Model training method, alarm classification method and device based on machine learning model, electronic equipment and storage medium

Abstract

The invention provides a model training method, an alarm classification method and device based on a machine learning model, electronic equipment and a storage medium, and relates to the technical field of medical equipment networking. After a key data set is constructed, a plurality of preset algorithm models can be trained by utilizing the key data set so as to screen out a target classification model from a plurality of obtained candidate classification models, and then whether an alarm generated by medical equipment belongs to a clinical alarm or not can be identified by utilizing the target classification model. In this way, the pressure of medical staff on a large number of non-clinical alarms is relieved, the clinical alarms are rapidly recognized through the target classification model, and the medical staff can conveniently and rapidly take treatment measures.

Description

Technical Field

[0001] The present invention relates to the technical field of medical device networking. Specifically, the present invention relates to a model training method, an alarm classification method based on a machine learning model, a device, an electronic device, and a storage medium. Background Art

[0002] At present, in the intensive care unit (ICU), medical devices such as ventilators and monitors are needed to maintain and monitor the vital signs of patients. Relevant investigation results show that 85%-99% of the alarms of such medical devices are either false alarms or technical alarms that have nothing to do with the actual clinical conditions of patients. In fact, only a small part of the alarms belong to clinical alarms related to the clinical conditions of patients. In this case, medical staff need to distinguish real clinical alarms from a large number of device alarms every day.

[0003] Therefore, how to effectively identify clinical alarms of medical devices is the key to ensuring the life safety of patients and reducing waste of medical resources. Summary of the Invention

[0004] The purpose of the present invention is to provide a model training method, an alarm classification method based on a machine learning model, a device, an electronic device, and a storage medium to improve the problems existing in the prior art.

[0005] The embodiments of the present invention can be implemented as follows: In a first aspect, the present invention provides a model training method, including: Constructing a key data set, where each key sample in the key data set reflects whether a clinical alarm occurs in a medical device within a set observation period; Training a plurality of preset algorithm models by using the key data set to screen out a target classification model from the obtained plurality of candidate classification models, where the target classification model is used to determine whether an alarm generated by the medical device belongs to a clinical alarm.

[0006] In a second aspect, the present invention further provides an alarm classification method based on a machine learning model, including: Receiving the acquisition data of a medical device collected by an acquisition device each time, where the acquisition data includes the status data and status flag values of a plurality of monitoring indicators; wherein, the status flag value of 0 represents that the acquisition data is collected by the acquisition device at regular intervals, and the status flag value of 1 represents that the acquisition data is collected when the acquisition device senses that the medical device triggers an alarm; If the status flag value of any one of the monitoring indicators in the received acquisition data is 1, then using the moment after waiting for a preset duration as the judgment moment; Generating a feature matrix to be recognized based on all the acquisition data within a set observation period before the judgment moment; Input the feature matrix to be recognized into the target classification model to obtain a classification result; wherein, the classification result reflects whether the alarm triggered by the medical device is a clinical alarm or a non-clinical alarm, and the target classification model is obtained by training the model using the method provided in the first aspect above.

[0007] In a third aspect, the present invention further provides a model training device, including: A dataset construction module for constructing a key dataset, where each key sample in the key dataset reflects whether a clinical alarm occurs in a medical device within a set observation duration; A model screening module for training a plurality of preset algorithm models using the key dataset to screen out a target classification model from the obtained plurality of candidate classification models, where the target classification model is used to determine whether the alarm generated by the medical device belongs to a clinical alarm.

[0008] In a fourth aspect, the present invention further provides an alarm classification device based on a machine learning model, including: A data receiving module for receiving the acquisition data of a medical device collected by an acquisition device each time, where the acquisition data includes the status data and status flag values of a plurality of monitoring indicators; wherein, the status flag value of 0 represents that the acquisition data is collected by the acquisition device at regular intervals, and the status flag value of 1 represents that the acquisition data is collected when the acquisition device senses that the medical device triggers an alarm; A vectorization module for, when the status flag value of any one of the monitoring indicators in the received acquisition data is 1, using the moment after waiting for a preset duration as the judgment moment; The vectorization module is further configured to generate a feature matrix to be recognized based on all the acquisition data within a set observation duration before the judgment moment; An alarm classification module for inputting the feature matrix to be recognized into the target classification model to obtain a classification result; wherein, the classification result reflects whether the alarm triggered by the medical device is a clinical alarm or a non-clinical alarm, and the target classification model is obtained by training the model using the method provided in the first aspect above.

[0009] In a fifth aspect, the present invention further provides an electronic device, including: a memory and a processor, where the memory stores a software program, and when the electronic device runs, the processor executes the software program to implement the method as described in the first aspect, or the method as described in the second aspect above.

[0010] In a sixth aspect, the present invention further provides a computer-readable storage medium, where the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the method as described in the first aspect, or the method as described in the second aspect above.

[0011] Compared with the prior art, the embodiments of the present invention provide a model training method, an alarm classification method based on a machine learning model, a device, an electronic device and a storage medium. After constructing a key data set, the key data set can be used to train a plurality of preset algorithm models, so as to screen out a target classification model from the obtained plurality of candidate classification models, and then the target classification model can be used to identify whether the alarm generated by the medical device belongs to a clinical alarm, so as to relieve the pressure of medical staff facing a large number of non-clinical alarms, and quickly identify clinical alarms by using the target classification model, which is convenient for medical staff to quickly take rescue measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0013] Figure 1 It is a schematic flowchart of a model training method provided by an embodiment of the present invention.

[0014] Figure 2 It is a schematic diagram of the acquisition time distribution of multiple historical acquisition data in an observation set provided by an embodiment of the present invention.

[0015] Figure 3 It is a schematic diagram of the sorting of the importance coefficients of multiple monitoring indicators provided by an embodiment of the present invention.

[0016] Figure 4 It is the performance evaluation result of multiple algorithm models provided by an embodiment of the present invention.

[0017] Figure 5 It is a schematic flowchart of an alarm classification method based on a machine learning model provided by an embodiment of the present invention.

[0018] Figure 6 It is a functional module diagram of a model training device provided by an embodiment of the present invention.

[0019] Figure 7 It is a functional module diagram of an alarm classification device based on a machine learning model provided by an embodiment of the present invention.

[0020] Figure 8 It is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. Components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a variety of different configurations.

[0022] Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0023] It should be noted that like reference numerals and letters denote like items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0024] In addition, terms such as "first" and "second" are used for descriptive purposes only and cannot be construed as indicating or implying relative importance.

[0025] It should be noted that, without conflict, the features in the embodiments of the present invention may be combined with each other.

[0026] Please refer to Figure 1 , Figure 1 which is a schematic flowchart of a model training method provided by an embodiment of the present invention. This method can be applied to electronic devices such as smartphones, personal notebooks, personal computers, and servers. As Figure 1 shown, this method includes the following steps S101 to S102.

[0027] S101. Construct a key data set.

[0028] In this embodiment, the key data set may include a number of key samples, and each key sample can reflect whether a clinical alarm occurs in a medical device within a set observation duration.

[0029] Among them, the medical device may be, but is not limited to, devices such as ventilators and monitors for monitoring patients' vital signs. The occurrence of a clinical alarm in the medical device indicates that the patient is in a relatively urgent situation and medical staff need to take rescue measures.

[0030] S102. Use the key data set to train a plurality of preset algorithm models to screen out a target classification model from the obtained plurality of candidate classification models.

[0031] In this embodiment, the target classification model is used to determine whether an alarm generated by a medical device belongs to a clinical alarm.

[0032] The model training method provided by the present invention, after constructing the key data set, can use the key data set to train a plurality of preset algorithm models, so as to screen out the target classification model from the obtained plurality of candidate classification models, and then can use the target classification model to identify whether an alarm generated by a medical device belongs to a clinical alarm, so as to relieve the pressure of medical staff facing a large number of non-clinical alarms, and quickly identify clinical alarms by using the target classification model, which is convenient for medical staff to quickly take treatment measures.

[0033] In an alternative implementation, in step S101, the process of constructing the key data set may include the following steps S1011 to S1013.

[0034] S1011. Construct an initial data set.

[0035] In this embodiment, the initial data set includes a number of initial samples, and each initial sample includes an initial feature matrix and its corresponding initial label matrix.

[0036] Among them, the initial feature matrix includes time series feature vectors of multiple monitoring indicators within a set observation duration, and one time series feature vector reflects the change of one monitoring indicator within the set observation duration. The initial label matrix includes an alarm label vector corresponding to each monitoring indicator, and one alarm label vector reflects the situation of triggering a clinical alarm by one monitoring indicator within the set observation duration.

[0037] Some of all the monitoring indicators of the medical device are device indicators and some are patient-related indicators. Taking a ventilator as an example, the device indicators may include: the operating state of the device (working, standby, shutdown, etc.), the set parameters of the device, the device ID, the device type, the device model, and the device manufacturer, etc., while the patient-related indicators may include, but are not limited to: (1) the patient and the patient's gender; (2) detailed information on respiratory support treatment, such as the type of mechanical ventilation (invasive or non-invasive), the intubation status, and the extubation event; (3) ventilator settings and monitoring parameters, including the ventilator mode, tidal volume, respiratory rate, inspiratory pressure, positive end-expiratory pressure, and fraction of inspired oxygen; (4) monitoring parameters, including heart rate, blood pressure, blood oxygen saturation, respiratory rate, and other vital sign data.

[0038] This example is only for illustration and is not limited here.

[0039] S1012. Based on all the initial feature matrices and their corresponding initial label matrices, screen out a plurality of key indicators from all the monitoring indicators.

[0040] In this embodiment, the key indicators may be monitoring indicators that are closely related to the occurrence of clinical alarms.

[0041] S1013. For each initial sample, delete the content other than the time series feature vectors and alarm label vectors of each key indicator in the initial sample to obtain key samples.

[0042] Exemplarily, assume that in an initial sample, the initial feature matrix includes time series feature vectors of 93 monitoring indicators, and the initial label matrix includes alarm label vectors of 93 monitoring indicators. After being processed by step S1013, in the obtained key sample, the initial feature matrix includes time series feature vectors of 15 key indicators, and the initial label matrix includes alarm label vectors of 15 key indicators. This example is only for illustration and is not limited herein.

[0043] Through the above steps S1011 - S1013, the present invention can eliminate the data of each monitoring indicator in the original data set that is not related to clinical alarms, and only retain the data of each key indicator related to clinical alarms, avoiding the interference of irrelevant data. For example, the ventilator device ID, device type, device model, and device manufacturer are all irrelevant to clinical alarms, while monitoring indicators such as oxygen concentration and positive end-expiratory pressure collected by the ventilator are all related to clinical alarms.

[0044] In an alternative implementation, an initial data set can be formed based on several consecutive historical acquisition data. Therefore, for step S1011, the process of constructing the initial data set may include the following sub-steps S10111 - S10116.

[0045] S10111. Obtain several historical acquisition data and their corresponding acquisition times.

[0046] In this embodiment, first, it is necessary to obtain several historical acquisition data and their corresponding acquisition times. The historical acquisition data includes the status data and trigger situations of each monitoring indicator when the medical device is running.

[0047] These historical acquisition data can be: those collected regularly when the medical device is running, and those automatically collected when the medical device triggers an alarm. Among them, when collecting, the status data and trigger situations of each monitoring indicator are recorded in detail. The status data reflects the specific value of the monitoring indicator within the acquisition time, and the trigger situation describes the type of alarm triggered by the monitoring indicator. The alarm types include three types: triggering clinical treatment, triggering technical treatment, and not triggering treatment, which provides a clear classification basis for the subsequent generation of label vectors.

[0048] S10112. Based on the set observation duration and the acquisition times of each historical acquisition data, divide the several historical acquisition data into multiple observation sets.

[0049] In this embodiment, a number of historical acquisition data can be divided into multiple observation sets based on the relationship between the set observation duration and the acquisition times of each piece of historical acquisition data.

[0050] The division rule here is to ensure that the difference between the acquisition times of any two pieces of historical acquisition data in each observation set does not exceed the set observation duration. This process aims to ensure that the data contained in each observation set has a certain temporal continuity, so as to accurately reflect the dynamic change characteristics of the monitoring indicators within the set observation duration.

[0051] Exemplarily, please combine Figure 2 , and assume that the acquisition times of 6 pieces of historical acquisition data in an observation set are Figure 2 t1 to t6 in , where except for the historical acquisition data corresponding to the t3 time which is automatically acquired due to a triggered alarm, the other 5 pieces of historical acquisition data are acquired at an interval period T, and in the observation set, the time difference ∆t between the t1 time and the t6 time satisfies ∆t ≤ the set observation duration T0.

[0052] Taking a ventilator as an example, the interval period T can be 5 minutes, 10 minutes, 15 minutes, etc., and the set observation duration T0 can be 25 minutes, 30 minutes, 40 minutes, etc. This example is only for illustration and is not limited here.

[0053] S10113. For any observation set, normalize all the status data of each monitoring indicator in the observation set to obtain the time series feature vectors corresponding to each monitoring indicator.

[0054] In this embodiment, all the status data of a monitoring indicator in an observation set can be normalized to obtain the time series feature vector corresponding to this monitoring indicator.

[0055] S10114. Perform encoding conversion on all the triggering situations of each monitoring indicator in the observation set to obtain the alarm label vectors corresponding to each monitoring indicator.

[0056] In this embodiment, all the triggering situations of a monitoring indicator in an observation set can be encoded and converted into a label for representation, and thus the alarm label vector corresponding to this monitoring indicator can be obtained. During the encoding conversion process, only when the triggering situation belongs to clinical treatment, the label is 1, and when the triggering situation belongs to technical treatment and no alarm is triggered, it is 0. In this way, the alarm label vector provides a clear supervision signal for subsequent model training, enabling the algorithm to learn to identify clinical alarms.

[0057] Exemplarily, if the number of monitoring indicators is M and an observation set includes S pieces of historical acquisition data, then for the m-th monitoring indicator among the M monitoring indicators: (1) The way to obtain the time series feature vector of the m-th monitoring index is as follows: The S state data corresponding to the m-th monitoring index can be normalized to obtain the time series feature vector corresponding to the m-th monitoring index. Therefore, a time series feature vector can represent the dynamic change of a monitoring index within the set observation duration. (2) The alarm label vector of the m-th monitoring index includes S labels, and the way to obtain it is as follows: If the m-th monitoring index triggers technical handling or does not trigger an alarm at the acquisition time of the s-th historical acquisition data, then the s-th label is set to 0. If the m-th monitoring index triggers clinical handling at the acquisition time of the s-th historical acquisition data, then the s-th label is set to 1.

[0058] Combined with Figure 2 , if an observation set includes 6 pieces of historical acquisition data, then the time series feature vector of each monitoring index includes 6 eigenvalues, and each alarm label vector includes 6 labels. If Figure 2 a monitoring index A triggers a clinical alarm at time t3 in , then the alarm label vector of monitoring index A is: 001000.

[0059] The above example is only for illustration, and the embodiments of the present invention do not limit the number of historical acquisition data, the number and time of clinical alarms in the observation set.

[0060] S10115. Based on the time series feature vectors and alarm label vectors corresponding to all monitoring indexes, obtain the initial sample corresponding to the observation set.

[0061] In this embodiment, for an observation set, the time series feature vectors and alarm label vectors obtained by processing the observation set through the above steps S10111~S10115 can form an initial sample.

[0062] This initial sample completely contains the state change information of all monitoring indexes and their corresponding alarm trigger situations within the set observation duration.

[0063] S10116. Traverse each observation set to obtain the initial data set.

[0064] By processing each observation set one by one through the above steps S10111~S10115, a complete initial data set can be obtained. The initial data set not only covers rich dynamic features of monitoring indexes, but also retains clear alarm label information, laying a solid foundation for the subsequent construction of the key data set and the training of the target classification model. The whole process is logically rigorous and closely linked, ensuring the efficient conversion from the original historical acquisition data to the structured initial data set, while maximizing the retention of the effective information in the original data.

[0065] In an alternative implementation, the correlation between each monitoring indicator and the triggering of a clinical alarm can be analyzed first based on mathematical statistics methods, and then the importance of the monitoring indicators can be analyzed using a machine learning model to ensure the accuracy of feature screening.

[0066] Correspondingly, for the process of screening multiple key indicators from all monitoring indicators in step S1012 above, the following sub-steps S10121 to S10124 can be included.

[0067] S10121. Based on all initial feature matrices and their corresponding initial label matrices, calculate the correlation coefficient between each monitoring indicator and the occurrence of a clinical alarm in the medical device.

[0068] In this embodiment, the correlation strength between each monitoring indicator and the occurrence of a clinical alarm is quantified by calculating the correlation coefficient. This calculation method of the correlation coefficient ensures the independent analysis of each monitoring indicator, so as to objectively measure its potential role in predicting clinical alarms.

[0069] Exemplarily, the correlation coefficient between a monitoring indicator A and the occurrence of a clinical alarm in the medical device can be the Spearman rank correlation coefficient, and when calculating, the time series feature vector of monitoring indicator A and the alarm label vector need to be used for calculation. The specific calculation process is prior art and will not be elaborated here.

[0070] S10122. Take each monitoring indicator with a correlation coefficient exceeding the preset threshold as a candidate indicator.

[0071] In this embodiment, the monitoring indicators with a correlation coefficient exceeding the preset threshold are taken as candidate indicators. In this way, during the indicator screening, only the candidate indicators with a strong correlation with the triggering of a clinical alarm are retained for the subsequent analysis stage.

[0072] Exemplarily, the size of the preset threshold can be set flexibly, such as 0.7, 0.75, etc.

[0073] S10123. Delete other contents in each initial sample except the time series feature vectors and time series feature vectors of each candidate indicator to obtain an intermediate data set.

[0074] In this embodiment, the initial data set is further converted into an intermediate data set, so that the intermediate data set only focuses on each candidate indicator, thus providing a clearer and more focused data basis for the subsequent importance analysis.

[0075] S10124. Based on the intermediate data set, use at least one machine learning algorithm to perform importance analysis on all candidate indicators to screen out multiple key indicators from all candidate indicators.

[0076] In this embodiment, at least one machine learning algorithm is used to perform importance analysis on all candidate indicators, with the aim of further identifying those key indicators that truly have a decisive impact on clinical alarm judgment from the candidate indicators.

[0077] Taking the adoption of two machine learning algorithms as an example, the process of screening out key indicators may include the following steps a1 to a4.

[0078] Step a1: Based on the intermediate data set, use the first classification algorithm and the second classification algorithm to respectively determine the first importance coefficient and the second importance coefficient of each candidate indicator.

[0079] In this embodiment, through two different classification algorithms, the importance of each candidate indicator is measured from different perspectives. The selection of the first classification algorithm and the second classification algorithm should be determined according to the actual application scenario and data characteristics.

[0080] Exemplarily, the first classification algorithm can adopt the logistic regression algorithm, and the second classification algorithm can adopt the adaptive boosting algorithm (i.e., AdaBoost algorithm) or the gradient learning algorithm.

[0081] Step a2: Sort all candidate indicators respectively in the order from large to small of the first importance coefficient and in the order from large to small of the second importance coefficient to obtain the first sorting result and the second sorting result.

[0082] Taking a ventilator as an example, the first importance coefficients of the top 15 monitoring indicators in the first sorting result are as Figure 3 shown. Figure 3 The 15 monitoring indicators in

[0083] Figure 3The above is only an example, and the embodiments of the present invention do not limit the number of monitoring indicators and the specific meanings of each monitoring indicator.

[0084] Step a3: Take each candidate indicator in the first sorting result and the second sorting result that is located in the preset first position interval as a key indicator.

[0085] Step a4: Determine the intersection between each candidate indicator in the first sorting result that is located in the preset second position interval and each candidate indicator in the second sorting result that is located in the second position interval, and take each candidate indicator in the intersection as a key indicator.

[0086] In this embodiment, the first position interval and the second position interval are adjacent.

[0087] Exemplarily, the first position interval can be [1, 4], and the second position interval can be [5, 10]. In this way, the first 4 candidate indicators in the first sorting result and the first 4 candidate indicators in the second sorting result are both taken as key indicators, and each one in the intersection of the first sorting result and the second sorting result from the 5th candidate indicator to the 10th candidate indicator is taken as a key indicator. This example is only for illustration and is not limited here.

[0088] In this way, by adopting a dual sorting mechanism, the priority of candidate indicators from different algorithm perspectives can be comprehensively considered. Since there may be limitations or biases in indicator selection by a single machine learning algorithm, the reliability and robustness of key indicator screening can be effectively improved through the collaborative approach of multiple machine learning algorithms.

[0089] Taking the target algorithm (i.e., the first classification algorithm or the second classification algorithm) as an example, the method for obtaining the target importance coefficient (i.e., the first importance coefficient or the second importance coefficient) of each candidate indicator in the above step a1 includes the following three steps: (1) Based on the target algorithm, construct a target model; (2) Use the intermediate data set to train the target model to obtain the trained target model; (3) Take the model weight corresponding to each candidate indicator in the trained target model as the target importance coefficient.

[0090] By using the intermediate data set to train the target model, the model will gradually adjust its internal parameters to minimize the prediction error. During this process, the weight values of different candidate indicators will be dynamically updated until the optimal state is reached.

[0091] Taking the target model as a logistic regression model as an example, in the initial state, the weight values of all candidate indicators are evenly distributed. Subsequently, the loss function is calculated through multiple iterations and the weight values are modified, so that the model prediction results gradually approach the true labels. After training, the weight values corresponding to some candidate indicators may tend to zero, indicating that these candidate indicators have no significant association with the prediction of event authenticity and can therefore be excluded; while the candidate indicators with larger weight values are considered to play a more important role in model prediction.

[0092] The present invention makes full use of the advantages of machine learning models, ensuring the accuracy and reliability of the importance evaluation of candidate indicators.

[0093] In an optional implementation manner, a key data set can be used to train and perform performance tests on a plurality of preset algorithm models, and finally select the one with the best performance as the target classification model. Correspondingly, in the above step S102, the process of using the key data set to train a plurality of preset algorithm models to screen out the target classification model from the obtained plurality of candidate classification models may include the following sub-steps S1021 to S1024.

[0094] S1021. Divide the key data set into a training set and a test set.

[0095] Optionally, the commonly used 10-fold cross-validation method in model evaluation technology can be adopted, that is, the data set is divided into several subsets to more accurately evaluate the performance of the model and enhance the reliability of model evaluation.

[0096] In this embodiment, the entire key data set is randomly divided into 10 folds, and the data is divided into a training set and a test set in a ratio of 7:3.

[0097] S1022. Use the training set to train a variety of algorithm models respectively to obtain a plurality of candidate classification models.

[0098] S1023. Based on the test set, perform performance evaluation on each candidate classification model respectively to obtain at least one performance index corresponding to each candidate classification model.

[0099] In this embodiment, the performance indexes may include: one or more of the area under the receiver operating characteristic curve (AUC-ROC), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), F1-score, Kappa value, etc.

[0100] Exemplarily, multiple algorithm models may include but are not limited to: Gaussian Naive Bayes classification model (Gaussian Naive Bayes, NB), K-Nearest Neighbors classification model (K-Nearest Neighbors, KNN), Linear Discriminant Analysis classification model (Linear Discriminant Analysis, LDA), Classification and Regression Trees classification model (Classification and Regression Trees, CART), Support Vector Machine classification model (Support Vector Machine, SVM), eXtreme Gradient Boosting classification model (eXtreme Gradient Boosting, XGBoost), Logistic Regression classification model (Logistic Regression, LR), Random Forest classification model (Random Forest, RF), etc. Please combine with Figure 4 , Figure 4 That is the performance evaluation result of the inventor for these several models.

[0101] S1024. Based on at least one performance metric corresponding to each candidate classification model, train a target classification model from each candidate classification model.

[0102] In an alternative implementation, each candidate classification model can be sorted separately in descending order according to each performance metric to obtain the model sorting result corresponding to each performance metric; then, among the model sorting results corresponding to all performance metrics, the candidate classification model with the most occurrences in the first place is used as the target prediction model.

[0103] In another alternative implementation, the respective performance metrics corresponding to each candidate classification model can be weighted and averaged respectively to obtain the performance evaluation score of each candidate classification model, and the candidate classification model with the highest performance evaluation score is used as the target classification model.

[0104] Combined with Figure 4 , if the weighted scoring method is used to process the performance metrics of the Figure 4 eight models respectively, the XGBoost model is finally determined as the target classification model.

[0105] The above content introduces the process of obtaining the target classification model. Next, the application process of the target classification model is introduced.

[0106] Please refer to Figure 5 , Figure 5 which is a schematic flowchart of an alarm classification method based on a machine learning model provided by an embodiment of the present invention. This method can be applied to electronic devices such as smartphones, personal notebooks, personal computers, servers, etc. AsFigure 5 , the method includes the following steps S201 to S204.

[0107] S201. Receive the acquisition data of the medical device collected by the acquisition device each time.

[0108] In this embodiment, the acquisition device may be an independent collector dedicated to acquiring medical devices. The acquisition data may include the status data and status flag values of multiple monitoring indicators; among them, the status flag value of 0 indicates that the acquisition data is acquired by the acquisition device at regular intervals, and the status flag value of 1 indicates that the acquisition data is acquired when the medical device triggers an alarm.

[0109] S202. If the status flag value of any monitoring indicator in the received acquisition data is 1, use the moment after waiting for a preset duration as the judgment moment.

[0110] In this embodiment, when the status flag value of any monitoring indicator in the received acquisition data is 1, it indicates that the medical device has triggered an alarm. At this time, the system will not immediately perform classification judgment, but set a waiting period of a preset duration (such as 5 minutes). This waiting strategy aims to ensure that there is enough data for subsequent analysis of whether the alarm is a clinical alarm, thereby improving the classification accuracy. The moment determined after the waiting is the judgment moment.

[0111] S203. Generate a feature matrix to be recognized based on all the acquisition data within the set observation duration before the judgment moment.

[0112] In this embodiment, generate a feature matrix to be recognized based on all the acquisition data within the set observation duration before the above judgment moment.

[0113] Optionally, among the M pieces of acquisition data within the set observation duration before the judgment moment in all the acquisition data, obtain the M status data of each key indicator, and perform normalization processing on the M status data of each key indicator respectively, then the feature matrix to be recognized can be obtained.

[0114] S204. Input the feature matrix to be recognized into the target classification model to obtain the classification result.

[0115] In this embodiment, input the generated feature matrix to be recognized into the target classification model to obtain the classification result. The classification result reflects whether the alarm triggered by the medical device is a clinical alarm or a non-clinical alarm.

[0116] The alarm classification method provided by the present invention can, when the medical device triggers an alarm, timely convert the status data of each key indicator within a certain time before and after the triggered alarm into a feature matrix to be recognized, and then input the feature matrix to be recognized into the target classification model to determine whether the alarm is a clinical alarm or a non-clinical alarm, ensuring the timeliness and accuracy of alarm recognition.

[0117] Based on the same inventive concept, an embodiment of the present invention also provides a model training device 200. Please refer to Figure 1 Figure Figure 6 , Figure 6 which is a functional module diagram of the model training device provided by the embodiment of the present invention, including: a data set construction module 210 and a model screening module 220.

[0118] The data set construction module 210 is configured to construct a key data set, and each key sample in the key data set reflects whether a clinical alarm occurs in a medical device within a set observation duration; The model screening module 220 is configured to train a plurality of preset algorithm models by using the key data set, so as to screen out a target classification model from the obtained plurality of candidate classification models, and the target classification model is used to determine whether an alarm generated by the medical device belongs to a clinical alarm.

[0119] Optionally, the key data set includes a plurality of key samples. During the process of constructing the key data set, the data set construction module 210 may specifically be configured to: Construct an initial data set, and the initial data set includes a plurality of initial samples; the initial samples include an initial feature matrix and its corresponding initial label matrix; the initial feature matrix includes time series feature vectors of a plurality of monitoring indicators within a set observation duration, and the initial label matrix includes an alarm label vector corresponding to each monitoring indicator; based on all the initial feature matrices and their corresponding initial label matrices, screen out a plurality of key indicators from all the monitoring indicators; for each initial sample, delete other contents in the initial sample except the time series feature vectors and alarm label vectors of each key indicator, so as to obtain a key sample.

[0120] Optionally, during the process of constructing the initial data set, the data set construction module 210 may specifically be configured to: obtain a plurality of historical acquisition data and their corresponding acquisition times, and the historical acquisition data includes status data and trigger conditions of each monitoring indicator when the medical device is running; based on the set observation duration and the acquisition times of each historical acquisition data, divide the plurality of historical acquisition data into a plurality of observation sets; wherein, the difference between the acquisition times of any two historical acquisition data in each observation set does not exceed the set observation duration; for any one observation set, perform normalization processing on all the status data of each monitoring indicator in the observation set to obtain time series feature vectors corresponding to each monitoring indicator; perform encoding conversion on all the trigger conditions of each monitoring indicator in the observation set to obtain alarm label vectors corresponding to each monitoring indicator; based on the time series feature vectors and alarm label vectors corresponding to all the monitoring indicators, obtain an initial sample corresponding to the observation set; traverse each observation set to obtain the initial data set.

[0121] Optionally, in the process of screening out multiple key indicators from all monitoring indicators based on all initial feature matrices and their corresponding initial label matrices, the dataset construction module 210 can specifically be used to: calculate the correlation coefficient between each monitoring indicator and the occurrence of clinical alarms in the medical device based on all initial feature matrices and their corresponding initial label matrices; use each monitoring indicator with a correlation coefficient exceeding a preset threshold as a candidate indicator; delete other content in each initial sample except the time series feature vectors of each candidate indicator to obtain an intermediate dataset; based on the intermediate dataset, use at least one machine learning algorithm to perform importance analysis on all candidate indicators to screen out multiple key indicators from all candidate indicators.

[0122] Optionally, in the process of performing importance analysis on all candidate indicators using at least one machine learning algorithm based on the intermediate dataset to screen out multiple key indicators from all candidate indicators, the dataset construction module 210 can specifically be used to: determine the first importance coefficient and the second importance coefficient of each candidate indicator using the first classification algorithm and the second classification algorithm respectively based on the intermediate dataset; sort all candidate indicators in descending order of the first importance coefficient and in descending order of the second importance coefficient respectively to obtain a first sorting result and a second sorting result; use each candidate indicator located in a preset first position interval in the first sorting result and the second sorting result as a key indicator; determine the intersection between each candidate indicator located in a preset second position interval in the first sorting result and each candidate indicator located in the second position interval in the second sorting result, and use each candidate indicator in the intersection as a key indicator.

[0123] Optionally, the target importance coefficient of each candidate indicator is obtained through the following method: construct a target model based on the target algorithm; the target algorithm is the first classification algorithm or the second classification algorithm; use the intermediate dataset to train the target model to obtain a trained target model; use the model weight corresponding to each candidate indicator in the trained target model as the target importance coefficient; the target importance coefficient is the first importance coefficient or the second importance coefficient.

[0124] Optionally, in the process of training a preset multiple algorithm models using the key dataset to screen out a target classification model from the obtained multiple candidate classification models, the model screening module 220 can specifically be used to: divide the key dataset into a training set and a test set; use the training set to train multiple algorithm models respectively to obtain multiple candidate classification models; based on the test set, perform performance evaluation on each candidate classification model respectively to obtain at least one performance indicator corresponding to each candidate classification model; based on at least one performance indicator corresponding to each candidate classification model, train a target classification model from each candidate classification model.

[0125] Optionally, during the process of training a target classification model from each candidate classification model based on at least one performance metric corresponding to each candidate classification model, the model screening module 220 can specifically be configured to: sort each candidate classification model in descending order according to each performance metric to obtain a model sorting result corresponding to each performance metric; use the candidate classification model with the most occurrences in the first place among the model sorting results corresponding to all performance metrics as the target prediction model; alternatively, the model screening module 220 can specifically be configured to: perform weighted averaging on each performance metric corresponding to each candidate classification model to obtain a performance evaluation score for each candidate classification model, and use the candidate classification model with the highest performance evaluation score as the target classification model.

[0126] Based on the same inventive concept as Figure 5 the present invention embodiment also provides an alarm classification device 400 based on a machine learning model. Please refer to Figure 7 , Figure 7 which is a functional module diagram of the alarm classification device based on a machine learning model provided by the present invention embodiment, including: a data receiving module 410, a vectorization module 420, and an alarm classification module 430.

[0127] The data receiving module 410 is configured to receive the acquisition data of the medical device collected by the acquisition device each time. The acquisition data includes the status data and status flag values of multiple monitoring metrics; wherein, the status flag value of 0 represents that the acquisition data is collected by the acquisition device at regular intervals, and the status flag value of 1 represents that the acquisition data is collected when the medical device is sensed to trigger an alarm; The vectorization module 420 is configured to, when the status flag value of any monitoring metric in the received acquisition data is 1, use the moment after waiting for a preset duration as the judgment moment; The vectorization module 420 is further configured to generate a feature matrix to be recognized based on all the acquisition data within a set observation duration before the judgment moment; The alarm classification module 430 is configured to input the feature matrix to be recognized into the target classification model to obtain a classification result; wherein, the classification result reflects whether the alarm triggered by the medical device is a clinical alarm or a non-clinical alarm, and the target classification model is obtained through model training by the above model training method.

[0128] Those skilled in the art can clearly understand that for the sake of convenience and brevity of description, the specific working processes of the above-described model training device 200 and the alarm classification device 400 based on a machine learning model can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0129] Please refer to Figure 8 , Figure 8A schematic structural diagram of an electronic device provided by an embodiment of the present invention. The electronic device 300 includes a processor 310, a memory 320, and a bus 330. The processor 310 is connected to the memory 320 through the bus 330.

[0130] The memory 320 can be used to store software programs. For example, software programs corresponding to the model training device 200 and / or the alarm classification device 400 based on a machine learning model provided by an embodiment of the present invention. The processor 310 executes various functional applications and data processing by running the software programs stored in the memory 320 to implement the model training method provided by an embodiment of the present invention, and / or, the alarm classification method based on a machine learning model.

[0131] Among them, the memory 320 can be, but is not limited to: RAM (Random Access Memory, random access memory), ROM (Read Only Memory, read only memory), FLASH (flash memory), PROM (Programmable Read-Only Memory, programmable read only memory), EPROM (Erasable Programmable Read-Only Memory, erasable read only memory), EEPROM (Electrically Erasable Programmable Read-Only Memory, electrically erasable read only memory), etc.

[0132] The processor 310 can be an integrated circuit chip with signal processing capabilities. The processor 310 can be a general-purpose processor, including: CPU (Central Processing Unit, central processor), NP (Network Processor, network processor), SoC (System on Chip, system-on-chip), etc.; it can also be: DSP (Digital Signal Processing, digital signal processor), ASIC (Application Specific Integrated Circuit, application-specific integrated circuit), FPGA (Field-Programmable Gate Array, field programmable gate array), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0133] It can be understood that Figure 8 The structure shown is only schematic. The electronic device 300 may further include more or fewer components than those shown in Figure 8 or have a different configuration from that shown in Figure 8 shown. Figure 8The components shown in [description] can be implemented by hardware, software, or a combination thereof.

[0134] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, it implements the model training method disclosed in the above embodiment and / or the alarm classification method based on a machine learning model. The computer-readable storage medium may be, but is not limited to, various media that can store program codes, such as a USB flash drive, a mobile hard disk, a ROM, a RAM, a PROM, an EPROM, an EEPROM, a FLASH disk, or an optical disc.

[0135] In summary, an embodiment of the present invention provides a model training method, an alarm classification method based on a machine learning model, a device, an electronic device, and a storage medium. After constructing a key data set, the key data set can be used to train a plurality of preset algorithm models to screen out a target classification model from the obtained plurality of candidate classification models, and then the target classification model can be used to identify whether the alarm generated by a medical device belongs to a clinical alarm, thereby reducing the pressure on medical staff facing a large number of non-clinical alarms, and quickly identifying clinical alarms using the target classification model, which is convenient for medical staff to quickly take rescue measures.

[0136] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A model training method, characterized in that, Including: Construct a key data set, where each key sample in the key data set reflects whether a clinical alarm occurs for a medical device within a set observation duration; Use the key data set to train a plurality of preset algorithm models, so as to screen out a target classification model from the obtained plurality of candidate classification models, and the target classification model is used to determine whether an alarm generated by the medical device belongs to a clinical alarm.

2. The method according to claim 1, characterized in that, The key data set includes a number of key samples, and the step of constructing the key data set includes: Construct an initial data set, where the initial data set includes a number of initial samples; the initial samples include an initial feature matrix and its corresponding initial label matrix; the initial feature matrix includes time series feature vectors of multiple monitoring indicators within the set observation duration, and the initial label matrix includes an alarm label vector corresponding to each monitoring indicator; Based on all the initial feature matrices and their corresponding initial label matrices, screen out multiple key indicators from all the monitoring indicators; For each initial sample, delete other content in the initial sample except the time series feature vectors and alarm label vectors of each key indicator to obtain the key sample.

3. The method according to claim 2, wherein The step of constructing the initial data set includes: Obtain a number of historical acquisition data and their corresponding acquisition times, where the historical acquisition data includes the status data and trigger conditions of each monitoring indicator when the medical device is running; Based on the set observation duration and the acquisition times of each historical acquisition data, divide the number of historical acquisition data into multiple observation sets; where the difference between the acquisition times of any two historical acquisition data in each observation set does not exceed the set observation duration; For any one of the observation sets, perform normalization processing on all the status data of each monitoring indicator in the observation set to obtain the time series feature vectors corresponding to each monitoring indicator; Perform encoding conversion on all the trigger conditions of each monitoring indicator in the observation set to obtain the alarm label vectors corresponding to each monitoring indicator; Based on the time series feature vectors and alarm label vectors corresponding to all the monitoring indicators, obtain the initial sample corresponding to the observation set; Traverse each observation set to obtain the initial data set.

4. The method according to claim 2, wherein The step of screening out multiple key indicators from all the monitoring indicators based on all the initial feature matrices and their corresponding initial label matrices includes: Based on all the initial feature matrices and their corresponding initial label matrices, calculate the correlation coefficient between each monitoring indicator and the occurrence of a clinical alarm for the medical device; Use each monitoring indicator with a correlation coefficient exceeding a preset threshold as a candidate indicator; Delete other content in each initial sample except the time series feature vectors and time series feature vectors of each candidate indicator to obtain an intermediate data set; Based on the intermediate data set, use at least one machine learning algorithm to perform importance analysis on all candidate indicators, so as to screen out multiple key indicators from all candidate indicators.

5. The method according to claim 4, wherein The step of performing importance analysis on all candidate metrics using at least one machine learning algorithm based on the intermediate data set to screen out multiple key metrics from all the candidate metrics includes: Based on the intermediate data set, use the first classification algorithm and the second classification algorithm to respectively determine the first importance coefficient and the second importance coefficient of each candidate metric; Sort all the candidate metrics respectively in descending order of the first importance coefficient and in descending order of the second importance coefficient to obtain a first sorting result and a second sorting result; Regard each candidate metric located in a preset first position interval in the first sorting result and the second sorting result as the key metric; Determine the intersection between each candidate metric located in a preset second position interval in the first sorting result and each candidate metric located in the second position interval in the second sorting result, and regard each candidate metric in the intersection as the key metric.

6. The method according to claim 5, characterized in that, The target importance coefficient of each candidate metric is obtained by the following method: Based on the target algorithm, construct a target model; the target algorithm is the first classification algorithm or the second classification algorithm; Use the intermediate data set to train the target model to obtain a trained target model; Regard the model weight corresponding to each candidate metric in the trained target model as the target importance coefficient; the target importance coefficient is the first importance coefficient or the second importance coefficient.

7. The method according to claim 1, characterized in that, The step of using the key data set to train a preset plurality of algorithm models to screen out a target classification model from the obtained plurality of candidate classification models includes: Divide the key data set into a training set and a test set; Use the training set to train various algorithm models respectively to obtain a plurality of candidate classification models; Based on the test set, perform performance evaluation on each candidate classification model respectively to obtain at least one performance metric corresponding to each candidate classification model; Based on at least one performance metric corresponding to each candidate classification model respectively, train the target classification model from each candidate classification model.

8. The method according to claim 7, wherein The step of training the target classification model from each candidate classification model based on at least one performance metric corresponding to each candidate classification model respectively includes: Sort each candidate classification model respectively in descending order of each performance metric to obtain a model sorting result corresponding to each performance metric; Regard the candidate classification model with the most occurrences in the first place in the model sorting results corresponding to all the performance metrics as the target classification model; Or, Respectively perform weighted averaging on each performance metric corresponding to each candidate classification model to obtain a performance evaluation score for each candidate classification model, and regard the candidate classification model with the largest performance evaluation score as the target classification model.

9. An alarm classification method based on a machine learning model, characterized in that, Include: Receive the acquisition data of the medical device collected each time by the acquisition device, where the acquisition data includes the status data of multiple monitoring indicators and the status flag value; among them, the status flag value of 0 indicates that the acquisition data is collected by the acquisition device at regular intervals, and the status flag value of 1 indicates that the acquisition data is collected when the medical device is sensed to trigger an alarm; If the status flag value of any one of the monitoring indicators in the received acquisition data is 1, then use the moment after waiting for a preset duration as the judgment moment; Generate a feature matrix to be recognized based on all the acquisition data within a set observation duration before the judgment moment; Input the feature matrix to be recognized into the target classification model to obtain a classification result; among them, the classification result reflects that the alarm triggered by the medical device is a clinical alarm or a non-clinical alarm, and the target classification model is obtained by training the model through the method described in any one of claims 1-8.

10. A model training device, characterized in that, Include: A dataset construction module for constructing a key dataset, where each key sample in the key dataset reflects whether a clinical alarm occurs in the medical device within a set observation duration; A model screening module for training a preset number of algorithm models using the key dataset to screen out a target classification model from the obtained multiple candidate classification models, and the target classification model is used to determine whether the alarm generated by the medical device belongs to a clinical alarm.

11. An alarm classification device based on a machine learning model, characterized in that, Include: A data reception module for receiving the acquisition data of the medical device collected each time by the acquisition device, where the acquisition data includes the status data of multiple monitoring indicators and the status flag value; among them, the status flag value of 0 indicates that the acquisition data is collected by the acquisition device at regular intervals, and the status flag value of 1 indicates that the acquisition data is collected when the medical device is sensed to trigger an alarm; A vectorization module for using the moment after waiting for a preset duration as the judgment moment when the status flag value of any one of the monitoring indicators in the received acquisition data is 1; The vectorization module is further configured to generate a feature matrix to be recognized based on all the acquisition data within a set observation duration before the judgment moment; An alarm classification module for inputting the feature matrix to be recognized into the target classification model to obtain a classification result; among them, the classification result reflects that the alarm triggered by the medical device is a clinical alarm or a non-clinical alarm, and the target classification model is obtained by training the model through the method described in any one of claims 1-8.

12. An electronic device, characterized in that, Include: A memory and a processor, where the memory stores a software program, and when the electronic device runs, the processor executes the software program to implement the method described in any one of claims 1-8, or the method described in claim 9.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, it implements the method described in any one of claims 1-8, or the method described in claim 9.

Images