a first electrocardiogram data set including a plurality of first electrocardiogram data including noise and a second electrocardiogram data set including a plurality of clean second electrocardiogram data corresponding to the plurality of first electrocardiogram data; inputting the first electrocardiogram data set to a noise removal unit and training a classifier to classify the first electrocardiogram data set with noise removed and the second electrocardiogram data set; and evaluating the performance of the classifier to obtain a noise removal unit having optimal performance.

[0017] According to a method for removing noise from a biosignal according to one embodiment of the present disclosure, a noise removal unit capable of achieving optimal performance in various environments can be implemented. Specifically, by objectively evaluating the performance of the noise removal unit using a classifier, and by exploring optimized denoising parameters or completing training of a neural network model, the performance of the noise removal unit can be continuously improved.

[0018] FIG. 1 is an exemplary diagram of a computing device for removing noise in a biosignal according to one embodiment of the present disclosure.

[0019] FIG. 2 is a schematic block diagram of a computing device for removing noise in a biosignal according to one embodiment of the present disclosure.

[0020] FIG. 3 is a flowchart schematically illustrating a method for removing noise in a biosignal according to an embodiment of the present disclosure.

[0021] FIG. 4 is an exemplary diagram illustrating a method for obtaining an optimal denoising module according to an embodiment of the present disclosure.

[0022] FIG. 5 is an exemplary diagram illustrating a method for obtaining an optimal denoising model according to an embodiment of the present disclosure.

[0023] FIG. 6 is a block diagram of a computing device according to another embodiment of the present disclosure.

[0024] Below, embodiments of the present disclosure are described in detail with reference to the attached 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 utilize or implement the contents of the present disclosure. Accordingly, 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 implemented in various different forms and is not limited to the embodiments described below.

[0025] Throughout the specification of this disclosure, identical or similar drawing numbers refer to identical or similar components. Furthermore, for the purpose of clearly describing the disclosure, drawing numbers for parts in the drawings that are not relevant to the description of the disclosure may be omitted.

[0026] The term "or" as used herein is intended to mean an inclusive "or" rather than an exclusive "or." That is, unless otherwise specified herein or clear from context, "X employs A or B" should be understood to mean either of its natural inclusive permutations. For example, unless otherwise specified herein or clear from context, "X employs A or B" can be interpreted to mean either X employs A, X employs B, or X employs both A and B.

[0027] The term "and / or" as used herein should be understood to refer to and include all possible combinations of one or more of the related concepts listed.

[0028] The terms "comprises" and / or "comprising" as used herein should be understood to mean the presence of certain features and / or components. However, it should be understood that the terms "comprises" and / or "comprising" do not exclude the presence or addition of one or more other features, other components, and / or combinations thereof.

[0029] Unless otherwise specified in this disclosure or unless the context makes it clear that the singular form is being referred to, the singular should generally be construed to include “one or more.”

[0030] The term "Nth (N is a natural number)" used in this disclosure can be understood as an expression used to distinguish components of this disclosure from each other based on a predetermined standard such as a functional perspective, a structural perspective, or convenience of explanation. For example, components performing different functional roles in this disclosure can be distinguished as a first component or a second component. However, components that are substantially the same within the technical spirit of this disclosure but must be distinguished for convenience of explanation may also be distinguished as a first component or a second component.

[0031] The term "acquisition" as used in this disclosure may be understood to mean not only receiving data through a wired or wireless communication network with an external device or system, but also generating data in an on-device form.

[0032] Meanwhile, the term "module" or "unit" used in the present disclosure can be understood as a term referring to an independent functional unit that processes computing resources, such as a computer-related entity, firmware, software or a part thereof, hardware or a part thereof, or a combination of software and hardware. At this time, the "module" or "unit" may be a unit composed of a single element, or a unit expressed as a combination or set of multiple elements. For example, as a narrow concept, a "module" or "unit" may refer to a hardware element of a computing device or a set thereof, an application program that performs a specific function of software, a processing process implemented through software execution, or a set of instructions for program execution, etc. In addition, as a broad concept, a "module" or "unit" may refer to the computing device itself that constitutes the system, or an application running on the computing device, etc. However, since the above-described concept is only an example, the concept of “module” or “part” may be defined in various ways within a range understandable to those skilled in the art based on the contents of the present disclosure.

[0033] The term "model" as used herein may be understood as a system implemented using mathematical concepts and language to solve a specific problem, a set of software units to solve a specific problem, or an abstract model of a processing process to solve a specific problem. For example, a neural network "model" may refer to the entire system implemented as a neural network that has problem-solving capabilities through learning. In this case, the neural network can have problem-solving capabilities by optimizing the parameters connecting nodes or neurons through learning. A neural network "model" may include a single neural network or a set of neural networks that are a combination of multiple neural networks.

[0034] The term "data" used in this disclosure may include "images," signals, and the like. The term "image" used in this disclosure may refer to multidimensional data composed of discrete image elements. In other words, "image" may be understood as a term referring to a digital representation of an object visible to the human eye. For example, "image" may refer to multidimensional data composed of elements corresponding to pixels in a two-dimensional image. "Image" may refer to multidimensional data composed of elements corresponding to voxels in a three-dimensional image.

[0035] The explanation of the above terms is intended to aid understanding of the present disclosure. Therefore, unless explicitly stated as limiting the contents of the present disclosure, it should be noted that the above terms are not intended to limit the technical ideas of the contents of the present disclosure.

[0036] FIG. 1 is an exemplary diagram of a computing device (100) that removes noise in a biosignal according to one embodiment of the present disclosure.

[0037] A computing device (100) according to an embodiment of the present disclosure may be a hardware device or a part of a hardware device that performs comprehensive processing and calculation of data, or may be a software-based computing environment connected via a communication interface. For example, the computing device (100) may be a server that performs intensive data processing functions and shares resources, or may be a client that shares resources through interaction with a server. In addition, the computing device (100) may be a cloud system in which multiple servers and clients interact to comprehensively process data. Since the above description is only one example related to the type of computing device (100), the type of computing device (100) may be configured in various ways within a category understandable to those skilled in the art based on the contents of the present disclosure. For example, the computing device (100) may be implemented as various electronic devices such as a desktop, a laptop, a smartphone, a server device, a smart watch, a smart band, a smart ring, etc.

[0038] The computing device (100) can acquire data regarding multiple bio-signals. Specifically, the computing device (100) can acquire bio-data regarding multiple patients. Here, the bio-data may be data converted by digital signal processing of a bio-signal measured from a patient by a bio-signal measuring device. The bio-data may include not only information related to the patient's bio-signal, but also biological information or identification information of the patient. The computing device (100) can acquire multiple bio-data regarding various patients and acquire a learning data set including the acquired multiple bio-data. Meanwhile, a bio-signal is a signal including bio-information of the patient that can be measured from the patient, and may include an electrocardiogram (ECG) signal, an electroencephalogram (EEG) signal, body temperature, blood pressure, blood flow, etc. For convenience of explanation, the present disclosure will be described below by assuming that the bio-signal is an ECG signal and the bio-data is ECG data.

[0039] Meanwhile, the computing device (100) can obtain learning data sets by distinguishing them. Specifically, the computing device (100) can obtain a learning data set (hereinafter, a first learning data set (10)) including a plurality of biometric data that includes noise (or the amount of noise included is greater than a preset value) and a learning data set (hereinafter, a second learning data set (20)) including a plurality of biometric data that does not include noise (or the amount of noise included is less than a preset value).

[0040] And, the computing device (100) can input the first learning data set (10) into the noise removal unit (610) to remove noise included in the plurality of biometric data included in the first learning data set (10). And, the computing device (100) can input the first learning data set (10) and the second learning data set (20) from which noise has been removed into the classifier (620) to train the classifier (620) or obtain a classification result. For example, the computing device (100) can input the biometric data included in the first learning data set (10) and the biometric data included in the second learning data set (20) that are matched with the same biometric data into the classifier (620), respectively. And, the computing device (100) can evaluate the performance of the noise removal unit (610) based on the performance of the classifier (620) for which learning has been completed or the classification result of the acquired classifier (620). The computing device (100) can evaluate the performance of the noise removal unit (610) higher as the performance of the classifier (620) is worse or the classification result of the classifier (620) is inaccurate.

[0041] In particular, the computing device (100) can obtain an optimal noise removal unit (610) for removing noise by adjusting the setting parameters of the noise removal unit (610) or by performing training of the neural network model (612) included in the noise removal unit (610) while monitoring the performance or classification result of the classifier (620). Hereinafter, embodiments of the present disclosure related thereto will be described in detail with reference to FIGS. 2 to 6.

[0042] FIG. 2 is a schematic block diagram of a computing device (100) that removes noise in a biosignal according to one embodiment of the present disclosure.

[0043] Referring to FIG. 2, a computing device (100) includes a processor (110) (hereinafter, processor (110)) including at least one core and a memory (120).

[0044] However, since FIG. 2 is only an example, the computing device (100) may further include other configurations for implementing a computing environment. Furthermore, only some of the disclosed configurations may be included in the computing device (100).

[0045] A processor (110) according to an embodiment of the present disclosure may be understood as a configuration unit 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 computational processes such as processing input data 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), or a field programmable gate array (FPGA). The above-described type of processor (110) is only one example, and thus, the type of processor (110) may be configured in various ways within a range understandable to those skilled in the art based on the contents of the present disclosure.

[0046] The processor (110) is electrically connected to other components of the computing device (100) (i.e., memory (120)) and controls the overall operation of the computing device (100).

[0047] The memory (120) according to one embodiment of the present disclosure may be understood as a configuration unit including hardware and / or software for storing and managing data processed in 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 of the computing device (100). For example, the memory (120) may include at least one type of storage medium among a flash memory (120) type, a hard disk type, a multimedia card micro type, a card type memory (120), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory (120), a magnetic disk, and an optical disk. Additionally, the memory (120) may include a database system that controls and manages data in a predetermined system. The types of memory (120) described above are merely examples, and thus, the types of memory (120) may be configured in various ways within a range understandable to those skilled in the art based on the contents of the present disclosure.

[0048] The memory (120) can structure and organize and manage data, combinations of data, and program codes executable by the processor (110) required for the processor (110) to perform operations. For example, the memory (120) can store a classifier (620) and an electrocardiogram data set used to train the classifier (620). In addition, the memory (120) can store a program code for training the classifier (620) based on the electrocardiogram data set or for operating the classifier (620) for which training has been completed, and an optimal noise removal unit (610) generated as the program code is executed.

[0049] FIG. 3 is a flowchart schematically illustrating a method for removing noise in a biosignal according to an embodiment of the present disclosure.

[0050] Referring to FIG. 3, the processor (110) can obtain a first electrocardiogram data set (11) including a plurality of first electrocardiogram data including noise and a second electrocardiogram data set (21) including a plurality of clean second electrocardiogram data corresponding to the plurality of first electrocardiogram data (S310).

[0051] Specifically, the processor (110) can obtain clean electrocardiogram data that does not contain noise and electrocardiogram data that contains noise as a first electrocardiogram data set (11) and a second electrocardiogram data set (21), respectively. At this time, the first electrocardiogram data set (11) and the second electrocardiogram data set (21) may each contain electrocardiogram data that contains noise and clean electrocardiogram data that are matched with the same electrocardiogram data. For example, the computing device (100) can identify whether electrocardiogram data contains noise (or identify whether noise contained in the electrocardiogram data is greater than or equal to a preset value), and include electrocardiogram data that contains noise (or whose amount of noise contained is greater than or equal to the preset value) in the first electrocardiogram data set (11), and include clean electrocardiogram data that does not contain noise (or whose amount of noise contained is less than or equal to the preset value) in the second electrocardiogram data set (21). Alternatively, the processor (110) may add randomly generated noise (e.g., Gaussian noise) to electrocardiogram data that does not include noise included in the second electrocardiogram data set (21) and include the noise in the first electrocardiogram data set (11), or perform preprocessing to remove noise from electrocardiogram data that includes noise included in the first electrocardiogram data set (11) to obtain clean electrocardiogram data and then include the obtained clean electrocardiogram data in the second electrocardiogram data set (21). In addition, the processor (110) may prepare the first electrocardiogram data set (11) and the second electrocardiogram data set (21) in various ways, such as data augmentation and a public data set.

[0052] In addition, the processor (110) can input the first electrocardiogram data set (11) into the noise removal unit (610) and train the classifier (620) to classify the first electrocardiogram data set (11) and the second electrocardiogram data set (21) from which noise has been removed (S320). Here, the noise removal unit (610) may be a configuration that performs a function of detecting and removing noise included in the electrocardiogram data. The noise removal unit (610) may include a denoising module designed to remove noise or a neural network model trained to remove noise.

[0053] The processor (110) can input the first electrocardiogram data set (11) into a noise removal unit (610) to remove noise from a plurality of electrocardiogram data included in the first electrocardiogram data set (11). In addition, the processor (110) can train a classifier (620) using the first electrocardiogram data set (11) from which noise has been removed (more specifically, the first electrocardiogram data set (11) including a plurality of electrocardiogram data from which noise has been removed) and the second electrocardiogram data set (21) including a plurality of electrocardiogram data that does not include noise.

[0054] Here, the classifier (620) may be implemented as a machine learning-based classifier (620) or a deep learning-based classification model. For example, the classifier (620) may be implemented as a multi-layer perceptron (MLP), a convolutional neural network (CNN), a recurrent neural network (RNN), etc.

[0055] The processor (110) can train a classifier (620) using a plurality of electrocardiogram data included in each of the first electrocardiogram data set (11) and the second electrocardiogram data set (21). At this time, the classifier (620) can be trained to distinguish between the electrocardiogram data included in the first electrocardiogram data set (11) (hereinafter, first electrocardiogram data) and the electrocardiogram data included in the second electrocardiogram data set (21) (hereinafter, second electrocardiogram data). Specifically, a plurality of electrocardiogram data (first electrocardiogram data with noise removed included in the first electrocardiogram data set (11) and clean second electrocardiogram data included in the second electrocardiogram data set (21)) are input to the classifier (620), and when the probability values ​​of classes for the plurality of electrocardiogram data are calculated from the classifier (620), a loss function (e.g., binary cross-entropy, categorical cross-entropy, etc.) is used to calculate the loss values ​​of different label values ​​and probability values ​​assigned to the plurality of electrocardiogram data corresponding to the electrocardiogram data set including the plurality of electrocardiogram data, and the weights of the classifier (620) can be updated based on the calculated loss values. By repeating this process, the classifier (620) can be trained to distinguish between the first electrocardiogram data set (11) and the second electrocardiogram data.

[0056] At this time, when the learning of the classifier (620) is completed, the processor (110) can evaluate the performance of the classifier (620) to obtain a noise removal unit (610) having optimal performance. The processor (110) can evaluate the performance of the classifier (620) for which learning has been completed. To this end, the processor (110) can obtain a verification data set (first verification data set) corresponding to the first electrocardiogram data set (11) and a verification data set (second verification data set) corresponding to the second electrocardiogram data set (21). The first and second verification data sets may be composed of a plurality of electrocardiogram data that are not used for learning in the first electrocardiogram data set (11) and the second electrocardiogram data set (21), respectively. The processor (110) can evaluate the performance of the classifier (620) for which learning has been completed using the obtained first verification data set and second verification data set. In particular, the processor (110) can calculate a performance index of the classifier (620), and obtain a noise removal unit (610) that optimally removes noise based on the calculated performance index. At this time, the processor (110) evaluates that the performance of the noise removal unit (610) is better when the performance of the classifier (620) is poor, that is, the lower the performance index, and thus obtains an optimal noise removal unit (610).

[0057] For example, the processor (110) may obtain a plurality of first electrocardiogram data sets (11) from which noise has been removed through a plurality of noise removal units (610), and train a classifier (620) using each of the first electrocardiogram data sets (11) (and the second electrocardiogram data set (21)) from which noise has been removed. Then, the performance indices of the plurality of classifiers (620) from which learning has been completed may be calculated, and based on the calculated performance indices, an optimal noise removal unit (610) may be obtained from the plurality of noise removal units (610) used for learning the plurality of classifiers (620).

[0058] Meanwhile, the plurality of electrocardiogram data constituting the first and second electrocardiogram data sets (11 and 21) are obtained and configured from a plurality of patients having the same or similar biological information, regional characteristics, etc., so that the processor (110) can obtain an optimal noise removal unit (610) for the electrocardiogram data of specific patients.

[0059] FIG. 4 is an exemplary diagram showing a method for obtaining an optimal denoising module (611) according to one embodiment of the present disclosure.

[0060] According to one embodiment of the present disclosure, the noise removal unit (610) may include a denoising module (611) in which a plurality of parameters necessary for noise removal are set. Here, the denoising module (611) may be a module (611) designed to remove noise from electrocardiogram data when electrocardiogram data is input. The denoising module (611) may be set with a plurality of parameters for removing noise. Here, the plurality of parameters are elements necessary for noise removal and may be set based on the type of noise, the size of noise, etc.

[0061] The processor (110) calculates a performance index of the classifier (620), identifies optimal values ​​for a plurality of parameters based on the obtained performance index, and sets the parameters of the denoising module (611) to the identified optimal values, thereby obtaining a noise removal unit (610) having optimal performance. For example, referring to FIG. 4, the plurality of parameters may be at least one of powerline noise removal, Gaussian smoothing, and bandpass filtering. A specific parameter value may be set for each parameter.

[0062] For example, in the case of Gaussian smoothing, parameter values ​​for the Gaussian kernel size and standard deviation may be set, in the case of a band-pass filter, parameter values ​​for the frequency domain may be set, and in the case of power line noise removal, parameter values ​​for the power line frequency and removal range may be set. The processor (110) may identify optimal parameter values ​​for a plurality of parameters based on performance indicators. In addition, the processor (110) may identify an optimal parameter combination to be set in the denoising module (611).

[0063] In this regard, according to one embodiment of the present disclosure, the processor (110) may generate combinations of a plurality of parameters applicable to the denoising module (611), calculate performance indices corresponding to the combinations of the plurality of parameters, and identify a parameter combination having an optimal value among the plurality of parameter combinations based on the calculated plurality of performance indices. Specifically, the processor (110) may train a classifier (620) based on a denoising module (611) that sets a single parameter, and may train a classifier (620) based on a denoising module (611) that sets at least two parameters among the plurality of parameters.

[0064] In this way, by applying parameters in various combinations to the denoising module (611), the processor (110) can train the classifier (620). At this time, the processor (110) can train the classifier (620) by setting not only the combination of parameters but also the parameter values ​​set for the same parameter combination differently. In this way, after assuming various cases in which various parameters are applied, the classifier (620) is trained, and by comparing the performance indices of the classifiers (620) for which each training has been completed, the processor (110) can identify the optimal parameter combination and the optimal parameter values. In particular, as described above, the plurality of electrocardiogram data constituting the first and second electrocardiogram data sets (11 and 21) are obtained and configured from a plurality of patients having the same or similar biological information, regional characteristics, etc., so that the processor (110) can obtain the denoising module (611) as the optimal noise removal unit (610) for the electrocardiogram data of specific patients.

[0065] Meanwhile, if there is an optimal other denoising module already obtained through an electrocardiogram data set corresponding to another object (e.g., a third electrocardiogram data set including noise and a fourth clean electrocardiogram data set) having biological characteristics and regional characteristics similar to the biological characteristics and regional characteristics of the object corresponding to the first and second electrocardiogram data sets (11 and 21), the processor (110) can reduce the time required to identify the optimal parameter combination and values ​​by applying the combination and values ​​of parameters of the other denoising module to the denoising module and then training the classifier. Information related to the other denoising module can be pre-stored in the memory (120). At this time, whether the biological characteristics and regional characteristics are similar can be determined by calculating the similarity, and the similarity can be determined based on the distance value between the feature vectors corresponding to the biological characteristics and the regional characteristics.

[0066] Meanwhile, the performance indicator is the AUROC (Area Under the ROC Curve) of the classifier (620), and the processor (110) can identify multiple parameters set in the denoising module (611) as optimal values ​​when the AUROC is within a preset range. For example, the processor (110) can identify the denoising module (611) used in the classifier (620) having an AUROC of 0.5 or less as the optimal noise removal unit (610).

[0067] FIG. 5 is an exemplary diagram showing a method for obtaining an optimal denoising model (612) according to one embodiment of the present disclosure.

[0068] Meanwhile, according to one embodiment of the present disclosure, the noise removal unit (610) may include a neural network model (612). At this time, the processor (110) may train the neural network model (612) to remove noise from the first electrocardiogram data set (11) including noise based on the first electrocardiogram data set (11) and the second electrocardiogram data set (21) including noise, and may train the classifier (620) to classify the first electrocardiogram data set (11) and the second electrocardiogram data set (21) from which noise has been removed based on the first electrocardiogram data set (11) and the second electrocardiogram data set (21) from which noise has been removed.

[0069] Specifically, the processor (110) can train a neural network model (612) to remove noise from electrocardiogram data. Here, the neural network model (612) can be implemented as an autoencoder, a multi-layer perceptron (MLP), a convolutional neural network (CNN), a recurrent neural network (RNN), etc. The processor (110) can train the neural network model (612) by matching a plurality of electrocardiogram data included in each of the first electrocardiogram data set (11) and the second electrocardiogram data set (21) against the same electrocardiogram data to form a pair of training data.

[0070] At this time, the neural network model (612) can be trained to remove noise from the electrocardiogram data including noise input based on the training data pair. Specifically, the electrocardiogram data including noise among the training data pairs (the electrocardiogram data from which noise has been removed included in the first electrocardiogram data set (11)) is input to the classifier (620), and when the electrocardiogram data restored from the neural network model (612) is output, the output electrocardiogram data and the clean electrocardiogram data not including noise among the training data pairs (the clean electrocardiogram data included in the second electrocardiogram data set (21)) are input to the classifier (620), and a loss value is calculated based on the difference between the classification result and the actual label, and the weights of the neural network model (612) are updated through a backpropagation algorithm based on the loss value. By repeating this, the neural network model (612) is gradually trained in the direction of restoring the input data including noise to be similar to the clean data. Meanwhile, the method for training the classifier (620) is omitted since the description of the present disclosure described above is equally applicable.

[0071] In particular, the processor (110) can train the neural network model (612) and the classifier (620) based on an adversarial learning method. Specifically, the processor (110) can perform training by repeating the process of training the classifier (620) while fixing the weight of the neural network model (612), and then training the neural network model (612) while fixing the weight of the classifier (620). In this process, the neural network model (612) is trained to receive electrocardiogram data containing noise and generate clean electrocardiogram data, and the classifier (620) is trained to compare the generated data with clean electrocardiogram data included in the second electrocardiogram data set (21) and to distinguish between the two data.

[0072] In addition, the processor (110) may set the neural network model (612) as a generator that generates electrocardiogram data with noise removed, and the classifier (620) as a discriminator that discriminates between electrocardiogram data with noise removed generated by the generator and clean electrocardiogram data included in the second electrocardiogram data set (21), thereby performing an adversarial learning method using the first electrocardiogram data set (11) and the second electrocardiogram data set (21). In this process, the generator (neural network model) is trained to generate data that is clean enough to fool the discriminator, and the discriminator (classifier) ​​is trained so that the generated data can be distinguished from the second electrocardiogram data. Through such adversarial learning, the noise removal performance of the neural network model (612) and the discriminative performance of the classifier (620) can be improved simultaneously.

[0073] Meanwhile, as described above, the plurality of electrocardiogram data constituting the first and second electrocardiogram data sets (11 and 21) are obtained and configured from a plurality of patients having the same or similar biological information, regional characteristics, etc., so that the processor (110) can obtain a denoising neural network model (612) with an optimal noise removal unit (610) for the electrocardiogram data of specific patients.

[0074] The processor (110) can calculate a performance index of the classifier (620), and based on the calculated performance index, complete the learning of the neural network model (612), and obtain the learned neural network model (612) with the optimal performance as the noise removal unit (610). That is, the processor (110) can determine the point in time to stop the learning process based on the calculated performance index for the classifier (620). In particular, the performance index is ACC (Accuracy), which is an index indicating the accuracy of the classifier (620), and if the ACC is equal to or lower than a preset value, the learning of the neural network model (612) can be completed. For example, if the ACC of the classifier (620) is equal to or lower than 0.5, the processor (110) can complete the learning of the neural network model (612), and obtain the learned neural network model (612) with the optimal noise removal unit (610).

[0075] Meanwhile, the processor (110) can remove noise from electrocardiogram data by using a denoising module (611) having optimal parameters obtained for the same classifier (620) and a pre-trained denoising neural network model (612). In particular, the processor can remove noise included in electrocardiogram data more accurately by merging and using the denoising module (611) and the pre-trained denoising neural network model (612).

[0076] The processor (110) can store a plurality of optimal denoising modules (611) corresponding to the biological characteristics or regional characteristics of the subject and a plurality of optimal pre-trained denoising neural network models (612) in the memory (120). Therefore, when electrocardiogram data is acquired, the processor can identify the biological characteristics or regional characteristics of the subject of the electrocardiogram data and select and use the corresponding denoising module (611) and denoising neural network model (612).

[0077] FIG. 6 is a block diagram of a computing device (600) according to another embodiment of the present disclosure.

[0078] Referring to FIG. 6, a computing device (600) according to an embodiment of the present disclosure includes a processor (610), a memory (620), a communication interface (630), a sensing unit (640), a display (650), a user interface (660), a camera (670), and a speaker (680). Among the configurations illustrated in FIG. 6, the processor (610) and the memory (620) correspond to the configurations of the processor (110) and the memory (120) of the computing device (100) illustrated in FIG. 2, and thus a detailed description thereof will be omitted.

[0079] A communication interface (630) according to an embodiment of the present disclosure may be understood as a component that transmits and receives data through any known wired or wireless communication system. For example, the communication interface (630) may perform data transmission and reception 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), fifth generation mobile communication (5G), ultrawide-band, ZigBee, radio frequency (RF) communication, wireless LAN, wireless fidelity, near field communication (NFC), or Bluetooth. Since the above-described communication systems are only examples, the wired and wireless communication system for data transmission and reception of the communication interface (630) may be applied in various ways other than the above-described examples.

[0080] The communication interface (630) can receive data necessary for the processor (110) to perform calculations through wired or wireless communication with any system or any client, etc. In addition, the communication interface (630) can transmit data generated through calculations of the processor (110) through wired or wireless communication with any system or any client, etc. For example, the communication interface (630) can receive medical data through communication with a cloud server that performs tasks such as databases in a hospital environment, standardization of medical data, or a computing device (600). The communication interface (630) can transmit output data of a neural network model, intermediate data derived from the calculation process of the processor (110), processed data, etc. through communication with the aforementioned database, server, or computing device (600). For example, the processor (110) can obtain a plurality of electrocardiogram data for each subject's biometric data from an external computing device (e.g., an external server device or an external biosignal measuring device) through the communication interface (630).

[0081] The sensing unit (640) can obtain biometric data of the subject. For example, the sensing unit (640) can include a plurality of electrodes (e.g., 12 leads). In this case, the processor (110) can obtain the user's electrocardiogram signal as biometric data through at least one electrode. In addition, the sensing unit (640) can include an image sensor or an optical sensor. In this case, the processor (110) can obtain the user's optical blood flow signal as biometric data through the image sensor (or optical sensor).

[0082] The display (650) can display various images. Here, the images include both still images and moving images. The display (650) can output guide information regarding activities generated based on the user status. The display (650) can be implemented as various types of displays, such as an LCD (Liquid Crystal Display Panel), an OLED (Organic Light Emitting Diodes), an LCoS (Liquid Crystal on Silicon), a DLP (Digital Light Processing), etc. In addition, the display (650) can also include a driving circuit, a backlight unit, etc., which can be implemented in a form such as an a-si TFT, an LTPS (low temperature poly silicon) TFT, an OTFT (organic TFT), etc.

[0083] Meanwhile, the display (650) may be implemented as a touch screen by combining with a touch panel, and in this case, the display (650) may perform the function of an input interface that receives a user's touch input as well as an output interface that outputs an image through the touch screen.

[0084] The user interface (660) is a component used by the computing device (600) to perform interaction with the user, and may include at least one of a touch sensor, a motion sensor, a button, a jog dial, and a switch, but is not limited thereto. The processor (610) may receive the user's biological information (occupation, age, gender, etc.) through the user interface (660).

[0085] The camera (670) captures images of objects around the user and obtains images of the objects. Specifically, the camera (670) can capture images of food consumed by the user. At this time, the processor (610) can determine the nutritional status of the user based on the user's condition and the image of the food consumed by the user and provide recommended dietary information related to heart disease as guide information. To this end, the camera (670) can be implemented with an imaging device such as an imaging device having a CMOS structure (CIS, CMOS Image Sensor) or an imaging device having a CCD structure (Charge Coupled Device). However, the present invention is not limited thereto, and the camera (670) can be implemented with a camera module having various resolutions capable of capturing an object. Meanwhile, the camera (670) can be implemented with a depth camera (e.g., an IR depth camera), a stereo camera, an RGB camera, etc.

[0086] The speaker (680) is a component that outputs various audio data on which various processing operations such as decoding, amplification, and noise filtering have been performed by an audio processing unit (not shown). The speaker (680) can output various notification sounds or voice messages. According to one embodiment of the present disclosure, the processor (610) can convert an electrical signal received from an external device into a user voice and output it through the speaker (680). For example, the speaker (680) can output a voice message warning of or suggesting diagnosis of a heart disease based on the judgment result on the possibility of a heart disease identified through a neural network model.

[0087] The various embodiments of the present disclosure described above can be combined with additional embodiments and modified within the scope understood by those skilled in the art in light of the detailed description above. It should be understood that the embodiments of the present disclosure are illustrative in all respects and not restrictive. For example, each component described as a single component may be implemented in a distributed manner, and likewise, components described as distributed may be implemented in a combined manner. Accordingly, all changes or modifications derived from the meaning, scope, and equivalent concepts of the claims of the present disclosure should be construed as being included within the scope of the present disclosure.

Claims

1. A method for removing noise in a biosignal, performed by a computing device including at least one processor, A step of obtaining a first electrocardiogram data set including a plurality of first electrocardiogram data including noise and a second electrocardiogram data set including a plurality of clean second electrocardiogram data; A step of training a classifier to classify the first electrocardiogram data set obtained by inputting the first electrocardiogram data set to a noise removal unit and the second electrocardiogram data set from which noise has been removed; and A step of evaluating the performance of the above classifier and obtaining a noise removal unit having optimal performance; Including, method.

2. In paragraph 1, The above noise removal unit, Includes a denoising module with multiple parameters set for noise removal, The steps for obtaining a noise removal unit with optimal performance are: A step of calculating a performance index of the classifier and identifying an optimal value for the plurality of parameters based on the obtained performance index; and A step of setting the parameters of the denoising module to the identified optimal values ​​to obtain a noise removal unit having the optimal performance; Including, method.

3. In paragraph 2, The above performance indicators are, The AUROC (Area Under the ROC Curve) of the above classifier is identified, and when the AUROC is within a preset range, multiple parameters set in the denoising module are identified as the optimal values. method.

4. In paragraph 2, The above multiple parameters are, At least one of powerline noise removal, Gaussian smoothing and bandpass filtering is set. method.

5. In paragraph 2, The steps for obtaining a noise removal unit with optimal performance are: A step of generating a combination of a plurality of parameters applicable to the denoising module and calculating the performance indicators corresponding to the combination of the plurality of parameters; and A step of identifying a parameter combination having an optimal value among the plurality of parameter combinations based on the plurality of performance indicators produced above; Including, method.

6. In paragraph 1, The above noise removal unit, Contains a neural network model, The step of training the above classifier is: A first learning step for training the neural network model to remove noise from the electrocardiogram data based on a first electrocardiogram data set including noise, and a second learning step for training the classifier based on the first electrocardiogram data set from which noise has been removed by the neural network model and the second electrocardiogram data set. method.

7. In paragraph 6, The step of training the neural network model to remove the above noise is: Training the neural network model and the classifier based on the adversarial learning method, method.

8. In paragraph 6, The steps for obtaining a noise removal unit with optimal performance are: A step of calculating a performance index of the classifier, completing learning of the neural network model based on the calculated performance index, and obtaining the neural network model for which learning has been completed as a noise removal unit having the optimal performance; Including, method.

9. In paragraph 8, The above performance indicators are, The ACC (Accuracy) of the above classifier is, and if the ACC is less than or equal to a preset value, the learning of the neural network model is completed. method.

10. In a device for removing noise in a biological signal, a processor comprising at least one core; and A memory including program codes executable by the processor; The above processor, Obtaining a first electrocardiogram data set including a plurality of first electrocardiogram data including noise and a second electrocardiogram data set including a plurality of clean second electrocardiogram data corresponding to the plurality of first electrocardiogram data, inputting the first electrocardiogram data set to a noise removal unit, training a classifier to classify the first electrocardiogram data set with noise removed and the second electrocardiogram data set, and evaluating the performance of the classifier to obtain a noise removal unit having optimal performance. device.

11. A computer program stored in a computer-readable storage medium, wherein the computer program, when executed on one or more processors, performs operations for quantifying the quality of a biosignal. The above actions are, An operation of obtaining a first electrocardiogram data set including a plurality of first electrocardiogram data including noise and a second electrocardiogram data set including a plurality of clean second electrocardiogram data corresponding to the plurality of first electrocardiogram data; An operation of training a classifier to classify the first electrocardiogram data set obtained by inputting the first electrocardiogram data set to a noise removal unit and the second electrocardiogram data set from which noise has been removed; and An operation of evaluating the performance of the above classifier and obtaining a noise removal unit having optimal performance; Including, Computer program.

Images