Extubation timing prediction support device and method

Abstract

The present invention provides an artificial intelligence-based method for predicting the probability of successful extubation of a low birth weight infant or for assisting in predicting the timing of extubation of a low birth weight infant, the method comprising the steps of acquiring information about the low birth weight infant (including at least one of patient information, ventilator information, and vital sign information), and calculating the probability of successful extubation of the low birth weight infant using a neural network model, and further comprising the steps of recommending extubation if the probability of successful extubation is equal to or greater than a threshold, and recommending continued use of the ventilator if the probability of successful extubation is less than the threshold.

Description

[Technical Field]

[0001] The present application relates to an apparatus and method for supporting prediction of extubation timing.

[0002] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority based on Korean Patent Application No. 10-2022-0168057, filed on December 5, 2022, the entire specification of which is incorporated herein by reference.

[0003] [Explanation regarding government-supported research and development] This research was supported by the Ministry of Health and Welfare of the Republic of Korea [Project name: Data labeling and practical clinical research for the development of emergency and real-time diagnostic artificial intelligence systems, Project specific number: 1465036287, Detailed project number: HI18C0022020022]. [Background technology]

[0004] Low birth weight infants require multiple mechanical ventilation (MV) sessions to support the development of their immature lungs. However, excessive intubation can lead to problems such as oxygen toxicity, bronchopulmonary dysplasia, and neurodevelopmental disorders, so extubation should be performed as quickly as possible. Because extubation requires the integration of multiple clinical information and consideration of complex causal relationships, there is currently no internationally agreed-upon extubation protocol, and the success rate and specificity of extubation vary depending on the medical staff and the amount of medical information available.

[0005] Previously, spontaneous breathing trials (SBTs) have been used to determine extubation readiness in extremely premature infants. However, the accuracy of SBTs is low, with as many as one-third of extubations failing. Furthermore, extremely premature infants commonly experience adverse reactions during SBTs. Few studies have investigated the dynamics of physiological signals as predictors of readiness for ventilation or extubation in low-birth-weight infants, and small studies have shown that using physiological signals with SBTs, such as changes in heart rate or respiratory rate, only marginally improved the ability to predict extubation success. Summary of the Invention [Problem to be solved by the invention]

[0006] The present application aims to solve the aforementioned problems by providing an accurate and field-applicable prediction model that identifies physiological signals apart from SBT to determine the success or failure of extubation prediction, and calculates extubation readiness probability based on such physiological signals.

[0007] Specifically, this application is broadly composed of a prediction model (neural network model) and outcome analysis to determine whether to extubate a low birth weight infant. The prediction model uses demographic information, such as the gestational age and birth weight of the intubated patient, periodically measured and recorded vital signs, such as oxygen saturation, the hourly variance and fluctuation of heart rate and respiratory rate, ventilator information records, such as FiO2, MAP, and PEEP, and bioindicators, such as the SF ratio (SpO2 / FiO2) and the ROX (ratio of oxygen saturation) index, which reflect the effectiveness of the patient's ventilator. The classifier for the prediction model can apply a complementary naive Bayesian model and a logistic regression model, which can achieve consistent performance even with highly imbalanced data. The results are analyzed using the theoretically verified SHAP (SHapley Additive exPlanations), which provides numerical and visual evidence of the basis and probability behind the success rate derived by the predictive model. [Means for solving the problem]

[0008] In one embodiment of the present application, a method for assisting in predicting the probability of successful extubation of a low birth weight infant or predicting the timing of extubation of a low birth weight infant, the method being executed by a processor and based on artificial intelligence, includes the steps of obtaining information about the low birth weight infant (including at least one of patient information, ventilator information, and vital sign information), and calculating the probability of successful extubation of the low birth weight infant using a neural network model.

[0009] In one embodiment, the method further includes a step of recommending extubation if the probability of successful extubation is equal to or greater than a threshold, and a step of recommending continued ventilator support if the probability of successful extubation is less than the threshold, wherein the threshold may represent a minimum point at which extubation of the low birth weight infant is recommended.

[0010] In one embodiment, the method may further include analyzing the results of the calculation of the probability of successful extubation and providing a report.

[0011] In one embodiment, the step of analyzing the results and providing a report may include the steps of: numerically providing information on the prediction parameters used by the neural network model to calculate the probability of successful extubation; and visualizing and providing a graph showing the contribution of the prediction parameters.

[0012] In one embodiment, the step of visualizing and providing the contribution of the prediction parameter as a graph may include the step of providing a change in the value of the prediction parameter from a specific time point in the past to the current time in the form of a heatmap.

[0013] In one embodiment, analyzing the results and providing a report may include providing a graphical visualization of the probability of successful extubation from a particular time in the past to the present time.

[0014] In one embodiment, the predictive parameters may be at least one of ventilator setting information including heart rate, respiratory rate, body temperature, oxygen saturation (SpO2), blood pressure, GA, birth weight, PMA at extubation, male gender, blood gases before extubation (pH, pCO2), and fraction of inspired oxygen (FiO2), and positive end-expiratory pressure (PEEP), mean arterial pressure (MAP), frequency, SpO2 / FiO2 (SF) ratio, ratio of SpO2 / FiO2 to respiratory rate (ROX), and respiratory severity score (RSS) obtained by multiplying MAP and FiO2.

[0015] In one embodiment, the prediction parameter that most contributes to the calculation of the probability of successful extubation by the neural network model may be the SF ratio.

[0016] In one embodiment, the step of analyzing the results of the calculation of the extubation success probability may apply SHapley Additive exPlanation (SHAP) to calculate, analyze, and analyse the importance of features extracted by the neural network to evaluate the analyzability of potential predictive parameters.

[0017] In one embodiment, the neural network model may be either a logistic regression model or a complement naive Bayesian model.

[0018] In one embodiment, the extubation prediction support method may be realized by a computer program stored in a computer-readable recording medium.

[0019] In another embodiment of the present application, in an artificial intelligence-based device for assisting in predicting the probability of successful extubation of a low birth weight infant or predicting the timing of extubation of a low birth weight infant, the device may include: an acquisition unit that acquires information about the low birth weight infant, including at least one of patient information, ventilator information, and vital sign information; and a neural network model that calculates the probability of successful extubation of the low birth weight infant using the information about the low birth weight infant as an input.

[0020] In one embodiment, the device may further include an analysis unit that analyzes the results of the calculation of the extubation success probability and provides a report.

[0021] In one embodiment, the analysis unit may provide numerical information on the prediction parameters used by the neural network model when calculating the probability of successful extubation, or may provide a visualization of the contribution of the prediction parameters as a graph. [Effects of the Invention]

[0022] This extubation timing prediction support device and method predicts the difficult-to-predict MV extubation success rate based on the admission information and vital signs of low birth weight newborns, analyzes and provides the contribution of factors to the prediction, supports medical staff in making extubation decisions, and suggests the timing of extubation for patients. Therefore, the present invention provides medical staff with extubation success rates and prediction indicators to support evidence-based extubation decisions. This can reduce the mortality and morbidity rates of low birth weight newborns due to inappropriate extubation.

[0023] The effects of the present invention are not limited to the above-mentioned effects, and other effects not mentioned will be clearly understood by those skilled in the art from the description of the accompanying claims. [Brief explanation of the drawings]

[0024] In order to more clearly describe the technical solutions of the embodiments of the present application or the prior art, the drawings necessary for describing the embodiments are briefly introduced below. It should be understood that the following drawings are only for describing the embodiments of the present specification, and are not intended to be limiting. Furthermore, for clarity of description, the following drawings may show some elements with various modifications, such as exaggeration, omission, etc.

[0025] [Figure 1] 1 is a flowchart of an extubation timing prediction support device and method according to an embodiment of the present application. [Figure 2] 1 is a detailed flowchart of the steps of analyzing prediction results and providing a report according to one embodiment of the present application. [Figure 3] 1 is a table showing details of acquired information on low birth weight babies according to one embodiment of the present application; [Figure 4] 1 is a table of multivariate analysis of potential variables for predicting successful extubation according to one embodiment of the present application. [Figure 5A]FIG. 1 is a PDP plot analyzing the correlation between SF ratio and calculated probability of successful extubation, showing the average SF ratio, according to one embodiment of the present application. [Figure 5B] FIG. 10 is a PDP plot analyzing the correlation between the SF ratio and the calculated probability of successful extubation, showing the mean sequential difference of the SF ratio, according to an embodiment of the present application. [Figure 6A] FIG. 10 is a diagram illustrating a step of providing numerical prediction parameter information and visualizing the contribution of the prediction parameters as a graph in a step of analyzing the prediction results according to an embodiment of the present application. [Figure 6B] FIG. 10 is a diagram illustrating a step of providing numerical prediction parameter information and visualizing the contribution of the prediction parameters as a graph in a step of analyzing the prediction results according to an embodiment of the present application. [Figure 6C] FIG. 10 is a diagram illustrating a step of providing numerical prediction parameter information and visualizing the contribution of the prediction parameters as a graph in a step of analyzing the prediction results according to an embodiment of the present application. [Figure 6D] FIG. 10 is a diagram illustrating a step of providing numerical prediction parameter information and visualizing the contribution of the prediction parameters as a graph in a step of analyzing the prediction results according to an embodiment of the present application. [Figure 6E] FIG. 10 is a diagram illustrating a step of providing numerical prediction parameter information and visualizing the contribution of the prediction parameters as a graph in a step of analyzing the prediction results according to an embodiment of the present application. [Figure 6F] FIG. 10 is a diagram illustrating a step of providing numerical prediction parameter information and visualizing the contribution of the prediction parameters as a graph in a step of analyzing the prediction results according to an embodiment of the present application. [Figure 6G] FIG. 10 is a diagram illustrating a step of providing numerical prediction parameter information and visualizing the contribution of the prediction parameters as a graph in a step of analyzing the prediction results according to an embodiment of the present application. [Figure 6H]FIG. 10 is a diagram illustrating a step of providing numerical prediction parameter information and visualizing the contribution of the prediction parameters as a graph in a step of analyzing the prediction results according to an embodiment of the present application. [Figure 7] FIG. 1 is a block diagram of components that perform major functions when implementing the invention as software according to one embodiment of the present application. DETAILED DESCRIPTION OF THE INVENTION

[0026] The terminology used herein is merely for the purpose of referring to particular embodiments and is not intended to limit the present application. As used herein, the singular forms "a," "an," and "the" include the plural forms unless the context clearly dictates otherwise. As used herein, the meaning of "comprising" embodies certain properties, regions, integers, steps, operations, items, and / or components, and does not exclude the presence or addition of other properties, regions, integers, steps, operations, items, and / or components.

[0027] Unless otherwise defined, all terms, including technical and scientific terms, used herein have the same meaning as commonly understood by a person of ordinary skill in the art to which this application pertains. Commonly used predefined terms are further interpreted as having a meaning consistent with the relevant technical literature and the presently disclosed content, and should not be interpreted as an ideal or very formal meaning unless otherwise defined.

[0028] FIG. 1 is a flowchart of a method for supporting prediction of extubation timing according to an embodiment of the present application.

[0029] Referring to FIG. 1, the extubation timing prediction support method may include the steps of: acquiring information about a low birth weight infant (S10); inputting the information about the low birth weight infant into a neural network model (S12); calculating a probability of successful extubation of the low birth weight infant (S14); and recommending extubation if the calculated probability of successful extubation is equal to or greater than a threshold (S161); or recommending continued use of a ventilator if the calculated probability is less than the threshold (S162). The information about the low birth weight infant may include at least one of patient information, ventilator information, and vital signs information. The threshold refers to the lowest point at which extubation of the low birth weight infant is recommended. In one example, the threshold may be set to 0.98.

[0030] The neural network model may be any of various artificial intelligence models, including logistic regression, complement naive Bayesian model, random forest, gradient boosting, decision tree, stochastic gradient descent (SGD) classifier, and extreme gradient boosting (XGB), etc. In particular, the neural network model of the present application may be a logistic regression model or a complement naive Bayesian model, which can achieve consistent performance even in highly imbalanced data.

[0031] FIG. 2 is a detailed flowchart of the steps of analyzing the prediction results and providing a report according to one embodiment of the present application.

[0032] 2, the extubation timing prediction support method may further include a step (S18) of analyzing the calculation result of the extubation success probability and providing a report. The step (S18) of analyzing the calculation result of the extubation success probability and providing a report may include a step (S181) of numerically providing information on prediction parameters used when the neural network model calculates the extubation success probability, and a step (S182) of visualizing and providing the contribution of the prediction parameters as a graph. The prediction parameters may be at least one of heart rate, respiratory rate, body temperature, oxygen saturation (SpO2), blood pressure, gestational age (GA), birth weight (post-menstrual age), PMA (post-menstrual age) at extubation, male gender, blood gases (pH, pCO2) before extubation, ventilator setting information including the oxygen concentration (FiO2) supplied from the ventilator, positive end-expiratory pressure (PEEP), mean arterial pressure (MAP), ventilator set respiratory frequency, SpO2 / FiO2 (SF) ratio, SpO2 / FiO2 to respiratory rate ratio (ROX), and respiratory severity score (RSS) obtained by multiplying MAP and FiO2. In particular, the prediction parameter that most contributes to calculation of the extubation success probability using the neural network model may be the SF ratio. The step of analyzing the calculation result of the extubation success probability may apply SHapley Additive exPlanation (SHAP) to calculate, analyze, and analyse the importance of features extracted by the neural network, thereby evaluating the analyzability of potential predictive parameters.

[0033] The step of visualizing and providing the contribution of the prediction parameter as a graph may provide a change in the value of the prediction parameter from a specific time point in the past to the current time in the form of a heat map. The step of analyzing the results and providing a report may provide a visualization of the probability of successful extubation from a specific time point in the past to the current time in the form of a graph. In one example, the specific time point in the past may be 24 hours prior to the current time.

[0034] FIG. 3 is a table showing details of the acquired information of low birth weight babies according to one embodiment of the present application.

[0035] 3, in the method for supporting prediction of extubation timing, in the step (S10) of acquiring information on the low birth weight infant, the information on the low birth weight infant may include at least one of patient information, ventilator information, and vital sign information. In one embodiment, the patient information (Demographic) may include gestational age, sex (Sex), birth weight, and Apgar score. As a basis for data collection and coding, sex may be set to 0 for males and 1 for females, and the Apgar score, which is a health index of the newborn, may be set to an integer value.

[0036] The ventilator information (Ventilation) may include MAP (Mean Airway Pressure), which is the average pressure of the ventilator, PEEP (Positive end expiratory pressure), which is the pressure in the airway after expiration, FiO2 (Fration of Inspired Oxygen), which is the oxygen content of the air supplied from the ventilator, and Freq (Frequency), which is the set breathing cycle of the ventilator.

[0037] Vital signs may include respiratory rate (RR), systolic blood pressure (Systolic BP), diastolic blood pressure (Diastolic BP), heart rate, and oxygen saturation.

[0038] FIG. 4 is a table of multivariate analysis of potential variables for predicting successful extubation according to one embodiment of the present application.

[0039] Referring to Figure 4, the SF ratio analyzed by SHAP and PDP was a significant predictor of extubation success in all cohorts. Among the potential variables, only the SF ratio and SpO2 were robust predictors of extubation success. Specifically, FiO2, PEEP, and RSS, which have been primarily included in other studies, were associated with extubation success in the internal validation cohort but were not significant in the external validation cohort. MAP, SpO2, and the SF ratio significantly discriminated extubation success in the external validation cohort, but in multivariate analysis, only the SF ratio and SpO2 were significant predictors of extubation success.

[0040] 5A and 5B are PDP plots analyzing the correlation between the SF ratio and the calculated probability of successful extubation according to one embodiment of the present application, where FIG. 5A shows the average SF ratio and FIG. 5B shows the mean sequential difference value of the SF ratio.

[0041] Referring to Figures 5A and 5B, the probability of successful extubation is negatively correlated with the increasing mean value of the sequential difference of the SF ratio. The mean value of the SF ratio and the mean value of the sequential difference of the SF ratio were confirmed to be important predictors of successful extubation. When the SF ratio was analyzed using SHAP and PDP, high classification power was consistently observed in all cohorts. The SHAP value of the mean value of the sequential difference of the SF ratio was 0.32, the highest ratio. In contrast to the prior art, which used birth weight and gestational age as the main determinants of extubation success, the present application predicts the probability of successful extubation using only the SF ratio.

[0042] 6A to 6H are exemplary diagrams illustrating steps of providing numerical prediction parameter information and visualizing the contribution of the prediction parameters as graphs in the step of analyzing the prediction results according to an embodiment of the present application.

[0043] The method of providing the predicted parameter information numerically and visualizing the contribution of the predicted parameters as a graph may be displayed on an output device including a display, a speaker, etc. As another example, the method may be provided on a device that has an integrated input / output function such as a touch screen.

[0044] 6A to 6H, the method of providing the predicted parameter information numerically and visualizing the contribution of the predicted parameters as a graph may include displaying the name and identifier of the patient as detailed information on an output device (1), along with the patient's extubation success rate (2) and a threshold value representing the minimum point at which extubation is recommended for the patient (3). Alternatively, the extubation success rate from a specific point in the past (e.g., 24 hours prior to the current point) to the current point may be displayed in the form of a visualized timeline (4), or in the form of a factor contribution graph, which is a graph showing the results of a SHAP analysis at the current prediction point (5). In addition, the graphs may be displayed in the form of a heatmap graph (6) showing the change in SHAP values ​​by factor from a specific time point in the past to the present, a graph (7) showing the biosignal values ​​from a specific time point in the past to the present, a graph (8) showing the change in ventilator setting information from a specific time point in the past to the present, a graph (9) showing outlier values ​​of the biosignals, or a box plot showing statistical summary information of the biosignals for each time unit.

[0045] FIG. 7 is a block diagram of the main functions when the present invention is realized as software according to one embodiment of the present application.

[0046] Referring to Figure 7, this application may be composed of a front-end service, an EMR integration service, an AI service, a relational database (RDB), an in-memory database, etc. The front-end service is composed of an authentication module, a WebAPI module, a configuration management module, and a DB integration module. The EMR integration service includes a source data analysis module, a data monitoring module, and a DB integration module, and the AI ​​service is composed of an AI module, a DB integration module, and a factor analysis module. The detailed structure is as follows:

[0047] The authentication module is used to verify software access privileges and execution, such as login, membership registration, existence of administrator account, current session privileges, and time restrictions.

[0048] The WebAPI module is used to provide front-end functions such as web page access and HTTP protocol communication for users to use the software. The WebAPI module displays information on admitted patients through a patient list, allows users to query admitted patient information and vital signs, query patient medical history information and past records, input and update patient ventilator information, and query extubation success rates based on user-entered information and patient data.

[0049] The setting management module is used to reflect and manage settings related to service execution such as host, port, log settings, database URI, etc. The setting management module sets the URI information of the DB to be used, the log level and output method, the allowed IP band and port, and sets web access.

[0050] The DB linkage module is used by the service to perform functions of querying, correcting, deleting, and updating data.

[0051] The source data analysis module converts the original data containing patient information and biosignals into a processable data format and analyzes it. The source data analysis module parses the source data into processable objects, verifies the validity of the input data, and determines which item the source data format corresponds to (patient information, biosignals, lab data, etc.).

[0052] The data monitoring module checks whether original data has been added or updated. The data monitoring module checks whether data has been added or updated at set intervals, and if there is any updated or added data, it sends it to an in-memory DB so that connected services can query it. It also converts the data into a JSON format that can be linked.

[0053] The AI ​​module is used to load a prediction model, which is a neural network model, input patient information into the prediction model, and calculate the extubation success prediction result.

[0054] The factor analysis module is used to analyze the analytical information provided to medical staff using SHAP etc. based on the results calculated by the AI ​​module. Specifically, the factor analysis module loads the SHAP explainer so that the results of the neural network model can be used on the front end and the results of the registered model can be analyzed, and uses the explainer to generate information on the contribution of factors to the predicted results.

[0055] The software may include a computer program, code, instructions, or at least one combination thereof, and may configure a processing device to perform a desired operation or instruct the processing device, either individually or collectively. The software and / or data may be embodied in a specific type of machine, component, physical device, virtual device, computer storage medium, or device that can provide commands or data to the processing device or be interpreted by the processing device. The software may be distributed across computer systems connected by a network, or may be stored or executed in a distributed manner. The software and data may be stored on at least one computer-readable recording medium.

[0056] The extubation timing prediction support method according to another aspect of the present application may be executed by a computing device including a processor. For example, the steps of (S10) acquiring information about the premature birth weight infant (including at least one of patient information, ventilator information, and vital sign information), (S12) inputting the information about the premature birth weight infant into a neural network model, (S14) calculating the probability of successful extubation of the premature birth weight infant, and (S18) analyzing the calculation result of the probability of successful extubation and providing a report may be executed by an acquisition unit, a neural network model, and an analysis unit, each of which is a computing device including a processor. The computing device may be a desktop computer, a laptop computer, a smartphone, or a similar computing device, or any device that can be integrated. The computer is a device having at least one alternative and special-purpose processor, memory, storage space, and networking components (either wireless or wired). The computer may run an operating system such as, for example, Microsoft's Windows-compatible operating systems, Apple's OS X or iOS, a Linux distribution, or Google's Android OS.

[0057] The operations of the extubation timing prediction support method according to the above-described embodiment may be implemented at least in part by a computer program and stored in a computer-readable storage medium, for example, by a program product consisting of a computer-readable medium containing program code, which may be executed by a processor to perform any or all of the steps, operations, or processes described above.

[0058] The computer-readable storage medium includes any type of recorded identification device in which data that can be read by a computer is stored. Examples of the computer-readable storage medium include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage identification device, etc. The computer-readable storage medium may also be distributed among computer systems connected via a network, so that computer-readable code is stored and executed in a distributed manner. Furthermore, functional programs, codes, and code segments for implementing the present embodiment will be easily understood by those skilled in the art to which the present embodiment pertains.

[0059] This extubation timing prediction support device and method predicts the difficult-to-predict MV extubation success rate based on the admission information and vital signs of low birth weight newborns, analyzes and provides the contribution of factors to the prediction, supports medical staff in making extubation decisions, and suggests the timing of extubation for patients. Therefore, this application supports evidence-based extubation decisions by providing medical staff with extubation success rates and prediction indicators. It also aims to reduce the mortality and morbidity rates of low birth weight newborns through appropriate extubation.

[0060] Although the present application has been described with reference to the embodiments shown in the drawings, these are merely illustrative, and those skilled in the art will understand that various modifications and variations of the embodiments are possible. However, such modifications should be considered to fall within the technical scope of protection of the present application. Therefore, the true technical scope of protection of the present application should be determined by the technical ideas of the appended claims. [Industrial Applicability]

[0061] The extubation timing prediction support device and method according to an embodiment of the present application predicts the difficult-to-predict extubation success rate of MV based on the admission information and vital signs of low birth weight newborns, analyzes and provides the contribution of factors to the prediction, supports medical staff in making extubation decisions, and suggests the timing of extubation for patients. Therefore, the present application supports evidence-based extubation decisions by providing medical staff with extubation success rates and prediction indicators, thereby aiming to reduce the mortality and morbidity rates of low birth weight newborns due to inappropriate extubation.

Claims

1. 1. A method for assisting in predicting the probability of successful extubation of a low birth weight infant or predicting the timing of extubation of a low birth weight infant, the method comprising: obtaining information about the premature birth weight infant, including at least one of patient information, ventilator information, and vital signs information; and calculating the probability of successful extubation of the low birth weight infant using a neural network model.

2. recommending extubation if the probability of successful extubation is equal to or greater than a threshold; If the probability of successful extubation is less than a threshold, recommending continued mechanical ventilation; The extubation prediction support method according to claim 1 , wherein the threshold value represents a minimum point at which extubation of the low birth weight infant is recommended.

3. The extubation prediction support method according to claim 1 , further comprising the step of analyzing a result of the calculation of the extubation success probability and providing a report.

4. analyzing the results and providing a report, Numerical information on the prediction parameters used by the neural network model to calculate the probability of successful extubation is provided; and providing a visualization of the contribution of the prediction parameters as a graph.

5. providing a graphical visualization of the contributions of the prediction parameters, The extubation prediction support method according to claim 4, further comprising a step of providing an amount of change in the value of the prediction parameter from a specific time point in the past to the current time point in the form of a heat map.

6. analyzing the results and providing a report, The extubation prediction support method according to claim 3 , further comprising a step of visualizing and providing the probability of successful extubation from a specific time point in the past to the current time point as a graph.

7. 5. The extubation prediction support method according to claim 4, wherein the prediction parameters are at least one of ventilator setting information including heart rate, respiratory rate, body temperature, oxygen saturation (SpO2), blood pressure, GA, birth weight, PMA at extubation, male gender, blood gases before extubation (pH, pCO2), and fraction of inspired oxygen (FiO2), and positive end-expiratory pressure (PEEP), mean arterial pressure (MAP), frequency, SpO2 / FiO2 (SF) ratio, ratio of SpO2 / FiO2 to respiratory rate (ROX), and respiratory severity score (RSS) obtained by multiplying MAP and FiO2.

8. 8. The extubation prediction support method according to claim 7, wherein the prediction parameter that most contributes to the calculation of the probability of successful extubation by the neural network model is the SF ratio, among the prediction parameters.

9. The step of analyzing the result of the calculation of the extubation success probability, The extubation prediction support method according to claim 3, characterized in that the analyzability of potential prediction parameters is evaluated by applying SHAP to calculate, analyze, and derive the importance of features extracted by the neural network.

10. The neural network model 2. The method for supporting prediction of extubation according to claim 1, wherein the model is either a logistic regression model or a complement naive Bayesian model.

11. A computer program stored in a computer-readable recording medium for realizing the extubation prediction support method according to any one of claims 1 to 10.

12. An apparatus for assisting in predicting the probability of successful extubation of a low birth weight infant or predicting the timing of extubation of a low birth weight infant based on artificial intelligence, an acquisition unit that acquires information about the low birth weight infant, the information including at least one of patient information, ventilator information, and vital sign information; and a neural network model that receives information about the low birth weight infant as input and calculates the probability of successful extubation of the low birth weight infant.

13. The extubation prediction support device according to claim 12 , further comprising an analysis unit that analyzes a result of the calculation of the extubation success probability and provides a report.

14. The analysis unit numerically provides information on the prediction parameters used by the neural network model when calculating the probability of successful extubation, or The extubation prediction support device according to claim 13, wherein the contribution of the prediction parameters is visualized and provided as a graph.

15. 15. The extubation prediction support device according to claim 14, wherein the prediction parameters are at least one of ventilator setting information including heart rate, respiratory rate, body temperature, oxygen saturation (SpO2), blood pressure, GA, birth weight, PMA at extubation, male gender, blood gases before extubation (pH, pCO2), and fraction of inspired oxygen (FiO2), and positive end-expiratory pressure (PEEP), mean arterial pressure (MAP), frequency, SpO2 / FiO2 (SF) ratio, ratio of SpO2 / FiO2 to respiratory rate (ROX), and respiratory severity score (RSS) obtained by multiplying MAP and FiO2.

16. 16. The extubation prediction support device according to claim 15, wherein the prediction parameter that most contributes to the calculation of the probability of successful extubation by the neural network model is the SF ratio, among the prediction parameters.

Images