Method, device and related equipment for detecting human-machine asynchrony during mechanical ventilation of ventilator

CN113689948BActive Publication Date: 2024-11-29SHENZHEN INST OF ADVANCED TECH +1
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202110957388.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-08-18
Publication Date
2024-11-29
Estimated Expiration
2041-08-18

AI Technical Summary

Technical Problem

In the prior art, human-machine asynchronous detection of ventilators relies on manual labeling characteristics, resulting in high cost, low timeliness and poor interpretation, making it difficult to meet the actual deployment needs.

Method used

Using the Gradient Boosting Decision Tree (GBDT) model and Logistic Regression (LR) model, the GBDT model is used to automatically extract the feature information of the ventilator during mechanical ventilation, and input these features into the LR model for classification and detection, realizing asynchronous automatic human-machine Test.

Benefits of technology

It reduces the cost of asynchronous detection of human-machine machines, improves the real-time and interpretability of detection, and can more quickly and accurately detect and identify human-machine asynchronous types.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113689948B_ABST
    Figure CN113689948B_ABST
Patent Text Reader

Abstract

The present invention discloses a method, device and related equipment for detecting human-machine asynchrony of mechanical ventilation of a ventilator, wherein the method comprises: constructing a GBDT model and an LR model, obtaining a first training data set, inputting the first training data set into the GBDT model for training, obtaining a first leaf node index value set and then inputting it into the LR model for training, obtaining a trained LR model, and processing the data set to be detected corresponding to the ventilator based on the trained GBDT model and the trained LR model to obtain a detection result of human-machine asynchrony. By adopting the embodiments of the present application, the cost of human-machine asynchrony detection can be reduced, and the real-time and interpretability of human-machine asynchrony detection can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of medical device technology, and in particular to a method, device and related equipment for detecting human-machine asynchrony during mechanical ventilation of a ventilator. Background Art

[0002] A ventilator is an important device that provides respiratory support. It is widely used in intensive care units and general departments in hospitals to provide important respiratory support for patients with respiratory dysfunction. In addition, many ventilators have entered families and become household devices to assist daily sleep.

[0003] Generally speaking, one of the most important functions of a ventilator is its ventilation sensitivity. That is, whether the ventilator can provide the same frequency of air supply / inflation support when the patient or user needs to inhale; and whether the ventilator can switch to the corresponding exhalation state in time when the patient or user needs to exhale. In other words, the ventilator needs to detect the patient's inhalation needs in time, detect the time to switch to exhalation after the end of inhalation, and so on.

[0004] However, during the process of ventilation between the ventilator and the patient or user, there is still a high probability that the ventilation of the ventilator will not match or conflict with the breathing needs of the patient or user, that is, human-machine asynchronous phenomenon, that is, asynchronous phenomenon, occurs in the mechanical ventilation process between the ventilator and the patient or user based on a certain type of airway. Specifically, when the patient or user needs to inhale and is in the inspiratory phase, the ventilator does not provide sufficient air supply / inflation support, or is not even in the air supply / inflation state; and when the patient or user needs to exhale and is in the expiratory phase, the ventilator does not switch to the expiratory state in time. If the ventilator is in a human-machine asynchronous state for a long time, it will not be able to effectively provide active help to the patient or user, and may even cause suffocation of the patient or user.

[0005] However, in the existing technology, domain experts and professional medical staff are often required to manually mark features to identify human-machine asynchrony, which naturally requires a lot of time from experts and professional medical staff, with high time and financial costs. In addition, the method of relying on expert manual identification is not real-time, which is not conducive to actual deployment, promotion and implementation. On the other hand, the scalability, generalization, and personalization of features manually extracted by experts are yet to be verified, so the interpretability is still limited. Therefore, how to provide a low-cost, high-real-time and highly interpretable ventilator human-machine asynchrony detection solution is an urgent problem to be solved. Summary of the invention

[0006] The embodiments of the present application provide a method, device and related equipment for detecting human-machine asynchronous detection of mechanical ventilation of a ventilator, which can reduce the cost of human-machine asynchronous detection and improve the real-time and interpretability of human-machine asynchronous detection.

[0007] In a first aspect, an embodiment of the present application provides a method for detecting human-machine asynchrony during mechanical ventilation of a ventilator, the method comprising:

[0008] Constructing a GBDT model and an LR model; the GBDT model includes one or more tree sets, each tree set includes one or more decision trees;

[0009] Acquire a first training data set, wherein the first training data set includes one or more of first flow characteristic data, first airway pressure characteristic data, and first volume characteristic data of mechanical ventilation;

[0010] Inputting the first training data set into the GBDT model, performing training based on the one or more decision trees, creating a new decision tree and updating the one or more decision trees, obtaining a trained GBDT model, and outputting a first leaf node index value set based on the trained GBDT model;

[0011] Inputting the first leaf node index value set into the LR model for training to obtain a trained LR model;

[0012] Based on the trained GBDT model and the trained LR model, the to-be-detected data set corresponding to the ventilator is processed to obtain a detection result of human-machine asynchrony.

[0013] In the embodiment of the present application, a GBDT model and an LR model are first constructed, and then the training data (i.e., the first training data set) is used to train the GBDT model and output the processing result (i.e., the first leaf node index value set), and then the processing result is used to train the LR model to obtain the trained GBDT model and LR model; finally, the trained GBDT model and LR model are used to process the data to be detected, so as to obtain the detection result of human-machine asynchrony. The embodiment of the present application aims at the problem of human-machine asynchrony discrimination in the prior art by relying on the method of manually annotating features, and realizes automatic extraction and processing of feature information in the original breathing signal of the ventilator during mechanical ventilation through the GBDT model, and uses the more expressive features obtained after processing as the input of the LR model, and uses the linearized relationship to complete the detection and classification of human-machine asynchrony types. Among them, because the GBDT model and the LR model are easy to deploy and have fast computing speed, the cost is low when actually deployed; and because the GBDT model and the LR model are also fast and efficient, they can instantly match the ventilation mode suitable for the patient's pathological condition. In addition, because the LR model is also highly interpretable, the solution of extracting and processing features based on the GBDT model, and then inputting the processed features into the LR model to complete the human-machine asynchronous classification and detection has strong interpretability. Therefore, compared with the method of relying on the manual feature annotation method to distinguish human-machine asynchronous in the prior art, the embodiment of the present invention can reduce costs in actual deployment and improve the real-time and interpretability of human-machine asynchronous detection.

[0014] In a possible implementation, the processing of the to-be-detected data set corresponding to the ventilator based on the trained GBDT model and the trained LR model to obtain a detection result of human-machine asynchrony includes:

[0015] Acquire a first test data set, wherein the first test data set includes one or more of second flow characteristic data, second airway pressure characteristic data, and second volume characteristic data of mechanical ventilation;

[0016] Inputting the first test data set into the trained GBDT model for processing to obtain a second leaf node index value set;

[0017] Inputting the second leaf node index value set into the trained LR model for processing until the accuracy of the processing result based on the trained LR model reaches a preset threshold, thereby obtaining a trained GBDT+LR model;

[0018] The data set to be tested corresponding to the ventilator is input into the trained GBDT+LR model for processing to obtain the detection result of human-machine asynchrony.

[0019] In the embodiment of the present application, before using the trained GBDT model and LR model to process the data to be detected, the performance of the trained GBDT model and LR model can be tested using test data. Among them, the test data (i.e., the first test data set) can be first input into the trained GBDT model to obtain the processing result (i.e., the second leaf node index value set), and then the processing result can be input into the trained LR model to obtain the test result; when the accuracy of the test result reaches the threshold, it is considered that the GBDT+LR model has been trained; finally, the data to be detected is input into the trained GBDT+LR model for processing, and the detection result of human-computer asynchrony can be obtained. In summary, before the actual detection, the embodiment of the present application first performs a performance test on the trained GBDT model and LR model through test data to ensure the accuracy of the model.

[0020] In a possible implementation manner, obtaining a first test data set includes:

[0021] Acquire an original test data set, wherein the original test data set includes one or more of original test flow characteristic data, original test airway pressure characteristic data, and original test volume characteristic data;

[0022] The original test data set is subjected to a first data preprocessing to obtain the first test data set; the first data preprocessing includes one or more of data labeling, data standardization, and zero padding alignment.

[0023] In the embodiment of the present application, the original test data is first obtained, and then the original test data is preprocessed to obtain the input data of the trained GBDT model (i.e., the first test data set). Before inputting the data into the model, the data specifications are first unified through data preprocessing, so that the data with unified specifications can be more conveniently input into the GBDT model for testing, thereby improving the operation speed during the model training process and further improving the timeliness of the solution.

[0024] In a possible implementation, inputting the first leaf node index value set into the LR model for training to obtain a trained LR model includes:

[0025] One Hot encoding is performed on the first leaf node index value set to obtain a second training data set;

[0026] The second training data set is input into the LR model for training to obtain a trained LR model.

[0027] In the embodiment of the present application, the processing result (i.e., the first leaf node index value set) output after the GBDT model training is input into the LR model for training, and the processing result is first processed by One Hot encoding. Using the data processed by One Hot encoding as the training input data of the LR model can solve the problem that the LR model is not good at processing discrete data, and to a certain extent, it also expands the effective features, thereby improving the accuracy and interpretability of the LR model analysis.

[0028] In a possible implementation, inputting the second leaf node index value set into the trained LR model for processing includes:

[0029] Performing One Hot encoding processing on the second leaf node index value set to obtain a second test data set;

[0030] The second test data set is input into the trained LR model for processing.

[0031] In the embodiment of the present application, the trained GBDT model is tested and the processing result (i.e., the second leaf node index value set) is output. Before the processing result is input into the trained LR model for testing, the processing result is first processed by One Hot encoding. Using the data processed by One Hot encoding as the test input data of the trained LR model can solve the problem that the LR model is not good at processing discrete data, and to a certain extent, it also expands the effective features, thereby improving the accuracy and interpretability of the LR model analysis.

[0032] In a possible implementation, an original training data set is obtained, where the original training data set includes one or more of original training flow characteristic data, original training airway pressure characteristic data, and original training volume characteristic data;

[0033] A second data preprocessing is performed on the original training data set to obtain the first training data set, where the first data preprocessing includes one or more of data labeling, data standardization, and zero padding alignment.

[0034] In the embodiment of the present application, the original training data is first obtained, and then the original training data is preprocessed to obtain the input data of the GBDT model (i.e., the first training data set). Before inputting the data into the model, the data specifications are first unified through data preprocessing, so that the data with unified specifications can be more conveniently input into the GBDT model for training, thereby improving the operation speed during the model training process and further improving the timeliness of the solution.

[0035] In a possible implementation, constructing the GBDT model and the LR model includes:

[0036] Setting training parameters of the GBDT model to obtain the GBDT model, wherein the training parameters of the GBDT model include a maximum number of iterations of a base learner of the GBDT model, a maximum depth of a CART tree of the GBDT model, a loss function of the GBDT model, and a learning rate of the GBDT model;

[0037] The training parameters of the LR model are set to obtain the LR model, wherein the training parameters of the LR model include the number of classifications and the weight vector of the LR model.

[0038] In the embodiment of the present application, when constructing the GBDT model and the LR model, their various parameters are first determined, and then the construction of the initial model is completed. The embodiment of the present application aims at the problem of human-machine asynchronous discrimination by relying on the method of manually annotating features in the prior art, and constructs the initial model of the GBDT model and the LR model, providing a prerequisite for subsequent training, testing and actual detection processing.

[0039] In a possible implementation, the detection result of human-machine asynchrony includes a human-machine asynchrony type, and the human-machine asynchrony type includes one or more of a double trigger type, an ineffective inspiratory effort type, and a normal type.

[0040] In the embodiment of the present application, the detection result of human-machine asynchrony is first clarified to determine the specific type of human-machine asynchrony, and the human-machine asynchrony type that can be determined includes one or more of a double trigger type, an ineffective inspiratory effort type, and a normal type. When the ventilator and the mechanical ventilation of the patient or user are in one of the above human-machine asynchrony types, the embodiment of the present application can quickly and accurately detect and identify.

[0041] In a possible implementation, the human-machine asynchronous type includes the double trigger type, the ineffective inspiratory effort type and the normal type;

[0042] The method of obtaining the detection result of human-machine asynchrony includes:

[0043] Obtaining three groups of probability values ​​corresponding to the double trigger type, the ineffective inspiratory effort type and the normal type;

[0044] The human-machine asynchrony type of the ventilator is determined according to the three groups of probability values ​​and preset conditions.

[0045] In an embodiment of the present application, when using the above-mentioned GBDT model and LR model to determine the test results of the data to be tested, three groups of probability values ​​corresponding to the double trigger type, the invalid inspiratory effort type and the normal type are first obtained, rather than directly determining and outputting a certain type as the final test result. This allows medical staff to obtain more data references, and medical staff can further determine the specific type of human-machine asynchrony based on the patient's actual symptoms and data references, thereby avoiding misjudgment due to the existence of extreme values ​​or critical values.

[0046] In a possible implementation, the method further includes:

[0047] When the human-machine asynchronous type of the ventilator is a double trigger type or an invalid inspiratory effort type, a prompt message is sent, and the prompt message is used to prompt medical personnel to adjust the settings of the ventilator according to the human-machine asynchronous type of the ventilator.

[0048] In an embodiment of the present application, when the detection result of the data to be detected is determined to be a double trigger type or an invalid inspiratory effort type, a prompt message is sent to the medical staff to remind the medical staff that the human-machine asynchronous type of the ventilator is abnormal and the settings of the ventilator need to be adjusted to restore it to a normal type.

[0049] In a second aspect, an embodiment of the present application provides a human-machine asynchronous detection device for mechanical ventilation of a ventilator, the device comprising:

[0050] A construction module, used to construct a GBDT model and an LR model; the GBDT model includes one or more tree sets, each tree set includes one or more decision trees;

[0051] An acquisition module, configured to acquire a first training data set, wherein the first training data set includes one or more of first flow characteristic data, first airway pressure characteristic data, and first volume characteristic data of mechanical ventilation;

[0052] A first training module, used to input the first training data set into the GBDT model, perform training based on the one or more decision trees, create a new decision tree and update the one or more decision trees, obtain a trained GBDT model, and output a first leaf node index value set based on the trained GBDT model;

[0053] A second training module, used for inputting the first leaf node index value set into the LR model for training to obtain a trained LR model;

[0054] A determination module is used to process the data set to be detected corresponding to the ventilator based on the trained GBDT model and the trained LR model to obtain a detection result of human-machine asynchrony.

[0055] In the embodiment of the present application, the GBDT model and the LR model are first constructed by the construction module, and then the training data (i.e., the first training data set) is obtained by the acquisition module, and then the GBDT model is trained based on the training data by the first training module and the processing result (i.e., the first leaf node index value set) is output, and then the LR model is trained based on the processing result by the second training module to obtain the trained GBDT model and LR model; finally, the determination module uses the trained GBDT model and LR model to process the data to be detected, so as to obtain the detection result of human-machine asynchrony. The embodiment of the present application aims at the problem of human-machine asynchrony discrimination by relying on the method of manually annotating features in the prior art, and realizes automatic extraction and processing of feature information in the original breathing signal of the ventilator during mechanical ventilation through the GBDT model, and uses the more expressive features obtained after processing as the input of the LR model, and uses the linearized relationship to complete the detection and classification of human-machine asynchrony types. In addition, the GBDT model has the characteristics of easy deployment and fast operation, and the LR model also has the characteristics of simple model and strong interpretability. Therefore, compared with the prior art method of relying on manual feature annotation to distinguish human-machine asynchrony, the embodiments of the present invention can reduce costs and improve the real-time and interpretability of human-machine asynchrony detection in actual deployment.

[0056] In a possible implementation manner, the determining module is specifically configured to:

[0057] Acquire a first test data set, wherein the first test data set includes one or more of second flow characteristic data, second airway pressure characteristic data, and second volume characteristic data of mechanical ventilation;

[0058] Inputting the first test data set into the trained GBDT model for processing to obtain a second leaf node index value set;

[0059] Inputting the second leaf node index value set into the trained LR model for processing until the detection result accuracy based on the trained LR model reaches a preset threshold, thereby obtaining a trained GBDT+LR model;

[0060] The data set to be tested corresponding to the ventilator is input into the trained GBDT+LR model for processing to obtain the detection result of human-machine asynchrony.

[0061] In a possible implementation manner, the determining module is specifically configured to:

[0062] Acquire an original test data set, wherein the original test data set includes one or more of original test flow characteristic data, original test airway pressure characteristic data, and original test volume characteristic data;

[0063] A second data preprocessing is performed on the original test data set to obtain the first test data set; the second data preprocessing includes one or more of data labeling, data standardization, and zero padding alignment.

[0064] In a possible implementation manner, the second training module is specifically used to:

[0065] One Hot encoding is performed on the first leaf node index value set to obtain a second training data set;

[0066] The second training data set is input into the LR model for training to obtain a trained LR model.

[0067] In a possible implementation manner, the determining module is specifically configured to:

[0068] Performing One Hot encoding processing on the second leaf node index value set to obtain a second test data set;

[0069] The second test data set is input into the trained LR model for processing.

[0070] In a possible implementation manner, the acquisition module is specifically configured to:

[0071] Acquire an original training data set, wherein the original training data set includes one or more of original training flow characteristic data, original training airway pressure characteristic data, and original training volume characteristic data;

[0072] The original training data set is subjected to a first data preprocessing to obtain the first training data set, wherein the first data preprocessing includes one or more of data labeling, data standardization, and zero padding alignment.

[0073] In a possible implementation manner, the construction module is specifically used to:

[0074] Setting training parameters of the GBDT model to obtain the GBDT model, wherein the training parameters of the GBDT model include a maximum number of iterations of a base learner of the GBDT model, a maximum depth of a CART tree of the GBDT model, a loss function of the GBDT model, and a learning rate of the GBDT model;

[0075] The training parameters of the LR model are set to obtain the LR model, wherein the training parameters of the LR model include the number of classifications and the weight vector of the LR model.

[0076] In a possible implementation, the detection result of human-machine asynchrony includes a human-machine asynchrony type, and the human-machine asynchrony type includes one or more of a double trigger type, an ineffective inspiratory effort type, and a normal type.

[0077] In a possible implementation manner, the human-machine asynchronous type includes the double trigger type, the ineffective inspiratory effort type and the normal type;

[0078] The determination module is specifically used for:

[0079] Obtaining three groups of probability values ​​corresponding to the double trigger type, the ineffective inspiratory effort type and the normal type;

[0080] The human-machine asynchrony type of the ventilator is determined according to the three groups of probability values ​​and preset conditions.

[0081] In a possible implementation manner, the device further includes:

[0082] The sending module is used to send a prompt message when the human-machine asynchronous type of the ventilator is a double trigger type or an invalid inspiratory effort type, wherein the prompt message is used to prompt medical staff to adjust the settings of the ventilator according to the human-machine asynchronous type of the ventilator.

[0083] In a third aspect, an embodiment of the present application provides a terminal, characterized in that it includes a processor, an input device, an output device and a memory, and the processor, input device, output device and memory are interconnected, wherein the memory is used to store a computer program, and the computer program includes program instructions, and the processor is configured to call the program instructions to execute the human-machine asynchronous detection method for mechanical ventilation of the ventilator according to the first aspect above.

[0084] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, characterized in that the computer storage medium stores a computer program, and the computer program includes program instructions, which, when executed by a processor, enable the processor to execute the human-machine asynchronous detection method for mechanical ventilation of a ventilator according to the first aspect above.

[0085] In a fifth aspect, an embodiment of the present application provides a ventilator, characterized in that it includes a main unit, an air-oxygen mixer, an air source, a humidifier, and an external pipeline, wherein the main unit, the air-oxygen mixer, the air source, the humidifier, and the external pipeline are interconnected, wherein the main unit is used to store a computer program, the computer program includes program instructions, and the host is configured to call the program instructions to execute the human-machine asynchronous detection method for mechanical ventilation of the ventilator according to the first aspect above. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0087] Figure 1 It is a structural schematic diagram of a ventilator 100 provided in an embodiment of the present application;

[0088] Figure 2a It is a flow chart of a method for detecting human-machine asynchrony during mechanical ventilation of a ventilator provided in an embodiment of the present application;

[0089] Figure 2b This is a schematic diagram of a GBDT+LR model processing flow provided in an embodiment of the present application;

[0090] Figure 3a This is a schematic diagram of a data preprocessing process provided by an embodiment of the present application;

[0091] Figure 3b It is a schematic diagram of the structure of a GBDT+LR model provided in an embodiment of the present application;

[0092] Figure 4a This is a confusion matrix diagram of test results based on the GBDT+LR model provided in an embodiment of the present application;

[0093] Figure 4b It is a visual structure diagram of the GBDT model training process provided in an embodiment of the present application;

[0094] Figure 4c This is a feature importance distribution diagram of features extracted by a GBDT model provided in an embodiment of the present application;

[0095] Figure 4d It is a Loss parameter diagram of a GBDT model training situation provided in an embodiment of the present application;

[0096] Figure 4e It is a schematic diagram of ROC curve and AUC value parameters of a GBDT+LR model provided in an embodiment of the present application;

[0097] Figure 5 This is a structural schematic diagram of a human-machine asynchronous detection device for mechanical ventilation of a ventilator provided in an embodiment of the present application;

[0098] Figure 6 It is a structural schematic diagram of another human-machine asynchronous detection device for mechanical ventilation of a ventilator provided in an embodiment of the present application. DETAILED DESCRIPTION

[0099] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0100] It should be understood that the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device comprising a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products or devices.

[0101] It should also be understood that reference to an "embodiment" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiment may be included in at least one embodiment of the present application. The presentation of the phrase at various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0102] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0103] First, some terms in this application are explained to facilitate understanding by those skilled in the art.

[0104] (1) Ineffective inspiratory effort is a phenomenon in which the inspiratory effort made by the patient or user when inhaling is not captured by the ventilator, resulting in the ventilator not providing ventilation support. Ineffective inspiratory effort is a common type of human-machine asynchrony.

[0105] (2) Double triggering is a phenomenon in which the patient's inspiratory action is captured twice by the ventilator in a short period of time, thus triggering the ventilator to provide ventilation support twice. Double triggering is characterized by two consecutive inhalations without time for exhalation. Generally speaking, the patient's or user's inspiratory effort exists throughout the entire process of double-triggered breathing, indicating that the patient's or user's continuous inspiratory effort in the initial stage of the inspiratory cycle and the expiratory cycle will trigger the second breath. Double triggering is a common type of human-machine asynchrony.

[0106] (3) Gradient Boosting Decision Tree (GBDT) model is a decision tree model based on ensemble thinking. Its essence is to learn based on residuals, and it has the characteristics of being able to process various types of data, having high accuracy, and being robust to outliers. The embodiment of the present application will utilize the GBDT model in the process of human-machine asynchronous detection.

[0107] (4) Logistic Regression (LR) model is a classification model formed by applying a logistic function on the basis of linear regression. It is simple, parallelizable and highly interpretable. The highly interpretable feature is particularly important in the field of algorithm models for medical devices. The embodiment of the present application will utilize the LR model in the process of human-machine asynchronous detection.

[0108] (5) One-hot encoding, also known as one-bit effective encoding, is a method of using an N-bit state register to encode N states, each state has its own independent register bit, and at any time, only one of them is effective. The embodiment of the present application will use One-hot encoding in the process of human-machine asynchronous detection.

[0109] First, the technical problem to be solved by this application is analyzed and proposed. In the prior art, the scheme for detecting human-machine asynchrony during mechanical ventilation of a ventilator includes scheme 1, which may specifically include the following steps S100 to S102:

[0110] Step S100: Domain experts and professional medical staff manually collect ventilator parameter features.

[0111] Step S101: Domain experts and professional medical staff manually label and analyze ventilator parameter features.

[0112] Step S102: Determine whether human-computer asynchronous phenomenon exists according to the analysis result.

[0113] This solution has the following disadvantages:

[0114] Disadvantage 1: High cost. The method of asynchronous human-machine discrimination based on manually labeled features consumes the valuable time of experts and professional medical staff in the field, and the financial cost is also not small.

[0115] Disadvantage 2: Low timeliness. The solution relies on manual feature collection, labeling, and analysis, which takes a lot of time and is difficult to meet the high timeliness requirements of medical devices such as ventilators. Therefore, solutions that rely on manual identification are also difficult to promote.

[0116] Disadvantage 3: The feature is not very interpretable. The features extracted manually have yet to be verified in terms of scalability, generalization, and personalization. Therefore, the interpretability of using the features for human-machine asynchronous discrimination is still not strong, and medical devices such as ventilators pay particular attention to the interpretability of the solution.

[0117] In order to solve the problems of high cost, low timeliness and weak feature interpretation of the current manual human-machine asynchronous detection technology, and to achieve the purpose of reducing costs, improving timeliness and interpretation, taking into account the shortcomings of the existing technology, the technical problems actually to be solved by this application are as follows:

[0118] 1. Use an automatic detection model that is simple to deploy, fast in operation, and highly interpretable (Disadvantages 1, 2, and 3 of Solution 1). The manual-based asynchronous detection technology partially meets the detection needs, but cannot meet the more stringent requirements for cost, timeliness, and feature interpretability in the actual deployment process. Therefore, an automatic detection model is needed to replace the manual method. The model should be simple and easy to deploy, thereby reducing the deployment cost; the model should also have the characteristics of fast operation, so as to solve the problem of low timeliness; finally, the model also needs to have strong interpretability to meet the interpretability requirements of medical devices such as ventilators.

[0119] In summary, the existing human-machine asynchrony detection technology based on manual methods has the problems of high cost, low timeliness and weak feature interpretation, which makes it impossible to meet the higher requirements of actual deployment detection. Therefore, the human-machine asynchrony detection method for mechanical ventilation of a ventilator provided in this application is based on the GBDT model and the LR model for human-machine asynchrony detection, which has the characteristics of simple and easy deployment, fast calculation and strong interpretation, and can solve the above technical problems.

[0120] In order to facilitate a better understanding of the present application, the following embodiment of the present application is briefly described through a ventilator 100, see Figure 1 , Figure 1 It is a structural schematic diagram of a ventilator 100 provided in an embodiment of the present application.

[0121] The ventilator 100 may include: a host 110, an air-oxygen mixer 120, an air source 130, a humidifier 140, an external pipe 150, etc. It is understood that the structure illustrated in the embodiment of the present invention does not constitute a specific limitation on the terminal device 100. In other embodiments of the present application, the terminal device 100 may include more or fewer components than shown in the figure, or combine certain components, or split certain components, or arrange the components differently. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0122] The host 110 may include one or more processing units, for example, the host 110 may include an application processor (AP), a modem processor, a graphics processor (GPU), an image signal processor (ISP), a controller, a memory, a digital signal processor (DSP), and / or a neural-network processing unit (NPU), etc. Different processing units may be independent devices or integrated into one or more processors.

[0123] The controller may be the nerve center and command center of the ventilator 100. The controller may generate an operation control signal according to the instruction operation code and the timing signal to complete the control of fetching and executing instructions.

[0124] The host 110 may also be provided with a memory for storing instructions and data. In some embodiments, the memory in the host 110 is a cache memory. The memory may store instructions or data that the host 110 has just used or cyclically used. If the host 110 needs to use the instruction or data again, it may be directly called from the memory. This avoids repeated access, reduces the waiting time of the host 110, and thus improves the efficiency of the system.

[0125] In some embodiments, the host 110 may include one or more interfaces. The interface may include an inter-integrated circuit (I2C) interface, an inter-integrated circuit sound (I2S) interface, a pulse code modulation (PCM) interface, a universal asynchronous receiver / transmitter (UART) interface, a mobile industry processor interface (MIPI), a general-purpose input / output (GPIO) interface, a subscriber identity module (SIM) interface, and / or a universal serial bus (USB) interface, etc. The host 110 may be connected to other components of the ventilator 100 or other external devices through one or more interfaces.

[0126] It is understandable that the interface connection relationship between the modules illustrated in the embodiment of the present invention is only a schematic illustration and does not constitute a structural limitation on the ventilator 100. In other embodiments of the present application, the terminal device 100 may also adopt different interface connection methods in the above embodiments, or a combination of multiple interface connection methods.

[0127] The input gas of the air-oxygen mixer 120 can be compressed air, ambient air or compressed oxygen. Oxygen accounts for 21% of the air, so compressed air and ambient air can also be used as the gas source 130 of the ventilator 100 to supply patients with breathing difficulties for treatment. For patients with impaired ventilation function, the oxygen concentration must be appropriately increased to meet the treatment needs. At this time, the air-oxygen mixer 120 can be connected to a compressed oxygen source and appropriately adjust the oxygen concentration according to the changes in the condition.

[0128] The air source 130 is generally divided into two types: electric air supply and compressed air source. If the ventilator 100 is powered by an electric motor and generates a certain positive pressure airflow through a compression pump or a folding leather bag to supply air to the patient, it is called an electric mode; if the ventilator 100 uses a compressed air pump and supplies air to the patient after filtering, decompression, humidification, etc., it is called a pneumatic mode.

[0129] The humidifier 140 mainly replaces the nasal cavity and the oral cavity to humidify the inhaled gas. The device that the ventilator 100 provides gas to the patient for humidification is called a humidifier; and the device that atomizes the liquid is called a nebulizer. Gas humidification and liquid atomization can protect the patient's trachea and bronchial mucosa. Usually, the humidifier 140 also has a heating device, that is, heating humidification or heating atomization, so that the gas is close to the human body temperature and reduces the stimulation to the patient.

[0130] The external pipe 150 provides the humidified or atomized gas to the patient, and directly discharges the gas exhaled by the patient through the respiratory valve. The external pipe 150 also feeds back the respiratory signal to the host 110 to achieve synchronous breathing and effectively improve the patient's ventilation efficiency.

[0131] based on Figure 1 The provided schematic diagram of the ventilator architecture is combined with the process of the human-machine asynchronous detection method for mechanical ventilation of the ventilator provided in this application to explain the technical problems raised in this application in detail.

[0132] See also Figure 2a and Figure 2b , Figure 2a is a flow chart of a method for detecting human-machine asynchrony during mechanical ventilation of a ventilator provided in an embodiment of the present application, Figure 2bThis is a schematic diagram of a GBDT+LR model processing flow provided by an embodiment of the present application. The human-machine asynchronous detection method can be applied to the above Figure 1 In the ventilator 100, the ventilator 100 can be used to support and perform Figure 2a The method flow steps S200 to S203 shown in FIG. 1 can refer to the processing flow Figure 2b The following will be combined with the attached Figure 2a The description is made from the perspective of the ventilator. The human-machine asynchrony detection method may include the following steps S200 to S203.

[0133] Step S200: construct a GBDT model and an LR model.

[0134] Specifically, the GBDT and LR models may be first constructed in the host 110 of the ventilator 100. The GBDT model may include one or more tree sets, and each tree set may include one or more decision trees.

[0135] Preferably, when constructing a GBDT model in the host 110 of the ventilator 100, the number of categories in the tree set, the maximum number of iterations of the base learner, the maximum depth of the CART tree, the loss function, the learning rate and other parameters can be set first to complete the GBDT model. For example, the number of categories in the tree set can be 3, the maximum number of iterations of the base learner can be 60, the maximum depth of the CART tree can be 3, the loss function can use the log-likelihood loss function, and the learning rate can be 0.1. When constructing the LR model in the host 110 of the ventilator 100, the number of categories and the weight vector of the LR model can be set first. For example, the number of categories can be 3. It can be understood that the above parameters can also be other values, and the embodiments of the present invention do not specifically limit this. It should be noted that the values ​​of the number of categories in the tree set in the GBDT model and the number of categories in the LR model should correspond to the actual number of categories. See Figure 3b , Figure 3b It is a schematic diagram of the structure of a GBDT+LR model provided in an embodiment of the present application. The schematic diagram shows that the upper half of the entire GBDT+LR model is the structure diagram of the GBDT model, where the value of N is 60, that is, the GBDT model has 60 trees; the lower half is the structure diagram of the LR model. In the embodiment of the present application, the output of the LR model can be used to determine the type of human-computer asynchrony.

[0136] Step S201: Acquire a first training data set.

[0137] Specifically, the first training data set can be the original historical data set about mechanical ventilation obtained by the ventilator 100 from the external pipe 150 through the host 110. Among them, the original historical data set may include one or more of flow characteristic data, airway pressure characteristic data, and volume characteristic data. For ease of understanding, the embodiment of the present application is explained by taking the original historical data set including flow characteristic data, airway pressure characteristic data and volume characteristic data as an example. It should be noted that the first training data set can be obtained before constructing the GBDT model and the LR model, or it can be obtained after constructing the GBDT model and the LR model, and no specific limitation is made here.

[0138] Preferably, the first training data set can be obtained by first dividing the original historical data set about mechanical ventilation obtained by the ventilator 100 from the external pipeline 150 by the host 110, and then performing data preprocessing. The original historical data set may include one or more of flow characteristic data, airway pressure characteristic data, and volume characteristic data; the original historical data set can be divided into a first training data set and a first test data set; data preprocessing may include one or more of data labeling, data standardization, and zero padding alignment. See Figure 3a , Figure 3a It is a data preprocessing flow diagram provided by the embodiment of the present application, which can be used to set up, debug and simulate equipment or devices, and then divide the collected raw data into training data and test data, and finally perform data annotation, data standardization and zero padding alignment on them. In the embodiment of the present application, the historical data set includes flow characteristic data, airway pressure characteristic data and volume characteristic data, and the data preprocessing can include data annotation, data standardization and zero padding alignment as an example. The historical data set can be firstly processed by data annotation, and then the data after data annotation processing is processed by data standardization, and finally the data after data standardization processing is processed by zero padding alignment. Among them, the historical data set can be divided into three categories of human-machine asynchrony: double triggering, invalid inspiratory effort and normal through data annotation; data standardization can be a linear method (such as extreme value method, standard deviation method), a broken line method (such as three broken line method), a curve method (such as half-normal distribution), etc., which is not specifically limited in the embodiment of the present application; zero padding alignment can be processed according to the actual data length, truncated if too long, supplemented if too short, to ensure the uniformity of data size.

[0139] Specifically, it can be assumed that the historical data set after data annotation processing includes the traffic feature data sequence F x =(f1,f2,…,f x ), airway pressure characteristic data sequence P x =(p1,p2,…,p x ), capacity characteristic data sequence V x=(v1,v2,…,v x ).

[0140] Then, we can x , P x 、V x Perform data standardization. The data standardization formula can be Among them, N represents one of F, P, and V, μ is the mean of the corresponding sequence, and σ is the standard deviation of the corresponding sequence.

[0141] Finally, we can assume a sufficient data length. If it is too long, it will be truncated, and if it is too short, it will be supplemented. For example, 100 can be selected as the longest data length. The length formula can be maxLen = max(len(P1, P2, ..., P L )), the subscript L = len(trainset) + len(testset) is used for calculation. That is to say, the length of the first training data set and the first test data set is 100. For all data with a length less than 100, "0" is added to form data of uniform size (size [-1,100,3]) for training and testing. Then, the matrix of size [-1,100,3] is flattened to obtain a matrix of size [-1,300]. Finally, the training data and test data matrices that meet the specification requirements of [-1,300] can be easily input into the GBDT+LR model for training and testing. Let Z x is a row element in the [-1,300] matrix, then it is obvious that Z x =[F x , P x 、V x ]=(f1,f2,…,f x , p1, p2, …, p x ,v1,v2,…,v x ).

[0142] Step S202: input the first training data set into the GBDT model for training, and output a first leaf node index value set.

[0143] Specifically, training is performed based on one or more decision trees in the constructed GBDT model, so as to create a new decision tree and update the above one or more decision trees. During the training process, the thresholds of the above various parameters of the model are reached to obtain the trained GBDT model, and the first leaf node index value set is output based on the trained GBDT model. As in the above example, training is performed based on one or more decision trees in the three tree sets to create a new decision tree and update the original one or more decision trees. When the number of decision trees in each tree set reaches 60, the training is suspended to obtain the trained GBDT model, and the index values ​​of all leaf nodes in the model are output based on the trained model. Among them, for the three different categories of human-computer asynchrony (i.e., double triggering, ineffective inspiratory effort, and normal) discrimination problems, three different tree sets are constructed to fit the probability of each category. Assuming that q is a certain type of human-computer asynchrony, use The probability of human-machine asynchronous category q can be calculated; in addition, the formula of the GBDT model obtained after training can be Where M represents the number of decision trees, R mj represents the set of leaf node index values ​​in all partitioning methods of the mth subtree, and I represents the indicator function. It should be noted that all leaf node index values ​​output by the trained GBDT model are features with more expressive power than the feature data of the training dataset.

[0144] Step S203: input the first leaf node index value set into the LR model for training to obtain a trained LR model.

[0145] Specifically, the first leaf node index value set output by the trained GBDT model is input into the constructed LR model for training, so as to obtain the trained LR model.

[0146] Preferably, the first leaf node index value set output by the trained GBDT model can be processed by OneHot to obtain a second training data set, and then the second training data set is input into the constructed LR model for training to obtain a trained LR model. The formula of the LR model obtained after training can be The value of K can be 3, which is the number of human-computer asynchronous classifications, n is the number of samples, and w is the k are the components of the weight vector.

[0147] Step S204: Based on the trained GBDT model and the trained LR model, the to-be-detected data set corresponding to the ventilator is processed to obtain a detection result of human-machine asynchrony.

[0148] Specifically, the data set to be detected of the ventilator 100 can be input into the GBDT model obtained after training for processing to obtain data features with stronger expression ability, and then the data features are input into the LR model obtained after training for processing to determine which of the detection results of human-machine asynchrony is double triggering, invalid inspiratory effort and normal. Among them, the trained LR model can output the probability values ​​of three types of human-machine asynchrony, namely double triggering, invalid inspiratory effort and normal, respectively, and the one with the largest probability value can be determined as the detection result of human-machine asynchrony. For example, the probability value of double triggering is 10%, the probability value of invalid inspiratory effort is 85%, and the probability value of normal is 5%, then the detection result of human-machine asynchrony is invalid inspiratory effort. It is understandable that the above probability value can also be other values ​​in the specific analysis process, and the embodiment of the present invention only makes an exemplary explanation for this, and does not make specific limitations. It should also be noted that medical staff can combine the actual symptoms of the patient and the probability value data of the human-machine asynchrony type as a reference to further determine the specific type of human-machine asynchrony, so as to avoid misjudgment caused by the existence of extreme values ​​or critical values.

[0149] Preferably, before the data set to be tested of the ventilator 100 is input into the trained GBDT model and LR model, the performance test of the trained GBDT model and LR model can be performed using the test data. A threshold can be set, for example, the threshold can be 95%. If the accuracy of the test does not meet the threshold requirement, it is necessary to obtain additional training data to train the model; if the accuracy of the test meets the threshold requirement, it is considered that the GBDT model and the LR model have been trained, and no additional training data can be temporarily obtained to train the model. It should be noted that the data set used for performance testing can be obtained when the original historical data set is divided into data (such as the first test data set in S201), and the data set used for performance testing can also be pre-processed. The specific processing method can refer to the content in S201, which will not be described in detail here. It should also be noted that when the first test data set is input into the trained GBDT model for processing, the feature data with stronger expression ability (i.e., the second leaf node index value set) is output, and before the feature data with stronger expression ability is input into the trained LR model, One Hot encoding processing can be performed first.

[0150] Optionally, the above step S204 further includes step S205: sending a prompt message.

[0151] Specifically, when the human-machine asynchronous type of the ventilator 100 is determined to be any of the double trigger type or the invalid inspiratory effort type, the ventilator 100 can send a prompt message through the host 110, and the prompt message is used to prompt the medical staff to adjust the ventilator according to the human-machine asynchronous type of the ventilator 100 (i.e., the double trigger type or the invalid inspiratory effort type) to restore it to the normal type. For example, the ventilator 100 can be connected to an external prompt device using an interface in the host 110. When the human-machine asynchronous type of the ventilator 100 is the double trigger type or the invalid inspiratory effort type, the host 110 sends a prompt message to the external prompt device, and then the external prompt device emits a prompt sound. It is understandable that the prompt method can also be other methods, and the embodiments of the present invention are only explained as an example and are not specifically limited.

[0152] Understandably, when the above-mentioned human-machine asynchronous detection method is applied to the human-machine asynchronous detection of ventilators, the time and financial cost of judging and deploying human-machine asynchronous detection can be reduced due to the adoption of the GBDT+LR model, which is simple, easy to deploy, fast in operation and highly interpretable. The fast operation also ensures the high timeliness of the detection. In addition, the model is highly interpretable, which is in line with the high requirements of the medical field for the interpretability of the solution.

[0153] The specific effect of the human-computer asynchronous detection in the embodiment of the present application is shown below with the experimental process and results of the embodiment of the present application. It should be noted that the original historical data used for the experiment in the embodiment of the present application includes flow characteristic data, airway pressure characteristic data and capacity characteristic data. During the experiment, these original historical data are divided with the help of trained technicians, and then the divided data are preprocessed (including data labeling, data standardization and zero padding alignment) to determine the data length to be 100. In addition, the parameters of the constructed GBDT model are that the number of categories in the tree set is 3, the maximum number of iterations of the base learner is 60, the maximum depth of the CART tree is 3, the loss function adopts the log-likelihood loss function, the learning rate is 0.1, and the number of categories of the LR model is 3; the probability formula of a certain human-computer asynchronous type q is The formula of the trained GBDT model is The formula of the trained LR model is The specific training process and model usage effect of the embodiments of the present application can be found in the following figures.

[0154] See also Figure 4a , Figure 4aThis is a confusion matrix diagram of the test results based on the GBDT+LR model provided in the embodiment of the present application. The figure shows that the GBDT+LR model is quite effective in the process of ventilator human-machine asynchronous detection, and the prediction accuracy of the double trigger type, invalid inspiratory effort type and normal type detection is above 95%. Among them, when predicting the double trigger type, a small part (5%) of the double triggers will be misjudged as invalid inspiratory effort types, while the prediction accuracy of invalid inspiratory effort types and normal types is close to 100%.

[0155] See also Figure 4b , Figure 4b This is a visual structure diagram of a GBDT model training process provided by an embodiment of the present application. The structure diagram shows some discrimination threshold parameters of the values ​​in the subtree nodes during the training of the first subtree of the GBDT model.

[0156] See also Figure 4c , Figure 4c It is a feature importance distribution diagram of features extracted by a GBDT model provided in an embodiment of the present application. The distribution diagram shows the importance distribution of the top 20 features for easy visual viewing. The horizontal axis represents the weight of the feature, and the higher the weight, the higher the importance of the feature; the vertical axis is the encoding of the feature, and different encodings correspond to different specific features. The correspondence between the specific encoding and the specific features is not reflected in detail here, only that the importance of different features will be different, and the embodiment of the present application can display the specific situation of the importance of these different features through visualization means.

[0157] See also Figure 4d , Figure 4d This is a Loss parameter diagram of a GBDT model training situation provided in an embodiment of the present application. The diagram shows that as the number of GBDT model training generations increases, the Loss of the GBDT model training is steadily decreasing, that is, the GBDT model is gradually converging to a good training state, and the training of the GBDT model is successful. Therefore, the more expressive features combined by the GBDT model analysis facilitate the final discrimination of the human-computer asynchronous type by the subsequent LR model.

[0158] See also Figure 4e , Figure 4eIt is a schematic diagram of the ROC curve and AUC value parameters of a GBDT+LR model provided in an embodiment of the present application. The schematic diagram shows the macro-average and micro-average ROC curves of the GBDT+LR model and their corresponding AUC values, where the AUC value is used to illustrate the model performance, and the closer the value is to 1, the better the model performance. Among them, the corresponding AUC value of the micro-average ROC curve is 0.9946; the corresponding AUC value of the macro-average ROC curve is 0.9994. The AUC values ​​of the macro-average and micro-average ROC curves of the GBDT+LR model provided in the embodiment of the present application are both quite close to 1, that is to say, the performance of the GBDT+LR model provided in the embodiment of the present application is good.

[0159] In summary, the solution of the present application for ventilator human-machine asynchronous detection through the GBDT+LR model is highly feasible, and can overcome the problems of high cost, low timeliness and poor interpretability of existing manual discrimination methods.

[0160] The method of the embodiment of the present application is described in detail above, and the related device of the embodiment of the present application is provided below.

[0161] See also Figure 5 , Figure 5 1 is a schematic diagram of the structure of a human-machine asynchronous detection device for mechanical ventilation of a ventilator provided in an embodiment of the present application. The human-machine asynchronous detection device 10 may include a construction module 101, an acquisition module 102, a first training module 103, a second training module 104 and a determination module 105, and optionally, a sending module 106. The detailed description of each module is as follows:

[0162] A construction module 101 is used to construct a GBDT model and an LR model; the GBDT model includes one or more tree sets, each tree set includes one or more decision trees;

[0163] An acquisition module 102 is used to acquire a first training data set, where the first training data set includes one or more of first flow characteristic data, first airway pressure characteristic data, and first volume characteristic data of mechanical ventilation;

[0164] A first training module 103 is used to input the first training data set into the GBDT model, perform training based on the one or more decision trees, create a new decision tree and update the one or more decision trees, obtain a trained GBDT model, and output a first leaf node index value set based on the trained GBDT model;

[0165] A second training module 104 is used to input the first leaf node index value set into the LR model for training to obtain a trained LR model;

[0166] The determination module 105 is used to process the to-be-detected data set corresponding to the ventilator based on the trained GBDT model and the trained LR model to obtain a detection result of human-machine asynchrony.

[0167] In a possible implementation, the determining module 105 is specifically configured to:

[0168] Acquire a first test data set, wherein the first test data set includes one or more of second flow characteristic data, second airway pressure characteristic data, and second volume characteristic data of mechanical ventilation;

[0169] Inputting the first test data set into the trained GBDT model for processing to obtain a second leaf node index value set;

[0170] Inputting the second leaf node index value set into the trained LR model for processing until the detection result accuracy based on the trained LR model reaches a preset threshold, thereby obtaining a trained GBDT+LR model;

[0171] The data set to be tested corresponding to the ventilator is input into the trained GBDT+LR model for processing to obtain the detection result of human-machine asynchrony.

[0172] In a possible implementation, the determining module 105 is specifically configured to:

[0173] Acquire an original test data set, wherein the original test data set includes one or more of original test flow characteristic data, original test airway pressure characteristic data, and original test volume characteristic data;

[0174] A second data preprocessing is performed on the original test data set to obtain the first test data set; the second data preprocessing includes one or more of data labeling, data standardization, and zero padding alignment.

[0175] In a possible implementation, the second training module 104 is specifically configured to:

[0176] One Hot encoding is performed on the first leaf node index value set to obtain a second training data set;

[0177] The second training data set is input into the LR model for training to obtain a trained LR model.

[0178] In a possible implementation, the determining module 105 is specifically configured to:

[0179] Performing One Hot encoding processing on the second leaf node index value set to obtain a second test data set;

[0180] The second test data set is input into the trained LR model for processing.

[0181] In a possible implementation, the acquisition module 102 is specifically configured to:

[0182] Acquire an original training data set, wherein the original training data set includes one or more of original training flow characteristic data, original training airway pressure characteristic data, and original training volume characteristic data;

[0183] The original training data set is subjected to a first data preprocessing to obtain the first training data set, wherein the first data preprocessing includes one or more of data labeling, data standardization, and zero padding alignment.

[0184] In a possible implementation, the construction module 101 is specifically configured to:

[0185] Setting training parameters of the GBDT model to obtain the GBDT model, wherein the training parameters of the GBDT model include a maximum number of iterations of a base learner of the GBDT model, a maximum depth of a CART tree of the GBDT model, a loss function of the GBDT model, and a learning rate of the GBDT model;

[0186] The training parameters of the LR model are set to obtain the LR model, wherein the training parameters of the LR model include the number of classifications and the weight vector of the LR model.

[0187] In a possible implementation, the detection result of human-machine asynchrony includes a human-machine asynchrony type, and the human-machine asynchrony type includes one or more of a double trigger type, an ineffective inspiratory effort type, and a normal type.

[0188] In a possible implementation, the human-machine asynchronous type includes the double trigger type, the ineffective inspiratory effort type and the normal type;

[0189] The determination module 105 is specifically used for:

[0190] Obtaining three groups of probability values ​​corresponding to the double trigger type, the ineffective inspiratory effort type and the normal type;

[0191] The human-machine asynchrony type of the ventilator is determined according to the three groups of probability values ​​and preset conditions.

[0192] In a possible implementation, the device further includes:

[0193] The sending module 106 is used to send a prompt message when the human-machine asynchronous type of the ventilator is a double trigger type or an invalid inspiratory effort type, and the prompt message is used to prompt medical staff to adjust the settings of the ventilator according to the human-machine asynchronous type of the ventilator.

[0194] It should be noted that the functions of the various functional units in the human-machine asynchronous detection device 10 described in the embodiment of the present application can be referred to in the above Figure 2a The relevant descriptions of steps S200 to S205 in the method embodiment are not repeated here.

[0195] See also Figure 6 , Figure 6 Schematic diagram of another human-machine asynchronous detection device for mechanical ventilation of a ventilator provided in an embodiment of the present application. Figure 6 As shown, the apparatus 20 may include: one or more processors 601; one or more input devices 602, one or more output devices 603 and a memory 604. The processor 601, the input device 602, the output device 603 and the memory 604 are connected via a bus 605. The memory 604 is used to store a computer program, the computer program includes program instructions, and the processor 601 is used to execute the program instructions stored in the memory 604.

[0196] Among them, the processor 601 is configured to call the program instructions to execute: construct a GBDT model and an LR model; the GBDT model includes one or more tree sets, each tree set includes one or more decision trees; obtain a first training data set, the first training data set includes one or more of the first flow characteristic data, the first airway pressure characteristic data, and the first volume characteristic data of mechanical ventilation; input the first training data set into the GBDT model, train based on the one or more decision trees, create a new decision tree and update the one or more decision trees, obtain the trained GBDT model, and output the first leaf node index value set based on the trained GBDT model; input the first leaf node index value set into the LR model for training to obtain the trained LR model; based on the trained GBDT model and the trained LR model, process the data set to be detected corresponding to the ventilator to obtain the detection result of human-machine asynchrony.

[0197] It should be understood that in the embodiment of the present application, the processor 601 may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0198] The input device 602 may include a touch pad, a microphone, etc., and the output device 603 may include a display (LCD, etc.), a speaker, etc.

[0199] The memory 604 may include a read-only memory and a random access memory, and provide instructions and data to the processor 601. A portion of the memory 604 may also include a non-volatile random access memory. For example, the memory 604 may also store information about the device type.

[0200] In a specific implementation, the processor 601, input device 602, and output device 603 described in the embodiment of the present application can execute the implementation method described in the human-machine asynchronous detection method for mechanical ventilation of the ventilator provided in the embodiment of the present application, and can also execute the implementation method of the human-machine asynchronous detection device for mechanical ventilation of the ventilator described in the embodiment of the present application, which will not be repeated here.

[0201] The present application also provides a computer-readable storage medium storing a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, Figure 2a The human-machine asynchronous detection method of mechanical ventilation of the ventilator is shown in the figure. For details, please refer to Figure 2a The description of the illustrated embodiment will not be repeated here.

[0202] The computer-readable storage medium may be an internal storage unit of the human-machine asynchronous detection device or electronic device described in any of the aforementioned embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium may also be an external storage device of the electronic device, such as a plug-in hard disk, a smart memory card (smartmedia card, SMC), a secure digital (secure digital, SD) card, a flash card (flash card), etc. equipped on the electronic device. Further, the computer-readable storage medium may also include both an internal storage unit of the electronic device and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by the electronic device. The computer-readable storage medium may also be used to temporarily store data that has been output or is to be output.

[0203] The present application also provides a ventilator 100, including a host 110, an air-oxygen mixer 120, an air source 130, a humidifier 140, and an external pipeline 150, see Figure 1 , Figure 1 1 is a schematic diagram of the structure of a ventilator 100 provided in an embodiment of the present application. The host 110, the air-oxygen mixer 120, the gas source 130, the humidifier 140 and the external pipe 150 are connected to each other, wherein the host 110 is used to store a computer program, the computer program includes program instructions, and the host is configured to call the program instructions to execute the above-mentioned human-machine asynchronous detection method for mechanical ventilation of the ventilator.

[0204] It is understandable that the above-mentioned ventilator 100 is only an example in the embodiment of the present application and should not be used as a limitation of the present invention. Other ventilators that only include one or more of the above-mentioned host, air-oxygen mixer, gas source, humidifier and external pipeline, when there are modules or units or components that perform the above-mentioned human-machine asynchronous detection method for mechanical ventilation of the ventilator for human-machine asynchronous detection, should also fall within the scope of protection of the present invention.

[0205] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0206] The present application is described with reference to the flowcharts and / or block diagrams of the methods, apparatuses, and computer program products of the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable diagnostic data processing device to produce a machine, so that instructions executed by the processor of the computer or other programmable diagnostic data processing device generate instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0207] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable diagnostic data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0208] These computer program instructions can also be loaded onto a computer or other programmable diagnostic data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0209] Although the present application has been described in conjunction with specific features and embodiments thereof, it is obvious that various modifications and combinations may be made thereto without departing from the spirit and scope of the present application. Accordingly, this specification and the drawings are merely exemplary illustrations of the present application as defined by the appended claims, and are deemed to have covered any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, those skilled in the art may make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for detecting human-machine asynchrony during mechanical ventilation of a ventilator, characterized in that: Applied to ventilators, including: Constructing a GBDT model and an LR model; the GBDT model includes one or more tree sets, each tree set includes one or more decision trees; Acquire an original historical data set, wherein the original historical data set includes one or more of original flow characteristic data, original airway pressure characteristic data, and original volume characteristic data; The original historical data set is divided and preprocessed to obtain a first training data set and a first test data set; the data preprocessing includes one or more of data labeling, data standardization, zero padding alignment and matrix flattening; the first training data set includes one or more of first flow characteristic data, first airway pressure characteristic data and first capacity characteristic data of mechanical ventilation, and the first test data set includes one or more of second flow characteristic data, second airway pressure characteristic data and second capacity characteristic data of mechanical ventilation; wherein the original historical data set after data labeling includes a flow characteristic data sequence F x =(f1,f2,…,f x ), airway pressure characteristic data sequence P x =(p1,p2,…,p x ), capacity characteristic data sequence V x =(v1,v2,…,v x ); the data standardization is used to x , P x and V x The data standardization formula is: Wherein, N represents one of F, P, and V, μ is the mean of the corresponding sequence, and σ is the standard deviation of the corresponding sequence; the zero-padding alignment is used to control the length of the first training data set and the first test data set, and the length formula of the zero-padding alignment is maxLen=max(len(P1,P2,…,P L )), subscript L = len(trainset) + len(testset); the matrix flattening is used to flatten and reduce the dimension of the data matrix after the zero-padding alignment; Inputting the first training data set into the GBDT model, performing training based on the one or more decision trees, creating a new decision tree and updating the one or more decision trees, obtaining a trained GBDT model, and outputting a first leaf node index value set based on the trained GBDT model; Inputting the first leaf node index value set into the LR model for training to obtain a trained LR model; Inputting the first test data set into the trained GBDT model for processing to obtain a second leaf node index value set; Inputting the second leaf node index value set into the trained LR model for processing until the accuracy of the processing result based on the trained LR model reaches a preset threshold, thereby obtaining a trained GBDT+LR model; Based on the trained GBDT model, feature extraction and processing are performed on the data set to be detected corresponding to the ventilator, and based on the trained LR model, the extracted and processed features are classified and matched to obtain a detection result of human-machine asynchrony.

2. The method according to claim 1, characterized in that The method extracts and processes features of the data set to be detected corresponding to the ventilator based on the trained GBDT model, and classifies and matches the extracted and processed features based on the trained LR model to obtain a detection result of human-machine asynchrony, including: Acquire a first test data set, wherein the first test data set includes one or more of second flow characteristic data, second airway pressure characteristic data, and second volume characteristic data of mechanical ventilation; Inputting the first test data set into the trained GBDT model for processing to obtain a second leaf node index value set; Inputting the second leaf node index value set into the trained LR model for processing until the accuracy of the processing result based on the trained LR model reaches a preset threshold, thereby obtaining a trained GBDT+LR model; The data set to be tested corresponding to the ventilator is input into the trained GBDT+LR model for processing to obtain the detection result of human-machine asynchrony.

3. The method according to claim 2, characterized in that The obtaining of a first test data set includes: Acquire an original test data set, wherein the original test data set includes one or more of original test flow characteristic data, original test airway pressure characteristic data, and original test volume characteristic data; The original test data set is subjected to a first data preprocessing to obtain the first test data set; the first data preprocessing includes one or more of data labeling, data standardization, and zero padding alignment.

4. The method according to any one of claims 1 to 3, characterized in that The step of inputting the first leaf node index value set into the LR model for training to obtain a trained LR model includes: One Hot encoding is performed on the first leaf node index value set to obtain a second training data set; The second training data set is input into the LR model for training to obtain a trained LR model.

5. The method according to any one of claims 2 to 4, characterized in that: The step of inputting the second leaf node index value set into the trained LR model for processing includes: Performing One Hot encoding processing on the second leaf node index value set to obtain a second test data set; The second test data set is input into the trained LR model for processing.

6. The method according to any one of claims 1 to 5, characterized in that The obtaining of a first training data set includes: Acquire an original training data set, wherein the original training data set includes one or more of original training flow characteristic data, original training airway pressure characteristic data, and original training volume characteristic data; A second data preprocessing is performed on the original training data set to obtain the first training data set, where the first data preprocessing includes one or more of data labeling, data standardization, and zero padding alignment.

7. The method according to any one of claims 1 to 6, characterized in that The construction of the GBDT model and the LR model includes: Setting training parameters of the GBDT model to obtain the GBDT model, wherein the training parameters of the GBDT model include a maximum number of iterations of a base learner of the GBDT model, a maximum depth of a CART tree of the GBDT model, a loss function of the GBDT model, and a learning rate of the GBDT model; The training parameters of the LR model are set to obtain the LR model, wherein the training parameters of the LR model include the number of classifications and the weight vector of the LR model.

8. The method according to any one of claims 1 to 7, characterized in that The detection result of human-machine asynchrony includes a human-machine asynchrony type, and the human-machine asynchrony type includes one or more of a double trigger type, an ineffective inspiratory effort type, and a normal type.

9. The method according to claim 8, characterized in that The human-machine asynchronous type includes the double trigger type, the ineffective inspiratory effort type and the normal type; The method of obtaining the detection result of human-machine asynchrony includes: Obtaining three groups of probability values ​​corresponding to the double trigger type, the ineffective inspiratory effort type and the normal type; The human-machine asynchrony type of the ventilator is determined according to the three groups of probability values ​​and preset conditions.

10. The method according to claim 9, characterized in that The method further comprises: When the human-machine asynchronous type of the ventilator is a double trigger type or an invalid inspiratory effort type, a prompt message is sent, and the prompt message is used to prompt medical personnel to adjust the settings of the ventilator according to the human-machine asynchronous type of the ventilator.

11. A human-machine asynchronous detection device for mechanical ventilation of a ventilator, characterized in that: Applied to ventilators, including: A construction module, used to construct a GBDT model and an LR model; the GBDT model includes one or more tree sets, each tree set includes one or more decision trees; An acquisition module, used to acquire an original historical data set, wherein the original historical data set includes one or more of original flow characteristic data, original airway pressure characteristic data, and original volume characteristic data; The original historical data set is divided and preprocessed to obtain a first training data set and a first test data set; the data preprocessing includes one or more of data labeling, data standardization, zero padding alignment and matrix flattening; the first training data set includes one or more of first flow characteristic data, first airway pressure characteristic data and first capacity characteristic data of mechanical ventilation, and the first test data set includes one or more of second flow characteristic data, second airway pressure characteristic data and second capacity characteristic data of mechanical ventilation; wherein the original historical data set after data labeling includes a flow characteristic data sequence F x =(f1,f2,…,f x ), airway pressure characteristic data sequence P x =(p1,p2,…,p x ), capacity characteristic data sequence V x =(v1,v2,…,v x ); the data standardization is used to x , P x and V x The data standardization formula is: Wherein, N represents one of F, P, and V, μ is the mean of the corresponding sequence, and σ is the standard deviation of the corresponding sequence; the zero-padding alignment is used to control the length of the first training data set and the first test data set, and the length formula of the zero-padding alignment is maxLen=max(len(P1,P2,…,P L )), subscript L = len(trainset) + len(testset); the matrix flattening is used to flatten and reduce the dimension of the data matrix after the zero-padding alignment; A first training module, used to input the first training data set into the GBDT model, perform training based on the one or more decision trees, create a new decision tree and update the one or more decision trees, obtain a trained GBDT model, and output a first leaf node index value set based on the trained GBDT model; A second training module, used for inputting the first leaf node index value set into the LR model for training to obtain a trained LR model; A determination module, used for inputting the first test data set into the trained GBDT model for processing to obtain a second leaf node index value set; Inputting the second leaf node index value set into the trained LR model for processing until the accuracy of the processing result based on the trained LR model reaches a preset threshold, thereby obtaining a trained GBDT+LR model; Based on the trained GBDT model, feature extraction and processing are performed on the data set to be detected corresponding to the ventilator, and based on the trained LR model, the extracted and processed features are classified and matched to obtain a detection result of human-machine asynchrony.

12. The device according to claim 11, characterized in that The determining module is specifically used for: Acquire a first test data set, wherein the first test data set includes one or more of second flow characteristic data, second airway pressure characteristic data, and second volume characteristic data of mechanical ventilation; Inputting the first test data set into the trained GBDT model for processing to obtain a second leaf node index value set; Inputting the second leaf node index value set into the trained LR model for processing until the detection result accuracy based on the trained LR model reaches a preset threshold, thereby obtaining a trained GBDT+LR model; The data set to be tested corresponding to the ventilator is input into the trained GBDT+LR model for processing to obtain the detection result of human-machine asynchrony.

13. The device according to claim 12, characterized in that The determining module is specifically used for: Acquire an original test data set, wherein the original test data set includes one or more of original test flow characteristic data, original test airway pressure characteristic data, and original test volume characteristic data; A second data preprocessing is performed on the original test data set to obtain the first test data set; the second data preprocessing includes one or more of data labeling, data standardization, and zero padding alignment.

14. The device according to any one of claims 11 to 13, characterized in that The second training module is specifically used for: One Hot encoding is performed on the first leaf node index value set to obtain a second training data set; The second training data set is input into the LR model for training to obtain a trained LR model.

15. The device according to any one of claims 12 to 14, characterized in that The determining module is specifically used for: Performing One Hot encoding processing on the second leaf node index value set to obtain a second test data set; The second test data set is input into the trained LR model for processing.

16. The device according to any one of claims 11 to 15, characterized in that The acquisition module is specifically used for: Acquire an original training data set, wherein the original training data set includes one or more of original training flow characteristic data, original training airway pressure characteristic data, and original training volume characteristic data; The original training data set is subjected to a first data preprocessing to obtain the first training data set, wherein the first data preprocessing includes one or more of data labeling, data standardization, and zero padding alignment.

17. The device according to any one of claims 11 to 16, characterized in that The construction module is specifically used for: Setting training parameters of the GBDT model to obtain the GBDT model, wherein the training parameters of the GBDT model include a maximum number of iterations of a base learner of the GBDT model, a maximum depth of a CART tree of the GBDT model, a loss function of the GBDT model, and a learning rate of the GBDT model; The training parameters of the LR model are set to obtain the LR model, wherein the training parameters of the LR model include the number of classifications and the weight vector of the LR model.

18. The device according to any one of claims 11 to 17, characterized in that The detection result of human-machine asynchrony includes a human-machine asynchrony type, and the human-machine asynchrony type includes one or more of a double trigger type, an ineffective inspiratory effort type, and a normal type.

19. The device according to claim 18, characterized in that The human-machine asynchronous type includes the double trigger type, the ineffective inspiratory effort type and the normal type; The determination module is specifically used for: Obtaining three groups of probability values ​​corresponding to the double trigger type, the ineffective inspiratory effort type and the normal type; The human-machine asynchrony type of the ventilator is determined according to the three groups of probability values ​​and preset conditions.

20. The device according to claim 19, characterized in that The device further comprises: The sending module is used to send a prompt message when the human-machine asynchronous type of the ventilator is a double trigger type or an invalid inspiratory effort type, wherein the prompt message is used to prompt medical staff to adjust the settings of the ventilator according to the human-machine asynchronous type of the ventilator.

21. A terminal, characterized in that: The method comprises a processor, an input device, an output device and a memory, wherein the processor, the input device, the output device and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the method according to any one of claims 1 to 10.

22. A computer-readable storage medium, characterized in that: The computer storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor is caused to perform the method according to any one of claims 1 to 10.

23. A ventilator, characterized in that: The invention comprises a host, an air-oxygen mixer, an air source, a humidifier, and an external pipeline, wherein the host, the air-oxygen mixer, the air source, the humidifier, and the external pipeline are interconnected, wherein the host is used to store a computer program, the computer program comprises program instructions, and the host is configured to call the program instructions to execute the method according to any one of claims 1 to 10.

Citation Information

Patent Citations

  • GBDT and LR fusion method and device based on federated learning, equipment and storage medium

    CN112288101A

  • Method and apparatus for predicting work of breathing

    US20040040560A1

  • Respiratory Equipment and Method for Controlling Respiratory Equipment

    US20090107498A1

  • Clinical Decision Support System for Patient-Ventilator Asynchrony Detection and Management

    US20200261674A1