Device and method for estimating renal function recovery using renal function recovery estimation model

Abstract

Disclosed are a device and method for estimating renal function recovery using a renal function recovery estimation model. The device for estimating renal function recovery using a renal function recovery estimation model may comprise: a data collection unit that receives vital sign data including user information about a user whose renal function recovery is to be estimated and biometric information about the user; a data preprocessing unit that replaces a missing value of the vital sign data with another value; a feature extraction unit that extracts feature data for renal function evaluation from the vital sign data in which the missing value has been replaced with the other value; and a renal function recovery estimation unit that estimates whether the renal function of the user is recovered on the basis of an output value output from a federated learning-based renal function recovery estimation model to which the feature data is input.

Description

Device and method for estimating renal function recovery using a renal function recovery estimation model

[0001] The following disclosure relates to a device and method for estimating renal function recovery using a renal function recovery estimation model.

[0002] Acute kidney injury (AKI) is a sudden loss of kidney function. The development of AKI increases healthcare costs, hospitalization duration, the rate of complications during hospitalization, and mortality. AKI is a common condition, occurring in 10-15% of hospitalized patients and 50-60% of intensive care unit patients. However, there is no clear definition of renal function recovery after AKI. To predict renal function recovery, the time required for renal function recovery or the results of biopsy data can be used.

[0003] According to one embodiment, a device for estimating kidney function recovery using a kidney function recovery estimation model may include a data collection unit that receives bio-sign data including user information of a user who is a target of kidney function recovery estimation and bio-information about the user, a data preprocessing unit that replaces missing values ​​in the bio-sign data with other values, a feature extraction unit that extracts feature data for evaluating kidney function from the bio-sign data in which missing values ​​are replaced with other values, and a kidney function recovery estimation unit that estimates whether or not the user's kidney function has recovered based on an output value output from a kidney function recovery estimation model based on associative learning into which feature data is input.

[0004] The feature data may include at least user information and the user's serum creatinine concentration among the data included in the vital signs data.

[0005] The data preprocessing unit can replace missing data with previous data corresponding to the missing data or replace missing data based on multiple imputation by chained equations (MICE) when the ratio of missing data in the vital signs data is below a threshold value.

[0006] The data preprocessing unit can replace missing data with a missing indicator if the proportion of missing data in the vital signs data exceeds a threshold.

[0007] The feature extraction unit can extract feature data based on LASSO (least absolute shrinkage and selection operator; LASSO), SHAP (shapley additive explanations)-value, and P-value.

[0008] The kidney function recovery estimation unit can estimate that the user's kidney function has recovered if the output value is less than the threshold output value, and can estimate that the user's kidney function has not recovered if the output value is greater than the threshold output value.

[0009] According to one embodiment, a learning device for learning a kidney function recovery estimation model may include a learning data collection unit that receives learning bio-sign data including user information of a user who is a target of kidney function recovery estimation and bio-information about the user, a data preprocessing unit that replaces missing values ​​in the learning bio-sign data with other values, a feature extraction unit that extracts feature data for evaluating kidney function from the learning bio-sign data in which missing values ​​are replaced, and a model learning unit that trains the kidney function recovery estimation model using output values ​​output from the kidney function recovery estimation model into which feature data is input and recovery estimation result values ​​corresponding to the learning bio-sign data.

[0010] The feature data may include at least user information and the user's serum creatinine concentration among the data included in the learning biosignature data.

[0011] The data preprocessing unit can replace missing data with previous data corresponding to the missing data or replace missing data based on multiple imputation by chained equations (MICE) when the proportion of missing data in the learning biosign data is below a threshold value.

[0012] The data preprocessing unit can replace missing data with a missing indicator if the proportion of missing data in the vital signs data exceeds a threshold.

[0013] The feature extraction unit can extract feature data based on LASSO (least absolute shrinkage and selection operator; LASSO), SHAP (shapley additive explanations)-value, and P-value.

[0014] The kidney function recovery estimation model may be a federated learning model based on soft voting or stacking of individual kidney function recovery estimation models.

[0015] The individual renal function recovery estimation model may be a model based on at least one of a logistic regression (LR) model, a random forest (RF) model, an extreme gradient boosting (XGBoost) model, a light gradient boosting machine (LightGBM) model, and a CatBoost model.

[0016] According to one embodiment, a method for estimating kidney function recovery using a kidney function recovery estimation model may include an operation of receiving bio-sign data including user information of a user who is a target of kidney function recovery estimation and bio-information about the user, an operation of replacing missing values ​​in the bio-sign data with other values, an operation of extracting feature data for evaluating kidney function from the bio-sign data in which the missing values ​​are replaced with other values, and an operation of estimating whether the user's kidney function has recovered based on an output value output from a kidney function recovery estimation model based on associative learning into which the feature data is input.

[0017] The replacement operation may include replacing missing data with previous data corresponding to the missing data or replacing missing data based on multiple imputation by chained equations (MICE) when the proportion of missing data in the biosignal data is below a threshold.

[0018] The replacing action may include replacing missing data with a missing indicator when the proportion of missing data in the vital signs data exceeds a threshold.

[0019] The operation of extracting feature data may include an operation of extracting feature data based on LASSO (least absolute shrinkage and selection operator; LASSO), SHAP (shapley additive explanations)-value, and P-value.

[0020] The estimating action may include an action of estimating that the user's renal function has recovered if the output value is less than a threshold output value, and an action of estimating that the user's renal function has not recovered if the output value is greater than or equal to the threshold output value.

[0021] According to one embodiment, a learning method for learning a kidney function recovery estimation model may include an operation of receiving learning bio-sign data including user information of a user who is a target of kidney function recovery estimation and bio-information about the user, an operation of replacing missing values ​​in the learning bio-sign data with other values, an operation of extracting feature data for evaluating kidney function from the learning bio-sign data in which the missing values ​​are replaced, and an operation of training the kidney function recovery estimation model using an output value output from the kidney function recovery estimation model into which the feature data is input and a recovery estimation result value corresponding to the learning bio-sign data.

[0022] FIG. 1 is a block diagram illustrating a kidney function recovery estimation device according to one embodiment.

[0023] FIG. 2 is a block diagram illustrating a learning device for learning a kidney function recovery estimation model according to one embodiment.

[0024] FIG. 3 is a diagram illustrating an individual renal function recovery estimation model according to one embodiment.

[0025] FIG. 4 is a diagram illustrating training a renal function recovery estimation model based on federated learning according to one embodiment.

[0026] FIG. 5 is a flowchart illustrating operations of a method for estimating kidney function recovery according to one embodiment.

[0027] FIG. 6 is a flowchart illustrating operations of a learning method for learning a kidney function recovery estimation model according to one embodiment.

[0028] Specific structural or functional descriptions of the embodiments are disclosed for illustrative purposes only and may be modified and implemented in various forms. Therefore, the actual implementation is not limited to the specific embodiments disclosed, and the scope of this specification includes modifications, equivalents, or alternatives within the technical concepts described in the embodiments.

[0029] Although terms such as "first" or "second" may be used to describe various components, these terms should be interpreted solely to distinguish one component from another. For example, a first component may be referred to as a second component, and similarly, a second component may also be referred to as a first component.

[0030] When it is said that a component is "connected" to another component, it should be understood that it may be directly connected or connected to that other component, but there may also be other components in between.

[0031] Singular expressions include plural expressions unless the context clearly dictates otherwise. In this document, phrases such as "A or B", "at least one of A and B", "at least one of A or B", "A, B, or C", "at least one of A, B, and C", and "at least one of A, B, or C" can each include any one of the items listed together in that phrase, or all possible combinations thereof. In this specification, it should be understood that the terms "comprises" or "has" and the like are intended to specify the presence of a described feature, number, step, operation, component, part, or combination thereof, but do not exclude the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0032] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art. Terms defined in commonly used dictionaries should be interpreted to have 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.

[0033] The term "~part" as used in this document refers to a software or hardware component such as an FPGA or ASIC, and the "~part" performs certain roles. However, the "~part" is not limited to software or hardware. The "~part" may be configured to reside on an addressable storage medium and may be configured to execute one or more processors. For example, the "~part" may include components such as software components, object-oriented software components, class components, and task components, as well as processes, functions, attributes, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. The functionality provided within the components and "~parts" may be combined into a smaller number of components and "~parts" or further separated into additional components and "~parts." Furthermore, the components and "~parts" may be implemented to execute one or more CPUs within a device or a secure multimedia card. Additionally, '~bu' may include one or more processors.

[0034] Hereinafter, embodiments will be described in detail with reference to the attached drawings. In the description with reference to the attached drawings, identical components are assigned the same reference numerals regardless of the drawing numbers, and redundant descriptions thereof will be omitted.

[0035]

[0036] FIG. 1 is a block diagram illustrating a kidney function recovery estimation device according to one embodiment.

[0037] Referring to FIG. 1, a renal function recovery estimation device (100) may include a data collection unit (110), a data preprocessing unit (120), a feature extraction unit (130), and a renal function recovery estimation unit (140). The renal function recovery estimation device (100) may include a memory (not shown) and a processor (not shown), and the processor may perform operations of the data collection unit (110), the data preprocessing unit (120), the feature extraction unit (130), and the renal function recovery estimation unit (140).

[0038] The memory can store instructions that can be executed by the processor. The memory can include a separate device, such as an external disk drive, a storage array, or other storage device accessible to the database system. The memory and the processor can be operatively coupled or can communicate with each other via an I / O port, a network connection, or the like, such that the processor can read a file stored in the memory. The memory can be a computer-readable storage medium that stores instructions, and the instructions stored in the memory, when executed by the processor, can prompt at least one processor to execute an image processing method or a method for training an image processing model.

[0039] Computer-readable storage media include read-only memory (ROM), programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, nonvolatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, BLU-RAY or optical disk memory, hard disk drive (HDD), solid state drive (SSD), card memory (e.g., multimedia card, secure digital (SD) card, or extreme digital (XD) card), magnetic tape, It may include floppy disks, magneto-optical data storage devices, optical data storage devices, hard disks, solid state disks, and other devices.

[0040] The processor can execute instructions stored in memory. The processor may include a central processing unit (CPU), a graphics processing unit (GPU), a neural network processing unit (NPU), a media processing unit (MPU), a data processing unit (DPU), a vision processing unit (VPU), a video processor, an image processor, a display processor, a microprocessor, a processor core, a multi-core processor, an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or any combination thereof. Hereinafter, the operations of the data collection unit (110), the data preprocessing unit (120), the feature extraction unit (130), and the renal function recovery estimation unit (140) performed by the processor will be described.

[0041] The data collection unit (110) may receive user information and vital sign data of a user who is a subject of estimation of kidney function recovery. The vital sign data may include biometric information about the user. According to one embodiment, the data collection unit (110) may receive user information and vital sign data based on test results for the user who is a subject of estimation of kidney function recovery. For example, the data collection unit (110) may receive vital sign data based on the results of a blood test, a lab test (e.g., a urine test), or a vital sign test (e.g., body temperature, blood pressure, pulse, etc.) for the user. The vital sign data acquired by the data collection unit (110) according to one embodiment may be data summarizing test results related to kidney disease at 24-hour intervals. For example, if multiple tests related to kidney disease are performed over a 24-hour period, the vital sign data may be data summarized including the maximum value, average value, minimum value, and number of measurements for the measured variables. The vital sign data may include data within a time period that affects the user who is a subject of estimation of kidney function recovery. For example, only data related to medication prescriptions (e.g., nephrotoxic antibiotic prescriptions, nonsteroidal anti-inflammatory drug prescriptions, or cytotoxic chemotherapy prescriptions), vascular imaging test data, overall anesthesia surgery data, contrast-enhanced computed tomography data, and intensive care unit transfer may be included in the vital signs data if they are dated within 7 days of the corresponding measurement time. In one embodiment, the vital signs data may include estimated glomerular filtration rate (eGFR) data. The estimated glomerular filtration rate may be used as an indicator for evaluating kidney function. The estimated glomerular filtration rate may be determined based on a creatinine equation (e.g., CKD-EPI 2021 creatinine equation). The vital signs data may include data related to creatinine.For example, vital signs data may include the amount of onset creatinine and the amount of baseline creatinine as variables, and may further include the difference between the amount of onset creatinine and the amount of baseline creatinine as variables.

[0042] The data preprocessing unit (120) can replace missing values ​​in the vital sign data with other values. The vital sign data acquired by the data collection unit (110) may have missing values ​​for measurement times or specific variables. Alternatively, the data preprocessing unit (120) may determine that specific data included in the vital sign data is a missing value if it exceeds a defined data range (or, if undefined, a data range of 2.5% to 97.5% of the entire interval).

[0043] The data preprocessing unit (120) can replace missing values ​​with other values. For example, the data preprocessing unit (120) can replace missing values ​​based on the proportion of missing values ​​in the vital sign data. If the proportion of missing data in the vital sign data is below a threshold value (e.g., 20%), the data preprocessing unit (120) can replace the missing data with previous data corresponding to the missing data or replace the missing data based on multiple imputation by chained equations (MICE). If the proportion of missing data in the vital sign data exceeds the threshold value, the data preprocessing unit (120) can replace the missing data with a missing indicator. When replacing missing data based on the corresponding previous data or MICE, the continuity of the vital sign data can be maintained, and when replacing missing data using a missing indicator, the missing values ​​can be identified and distinguished.

[0044] The feature extraction unit (130) can extract feature data for evaluating kidney function from vital sign data in which missing values ​​are replaced with other values. For example, the feature extraction unit (130) can extract feature data from the vital sign data based on the least absolute shrinkage and selection operator (LASSO), Shapley additive explanations (SHAP)-value, and P-value. The vital sign data can include features including at least basic patient information, vital measurements, and test results, and the feature data extracted from the vital sign data can include user information and the user's serum creatinine concentration. The feature extraction unit (130) can determine regression coefficients and Shapley additive explanations (SHAP) values ​​for all features included in the vital sign data based on a defined feature selection process. The feature extraction unit (130) can extract feature data for evaluating renal function based on a stepwise method that removes unnecessary variables by applying LASSO and logistic regression to determined values, and a correlation coefficient and missing value ratio between features.

[0045] The renal function recovery estimation unit (140) can estimate whether the user's renal function has recovered based on an output value output from a renal function recovery estimation model based on associative learning into which feature data has been input. The renal function recovery estimation unit (140) can estimate that the user's renal function has recovered if the output value of the renal function recovery estimation model is less than a threshold output value, and can estimate that the user's renal function has not recovered if the output value of the renal function recovery estimation model is greater than or equal to the threshold output value. For example, the threshold output value of the renal function recovery estimation model can be defined as 0.7. The renal function recovery estimation model can output a probability value of 0.4 if the creatinine concentration decreases by 33% or more of the creatinine concentration at the time of onset of acute renal failure within 7 days from the onset of acute renal failure, and the renal function recovery estimation unit (140) can estimate that the renal function has recovered.

[0046] Alternatively, the renal function recovery estimation model may output a probability value of 0.4 as an output value when the creatinine concentration is below the reference value used for diagnosing acute renal failure, and the renal function recovery estimation unit (140) may estimate that renal function has recovered. The renal function recovery estimation model based on associative learning will be described in more detail in FIGS. 2 to 4.

[0047] A device for estimating renal function recovery (100) can predict whether renal function is recovered after acute renal failure occurs. Acute renal failure can be diagnosed through the concentration of serum creatinine and the results of urine tests, and if symptoms of acute renal failure appear for more than 7 days after the onset of acute renal failure, it can be diagnosed as acute kidney disease. In order to diagnose acute kidney disease, it is necessary to confirm that symptoms of acute renal failure appear for more than 7 days, but the device for estimating renal function recovery (100) can be used to diagnose acute kidney disease by estimating whether renal function is recovered in advance (or early) at the time of the onset of acute renal failure.

[0048]

[0049] FIG. 2 is a block diagram illustrating a learning device for learning a kidney function recovery estimation model according to one embodiment.

[0050] Referring to FIG. 2, the learning device (200) may include a learning data collection unit (210), a data preprocessing unit (220), a feature extraction unit (230), and a model learning unit (240). The data preprocessing unit (220) and the feature extraction unit (230) correspond to the data preprocessing unit (120) and the feature extraction unit (130) of FIG. 1, and any further description thereof will be omitted. The learning device (200) may include a memory (not shown) and a processor (not shown), and the processor may perform the operations of the learning data collection unit (210), the data preprocessing unit (220), the feature extraction unit (230), and the model learning unit (240). The memory may store instructions that may be performed by the processor. The memory may include a separate device such as an external disk drive, a storage array, or another storage device usable by a database system. The memory and the processor may be operatively coupled or may communicate with each other via an I / O port, a network connection, or the like, such that the processor can read a file stored in the memory. The memory may be a computer-readable storage medium storing instructions, and the instructions stored in the memory, when executed by the processor, may prompt at least one processor to execute an image processing method or a method for training an image processing model.

[0051] Computer-readable storage media include read-only memory (ROM), programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, nonvolatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, BLU-RAY or optical disk memory, hard disk drive (HDD), solid state drive (SSD), card memory (e.g., multimedia card, secure digital (SD) card, or extreme digital (XD) card), magnetic tape, It may include floppy disks, magneto-optical data storage devices, optical data storage devices, hard disks, solid state disks, and other devices.

[0052] The processor can execute instructions stored in memory. The processor may include a central processing unit (CPU), a graphics processing unit (GPU), a neural network processing unit (NPU), a media processing unit (MPU), a data processing unit (DPU), a vision processing unit (VPU), a video processor, an image processor, a display processor, a microprocessor, a processor core, a multi-core processor, an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or any combination thereof. Hereinafter, the operations of the learning data collection unit (210) and the model learning unit (240) performed by the processor will be described.

[0053] The learning data collection unit (210) may receive learning bio-sign data including user information of a user who is a target of renal function recovery estimation and bio-information about the user. The learning data may include learning bio-sign data and label data for a renal function recovery estimation model. The label data is data indicating whether renal function has recovered. For example, the label data may indicate 0 if renal function has recovered and 1 if renal function has not recovered. The learning bio-sign data corresponds to the bio-sign data of FIG. 1, and any overlapping description will be omitted.

[0054] The model learning unit (240) can learn a renal function recovery estimation model. The model learning unit (240) can learn a renal function recovery estimation model using an output value output from a renal function recovery estimation model into which feature data is input and a recovery estimation result value corresponding to learning vital sign data. The model learning unit (240) can obtain a recovery estimation result value corresponding to learning vital sign data from an individual renal function recovery estimation model. The recovery estimation result value may correspond to an output value of an individual renal function recovery estimation model into which learning vital sign data is input. The individual renal function recovery estimation model may include a tree-based decision model and may be a model based on at least one of a logistic regression (LR) model, a random forest (RF) model, an extreme gradient boosting (XGBoost) model, a light gradient boosting machine (LightGBM) model, and a CatBoost model. The individual renal function recovery estimation model may include hyperparameters, and the hyperparameters may include a learning rate, a number of local epochs, a size of a minibatch, and a communication cycle between a local device and a central server. The hyperparameters may be adjusted through cross-validation. The model learning unit (240) may train the renal function recovery estimation model based on federated learning. Federated learning is a machine learning method that trains a machine learning model of a local device and then trains the machine learning model of the central server using the learning results of the local device, without collecting learning data distributed on the local devices to the central server. According to one embodiment, the model learning unit (240) may train the renal function recovery estimation model based on soft voting or stacking for the individual renal function recovery estimation model.For example, the model learning unit (240) can determine the average of the output values ​​of individual renal function recovery estimation models based on soft voting, and train the renal function recovery estimation model using the determined average value. Alternatively, the model learning unit (240) can train the renal function recovery estimation model using the output values ​​of individual renal function recovery estimation models based on stacking. As described above, the model learning unit (240) can train the renal function recovery estimation model without sharing the training data input to the individual renal function recovery estimation models. This can save data storage space and improve data security. The training of the renal function recovery estimation model by the model learning unit (240) will be described in more detail with reference to FIG. 4.

[0055] FIG. 3 is a diagram illustrating an individual renal function recovery estimation model according to one embodiment.

[0056] Referring to Fig. 3, the individual renal function recovery estimation model into which learning data is input can output predicted values ​​in parallel. For example, the first individual renal function recovery estimation model (310) into which the first learning data (305) is input can output predicted values. (311) and (313) and the second individual renal function recovery estimation model, into which the second learning data (306) is input, uses the learning data to predict the value. (321) and (323) can be printed.

[0057] The first learning data (305) and the second learning data (306) may be learning bio-sign data including user information of a user who is a target of kidney function recovery estimation and bio-information about the user, and may include label data.

[0058] The first learning data (305) and the second learning data (306) used to train the individual renal function recovery estimation model can be merged into a full learning data set to train the renal function recovery estimation model. The full learning data set can be used to train the renal function recovery estimation model through cross-validation.

[0059] The first learning data (305) and the second learning data (306) may have missing values. In order for the first learning data (305) and the second learning data (306) to be merged into the entire learning data, the missing values ​​must be replaced, and the missing values ​​can be replaced through a data preprocessing process. The data preprocessing process can be performed by a local device (not shown) corresponding to each individual renal function recovery estimation model, or by a data preprocessing unit (e.g., the data preprocessing unit (220) of FIG. 2) of the renal function recovery estimation model. For example, the first learning data (305) may include features a, b, and c, and the second learning data (306) may include features a and c. In this case, the second learning data (306) may replace the missing value corresponding to the b feature of the first learning data (305) through data preprocessing. Replacing missing values ​​by performing data preprocessing is described in detail in Fig. 1, so redundant description will be omitted.

[0060] The individual renal function recovery estimation model may be based on a decision-making model, as described in FIG. 2, and may be a model based on at least one of a neural network model, a logistic regression (LR) model, a random forest (RF) model, an extreme gradient boosting (XGBoost) model, a light gradient boosting machine (LightGBM) model, and a CatBoost model. For example, the first individual renal function recovery estimation model (310) may be a CatBoost model, and the second individual renal function recovery estimation model (320) may be a random forest model.

[0061] Predicted value (311) is an output value output from the first individual renal function recovery estimation model (310) into which the first learning data (305) is input, and the predicted value (313) is the output value from the second individual renal function recovery estimation model (320) into which the first learning data (305) is input. Predicted value (311) and predicted values (313) may represent a probability value regarding whether or not the kidney function recovery corresponding to the first learning data (305) is estimated. Predicted value (321) is the output value output from the first individual renal function recovery estimation model (310) into which the second learning data (306) is input, and the predicted value (323) is the output value from the second individual renal function recovery estimation model (320) into which the second learning data (306) is input. Predicted value (321) and predicted values (323) may represent a probability value regarding whether or not the kidney function recovery corresponding to the second learning data (306) is estimated. Predicted value (311), predicted value (313), predicted value (321) and predicted values Training a kidney function recovery estimation model using (323) is described in more detail in Fig. 4.

[0062]

[0063] FIG. 4 is a diagram illustrating training a renal function recovery estimation model based on federated learning according to one embodiment.

[0064] Referring to Figure 4, the model learning unit (240) predicts the value (311), predicted value (313), predicted value (321) and predicted values (323) to obtain the final predicted values ​​(e.g. predicted values (410) and predicted values (420)) can be obtained. For example, the model learning unit (240) can obtain a predicted value based on soft voting. (311) and predicted values Predicted value by finding the average of (321) (410) is obtained, and the predicted value (313) and predicted values Predicted value by finding the average of (323) (420) can be obtained. Or, the model learning unit (240) can obtain a predicted value based on stacking. (311) and predicted values Predicted values ​​using (321) (410) is obtained, and the predicted value (313) and predicted values Predicted values ​​using (323) (420) can be obtained.

[0065] Predicted value (311), predicted value (313) can be output from the first individual renal function recovery estimation model (310), and the predicted value (311) is the weight This may be the applied value.

[0066] Predicted value (321) and predicted values (323) can be output from the second individual kidney function recovery estimation model (310), and the weight may be a value applied. Weight and weights can be adjusted based on the amount of training data used for learning. For example, the predicted value The amount of training data used in (311) is the predicted value If the amount of training data used in (323) is three times greater, is 0.75 and can be 0.25.

[0067] The model learning unit (240) can learn a renal function recovery estimation model based on a federated learning method. Unlike general machine learning, federated learning does not directly train a machine learning model (e.g., a renal function recovery estimation model) of a central server using learning data, but can train a machine learning model of a central server without sharing data between local devices based on a learned lower machine learning model (e.g., an individual renal function recovery estimation model) of a local device. According to one embodiment, the model learning unit (240) can train a renal function recovery estimation model using a learned first individual renal function recovery estimation model (310) and a learned second individual renal function recovery estimation model (310). The model learning unit (240) obtains an output value from a renal function recovery estimation model into which all learning data is input, and compares the output value with a predicted value. (410) and predicted values A renal function recovery estimation model can be trained to estimate whether renal function has recovered based on comparing the average value of (420). For example, if the predicted value is less than 0.7, renal function is defined as recovered, and if the predicted value is less than 0.7, (410) is 0.8 and predicted value (420) can be 0.85. The model learning unit (240) is based on soft voting. (410) and (420) The average value can be determined as 0.825, and the average value 0.825 can be determined as the final prediction value. The model learning unit (240) estimates that renal function has not recovered since the final prediction value 0.825 is greater than or equal to 0.7, and can train a renal function recovery estimation model by comparing it with label data for renal function recovery estimation. Voting, soft voting, or weight-based soft voting can be used to train the renal function recovery estimation model, and is not affected by the number of machine learning models on the local device.

[0068] The training of the individual renal function recovery estimation model can be accomplished through general machine learning. If the individual renal function recovery estimation model (e.g., the first individual renal function recovery estimation model (310)) is a neural network model, the training process of the individual renal function recovery estimation model is a process of enabling the individual renal function recovery estimation model to recognize input data patterns and make predictions on its own using given training data (e.g., the first training data (305) of FIG. 3). This machine learning process may include the following steps: (1) training data preparation, (2) model initialization, (3) forward computation, (4) loss calculation, (5) backpropagation, and (6) parameter update. The training data preparation process collects and preprocesses the training data from which the individual renal function recovery estimation model will learn. The preprocessing process includes refining the training data and, if necessary, performing operations such as standardization, normalization, and feature selection to create a format suitable for the individual renal function recovery estimation model. The model initialization process sets the initial parameters of the individual renal function recovery estimation model. For example, if the individual renal function recovery estimation model is a neural network, this may include initializing the weights and biases. The forward propagation process inputs the prepared training data into the individual kidney function recovery estimation model to calculate the predicted value for the kidney function recovery estimation. In this process, the individual kidney function recovery estimation model processes the training data using the current parameters and generates the predicted value as the output value. The loss calculation process calculates the difference between the predicted value of the individual kidney function recovery estimation model and the actual correct answer (label) using a loss function. The loss function is a function that allows for calculating a value indicating how accurate (or inaccurate) the prediction of the individual kidney function recovery estimation model is. The backpropagation process adjusts the parameters of the individual kidney function recovery estimation model to reduce the loss derived through the loss function.The backpropagation algorithm differentiates the loss function (gradient) to calculate how much each parameter of the individual renal function recovery estimation model contributed to the loss. Based on this value, the parameters of the individual renal function recovery estimation model can be updated. The parameter update process uses the calculated gradient to update the parameters of the individual renal function recovery estimation model. Gradient descent or its variants can typically be used for parameter updating. Through this process, the individual renal function recovery estimation model can learn to make more accurate predictions. The above processes (e.g., forward calculation, loss calculation, backpropagation, parameter update) can be repeated multiple times on a large number of training data, and training can proceed multiple times until the individual renal function recovery estimation model is sufficiently trained. Through this learning process, the individual renal function recovery estimation model can learn patterns from the given data and gain the ability to make predictions on new data.

[0069]

[0070] FIG. 5 is a flowchart illustrating operations of a method for estimating kidney function recovery according to one embodiment.

[0071] The operations of the recovery estimation method can be performed by a kidney function recovery estimation device (e.g., the kidney function recovery estimation device (100) of FIG. 1).

[0072] In operation (510), the renal function recovery estimation device may receive vital sign data. The vital sign data may include user information of the user who is the subject of renal function recovery estimation and biometric information about the user.

[0073] In operation (520), the renal function recovery estimation device may replace missing values ​​in the vital sign data with other values. For example, if the ratio of missing data in the vital sign data is below a threshold value, the renal function recovery estimation device may replace the missing data with previous data corresponding to the missing data or replace the missing data based on multiple imputation by chained equations (MICE). Alternatively, if the ratio of missing data in the vital sign data exceeds a threshold value, the renal function recovery estimation device may replace the missing data with a missing value indicator.

[0074] According to one embodiment, a renal function recovery estimation device may replace missing data with a missing indicator when the ratio of missing data in vital sign data exceeds a threshold value.

[0075] In operation (530), the renal function recovery estimation device can extract feature data for evaluating renal function. The renal function recovery estimation device can extract feature data for evaluating renal function from vital sign data in which missing values ​​are replaced with other values. For example, the renal function recovery estimation device can extract feature data based on the least absolute shrinkage and selection operator (LASSO), Shapley additive explanations (SHAP)-value, and P-value.

[0076] In operation (540), the kidney function recovery estimation device can estimate whether kidney function has been recovered using a kidney function recovery estimation model. Whether the user's kidney function has been recovered can be estimated based on an output value output from a kidney function recovery estimation model based on associative learning into which feature data has been input. For example, the kidney function recovery estimation device can estimate that the user's kidney function has been recovered if the output value is less than a threshold output value, and can estimate that the user's kidney function has not been recovered if the output value is greater than or equal to the threshold output value. The individual kidney function recovery estimation model may be a model based on at least one of a logistic regression (LR) model, a random forest (RF) model, an extreme gradient boosting (XGBoost) model, a light gradient boosting machine (LightGBM) model, and a CatBoost model.

[0077]

[0078] FIG. 6 is a flowchart illustrating operations of a learning method for learning a kidney function recovery estimation model according to one embodiment.

[0079] The operations of a learning method for learning a kidney function recovery estimation model can be performed by a learning device (e.g., the learning device (200) of FIG. 2).

[0080] In operation (610), the learning device may receive learning bio-sign data. The learning bio-sign data may include information and bio-information of the recovery estimation target.

[0081] In operation (620), the learning device may replace missing values ​​in the learning biosignature data with other values. For example, if the ratio of missing data in the learning biosignature data is below a threshold value, the learning device may replace the missing data with previous data corresponding to the missing data or replace the missing data based on multiple imputation by chained equations (MICE). In one embodiment, if the ratio of missing data in the biosignature data exceeds a threshold value, the learning device may replace the missing data with a missing indicator.

[0082] In operation (630), the learning device can extract feature data for evaluating renal function from the learning bio-sign data. For example, the feature data can be extracted from the learning bio-sign data based on the least absolute shrinkage and selection operator (LASSO), Shapley additive explanations (SHAP)-value, and P-value.

[0083] In operation (640), the learning device may train a kidney function recovery estimation model using an output value output from a recovery estimation model and a recovery estimation result value corresponding to the learning vital sign data. For example, the learning device may determine a recovery estimation result value corresponding to the vital sign data based on soft voting or stacking for individual kidney function recovery estimation models, and train a kidney function recovery estimation model using the output value output from the recovery estimation model and the recovery estimation result value. The individual kidney function recovery estimation model may be at least one of a logistic regression (LR) model, a random forest (RF) model, an extreme gradient boosting (XGBoost) model, a light gradient boosting machine (LightGBM) model, and a CatBoost model.

[0084]

[0085] The embodiments described above may be implemented using hardware components, software components, and / or a combination of hardware components and software components. For example, the devices, methods, and components described in the embodiments may be implemented using a general-purpose computer or a special-purpose computer, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing instructions and responding to them. The processing device may execute an operating system (OS) and software applications running on the operating system. Furthermore, the processing device may access, store, manipulate, process, and generate data in response to the execution of the software. For ease of understanding, the processing device is sometimes described as being used alone; however, one of ordinary skill in the art will recognize that the processing device may include multiple processing elements and / or multiple types of processing elements. For example, a processing unit may include multiple processors, or a processor and a controller. Other processing configurations, such as parallel processors, are also possible.

[0086] Software may include computer programs, codes, instructions, or a combination of one or more of these, which may configure a processing device to perform a desired operation or may independently or collectively command the processing device. The software and / or data may be permanently or temporarily embodied in any type of machine, component, physical device, virtual equipment, computer storage medium, or device for interpretation by the processing device or for providing instructions or data to the processing device. The software may also be distributed over networked computer systems and stored or executed in a distributed manner. The software and data may be stored on a computer-readable recording medium.

[0087] The method according to the embodiment may be implemented in the form of program commands that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may store program commands, data files, data structures, etc., alone or in combination, and the program commands recorded on the medium may be those specially designed and configured for the embodiment or may be those known to and available to those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specially configured to store and execute program commands such as ROMs, RAMs, and flash memories. Examples of program commands include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter, etc.

[0088] The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

[0089] Although the embodiments described above have been described with limited drawings, those skilled in the art will appreciate that various technical modifications and variations can be applied based on the described embodiments. For example, appropriate results can still be achieved even if the described techniques are performed in a different order than described, and / or components of the described systems, structures, devices, circuits, etc. are combined or combined in a different manner than described, or are replaced or substituted with other components or equivalents.

[0090] Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims described below.

Claims

1. In a device for estimating renal function recovery using a renal function recovery estimation model, A data collection unit that receives user information of a user who is a subject of estimation of kidney function recovery and vital sign data including vital information about the user; A data preprocessing unit that replaces missing values ​​of the above vital signs data with other values; A feature extraction unit for extracting feature data for evaluating renal function from bio-sign data in which the missing value is replaced with the other value; and A kidney function recovery estimation unit that estimates whether the user's kidney function has recovered based on an output value output from a kidney function recovery estimation model based on federated learning into which the above feature data has been input. Including, A device for estimating renal function recovery.

2. In paragraph 1, The above feature data is, At least the data included in the above vital signs data includes the user information and the user's serum creatinine concentration. A device for estimating renal function recovery.

3. In paragraph 1, The above data preprocessing unit, If the ratio of the missing data in the above vital signs data is less than or equal to a threshold value, the missing data is replaced with previous data corresponding to the missing data or the missing data is replaced based on MICE (multiple imputation by chained equations; MICE). A device for estimating renal function recovery.

4. In paragraph 1, The above data preprocessing unit, If the ratio of the missing data in the above vital signs data exceeds the threshold, the missing data is replaced with a missing indicator. A device for estimating renal function recovery.

5. In paragraph 1, The above feature extraction unit, Extracting the feature data based on LASSO (least absolute shrinkage and selection operator; LASSO), SHAP (shapley additive explanations)-value and P-value. A device for estimating renal function recovery.

6. In paragraph 1, The above kidney function recovery estimation unit is, If the above output value is less than the threshold output value, it is assumed that the user's renal function has recovered. If the above output value is greater than or equal to the threshold output value, it is estimated that the user's renal function has not been recovered. A device for estimating renal function recovery.

7. In a learning device that learns a kidney function recovery estimation model, A learning data collection unit that receives learning bio-sign data including user information of a user who is a target of estimation of kidney function recovery and bio-information about the user; A data preprocessing unit that replaces missing values ​​of the above learning bio-sign data with other values; A feature extraction unit for extracting feature data for evaluating renal function from learning bio-sign data in which the above missing values ​​are replaced; A model learning unit that learns the renal function recovery estimation model by using the output value output from the renal function recovery estimation model into which the above feature data is input and the recovery estimation result value corresponding to the learning vital signs data. Learning device.

8. In paragraph 7, The above feature data is, At least the data included in the learning bio-sign data includes the user information and the user's serum creatinine concentration. Learning device.

9. In paragraph 7, The above data preprocessing unit, If the ratio of the missing data in the above learning bio-sign data is less than or equal to a threshold value, the missing data is replaced with previous data corresponding to the missing data or the missing data is replaced based on MICE (multiple imputation by chained equations; MICE). Learning device.

10. In paragraph 7, The data preprocessing unit is If the ratio of the missing data in the above vital signs data exceeds the threshold, the missing data is replaced with a missing indicator. Learning device.

11. In paragraph 7, The above feature extraction unit, Extracting the feature data based on LASSO (least absolute shrinkage and selection operator; LASSO), SHAP (shapley additive explanations)-value and P-value. Learning device.

12. In paragraph 7, The above kidney function recovery estimation model is, A federated learning model based on soft voting or stacking for individual kidney function recovery estimation models. Learning device.

13. In paragraph 12, The above individual renal function recovery estimation model is, A model based on at least one of a logistic regression model (LR), a random forest model (RF), an extreme gradient boosting (XGBoost) model, a light gradient boosting machine (LightGBM) model, and a CatBoost model. Learning device.

14. A method for estimating renal function recovery using a renal function recovery estimation model, An operation of receiving user information of a user who is a subject of estimation of kidney function recovery and vital sign data including vital information about the user; An action of replacing missing values ​​of the above vital signs data with other values; An operation of extracting feature data for evaluating renal function from biosignal data in which the missing value is replaced with the other value; and An operation for estimating whether the user's kidney function has recovered based on an output value output from a kidney function recovery estimation model based on federated learning into which the above feature data has been input. Including, Methods for estimating renal function recovery 15. In paragraph 14, The above replacing action is, An operation including replacing the missing data with previous data corresponding to the missing data or replacing the missing data based on MICE (multiple imputation by chained equations; MICE) when the ratio of the missing data in the above vital signs data is less than or equal to a threshold value. Methods for estimating renal function recovery 16. In paragraph 14, The above replacing action is, An operation for replacing the missing data with a missing indicator when the ratio of the missing data in the above vital signs data exceeds a threshold value, Methods for estimating renal function recovery 17. In paragraph 14, The operation of extracting the above feature data is as follows: Including an operation of extracting the feature data based on LASSO (least absolute shrinkage and selection operator; LASSO), SHAP (shapley additive explanations)-value and P-value. Methods for estimating renal function recovery 18. In paragraph 14, The above estimated action is, An operation for estimating that the user's renal function has been restored if the output value is less than the threshold output value; and Including an operation for estimating that the user's renal function has not been recovered if the output value is greater than or equal to a threshold output value. Methods for estimating renal function recovery 19. A learning method for learning a model for estimating renal function recovery, An operation of receiving learning bio-sign data including user information of a user who is a subject of estimation of kidney function recovery and bio-information about the user; An action of replacing missing values ​​in the above learning bio-sign data with other values; An operation of extracting feature data for evaluating renal function from learning bio-sign data in which the above missing values ​​are replaced; and An operation of training the renal function recovery estimation model by using the output value output from the renal function recovery estimation model into which the above feature data is input and the recovery estimation result value corresponding to the learning vital signs data. Including, How to learn.

20. A computer program stored on a computer-readable recording medium to execute the method of claim 14 in combination with hardware.

Images