Method and apparatus for training a respiratory waveform classification model to detect human-ventilator asynchrony

By sorting airway pressure data in ascending order and extracting features, and training a respiratory waveform classification model using basic machine learning algorithms, the problem of real-time detection and classification of asynchronous human-ventilator ventilation was solved, achieving fast and accurate waveform classification.

CN114534041BActive Publication Date: 2024-11-15SHENZHEN INST OF ADVANCED TECH CHINESE ACAD OF SCI
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210107144.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-01-28
Publication Date
2024-11-15
Estimated Expiration
2042-01-28

AI Technical Summary

Technical Problem

In existing technologies, the detection and classification of asynchronous human-machine ventilation mainly rely on doctors' judgment, and existing machine learning methods are complex and computationally intensive, making it impossible to perform real-time and effective waveform classification.

Method used

By collecting airway pressure data from consecutive respiratory cycles, sorting the data in ascending order and extracting features, and using basic machine learning algorithms such as support vector machines and nearest neighbor algorithms to train a respiratory waveform classification model, the system can achieve real-time detection and classification of asynchrony between human and ventilator ventilation.

Benefits of technology

It enables real-time detection and classification of asynchronous human-ventilator breathing, reduces computational complexity, and improves classification efficiency and accuracy. It can quickly identify normal breathing, double-triggered inspiration, and ineffective inspirational effort waveforms.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114534041B_ABST
    Figure CN114534041B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of respiratory equipment analysis, and specifically to a respiratory waveform classification model training method and device for detecting human-machine ventilation asynchrony. The method of the present invention observes various waveforms and arranges the airway pressure data of each respiratory cycle in ascending order. The data distribution characteristics of different respiratory types have obvious differences. Therefore, it is very effective to extract features after arranging the airway pressure data in ascending order, and the respiratory types according to the waveform can be distinguished very effectively. The most basic machine learning classification algorithm can be applied for rapid classification training. The respiratory waveform classification model trained by the method of the present application is very effective in detecting respiratory types.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of respiratory equipment analysis, and in particular to a respiratory waveform classification model training method and device for detecting human-machine ventilation asynchrony. Background Art

[0002] In modern clinical medicine, ventilators, as an effective means of artificially replacing spontaneous ventilation function, have been widely used in respiratory failure caused by various reasons, anesthesia respiratory management during major surgery, respiratory support therapy and emergency resuscitation, and occupy a very important position in the field of modern medicine.

[0003] The ventilator is a vital medical device that can prevent and treat respiratory failure, reduce complications, and save and prolong the patient's life. However, in the process of mechanical ventilation of patients with ventilators, abnormal ventilation often causes human-machine asynchrony. Human-machine asynchrony can have many negative effects on patients, and doctors need to adjust the ventilator parameters in time. However, it takes a certain amount of time from the doctor's observation of the asynchrony to the doctor's judgment and adjustment of the parameters. Therefore, the automatic detection and classification of human-machine asynchrony is of great benefit to patients and is a very meaningful research direction. There are many types of human-machine asynchrony, the most common of which are ineffective inspiratory effort and double-triggered inspiration. Ineffective inspiratory effort is defined as the failure to reach the trigger threshold of the ventilator after the inspiratory muscle exerts force, and no ventilator breathing occurs. Double-triggered inspiration is due to the fact that the short ventilator air delivery time is not coordinated with the patient's neural inspiratory time, causing the patient to have two ventilator air deliveries in one inspiratory effort.

[0004] Currently, the phenomenon of human-machine asynchrony is still mainly judged and handled by doctors. Moreover, in the existing technologies that use machine learning methods for automatic detection and classification, the algorithm process is generally very complex, the data dimension is too high, and it is difficult to implement. The main reason is that there is no very effective work in data processing and feature extraction, and a large amount of high-dimensional raw data needs to be input into the multi-layer neural network for machine learning. This results in the existing technology having a very large amount of calculation and a very complex model, making it impossible to perform instant and effective waveform classification. Summary of the invention

[0005] The embodiments of the present invention provide a method and device for training a respiratory waveform classification model for detecting human-machine ventilation asynchrony, so as to at least solve the technical problem of real-time detection and classification of existing human-machine ventilation asynchrony.

[0006] According to an embodiment of the present invention, a respiratory waveform classification model training method for detecting human-machine ventilation asynchrony is provided, comprising the following steps:

[0007] Collect respiratory data of continuous respiratory cycles, including airway pressure data;

[0008] The airway pressure data is divided into a number of signal segments, each of which corresponds to a complete breathing cycle;

[0009] According to the waveform characteristics of the airway pressure data, multiple different types of breathing type annotations are performed, including normal breathing waveform, double-triggered inspiratory waveform and ineffective inspiratory effort waveform;

[0010] Read the airway pressure data of each respiratory cycle, and arrange the airway pressure data of each respiratory cycle in ascending order;

[0011] Extracting characteristic data of each airway pressure data, and using each characteristic data as a characteristic value corresponding to each respiratory cycle;

[0012] All eigenvalues ​​are stored in a data table to obtain eigenvalue data sets of all respiratory cycles, and respiratory type annotations are stored in the data table in the order in which the eigenvalues ​​are stored in the data table to obtain an annotation list corresponding to the eigenvalue data set;

[0013] Divide the feature value data set into a training set and a test set;

[0014] Based on the machine learning classification algorithm, the training set and the test set were used for model training to obtain the respiratory waveform classification model.

[0015] Furthermore, the respiratory data of the continuous respiratory cycle is collected as follows:

[0016] The experimental platform was built using Test simulated lung and Mindray SV300 ventilator for data collection.

[0017] Furthermore, the airway pressure data is divided into several signal segments, where each signal segment corresponds to a complete breath:

[0018] Every two troughs constitute a complete breathing cycle.

[0019] Furthermore, the airway pressure data is divided into several signal segments, wherein each signal segment corresponds to a complete breath and includes:

[0020] After segmenting the airway pressure data, the airway pressure data for each respiratory cycle needs to be saved separately as a table file.

[0021] Furthermore, the characteristic data of each airway pressure data is extracted, and each characteristic data is used as the characteristic value corresponding to each respiratory cycle, specifically:

[0022] Extract feature data according to feature calculation rules;

[0023] The characteristic calculation rule is:

[0024] Divide each respiratory cycle sorted in ascending order into 10 equal number of characteristic data points;

[0025] The first feature data point is the minimum value of the entire set of sorted data, and 10 other feature points are taken, and the maximum value of each data point is taken in turn, adding up to a total of 11 feature points;

[0026] The 11 calculated characteristic data are combined as characteristic values ​​of each corresponding respiratory cycle.

[0027] Furthermore, based on the machine learning classification algorithm, the training set and the test set are used to perform model training, and the respiratory waveform classification model is obtained as follows:

[0028] Support vector machine algorithm, nearest neighbor algorithm, logistic regression algorithm, decision tree algorithm, random forest algorithm and naive Bayes algorithm are used for classification model training.

[0029] A respiratory waveform classification model training device for detecting human-machine ventilation asynchrony, comprising:

[0030] A data acquisition module, used for acquiring respiratory data of continuous respiratory cycles, the respiratory data including airway pressure data;

[0031] A data segmentation module, used for segmenting the airway pressure data into a number of signal segments, wherein each signal segment corresponds to a complete breathing cycle;

[0032] A data annotation module, used to annotate multiple different types of breathing types according to the waveform characteristics of airway pressure data;

[0033] A data sorting module is used to read the airway pressure data of each respiratory cycle and arrange the airway pressure data of each respiratory cycle in ascending order;

[0034] A data extraction module is used to extract characteristic data of each airway pressure data, and use each characteristic data as a characteristic value corresponding to each respiratory cycle;

[0035] A data set generation module is used to store all eigenvalues ​​into a data table to obtain eigenvalue data sets of all respiratory cycles, and store the respiratory type annotations into the data table in the order in which the eigenvalues ​​are stored in the data table to obtain an annotation list corresponding to the eigenvalue data set;

[0036] Data classification module, used to divide the feature value data set into training set and test set;

[0037] The model training module is used to perform model training based on the machine learning classification algorithm, apply the training set and the test set, and obtain the respiratory waveform classification model.

[0038] A method for real-time classification of respiratory waveform data, using any of the above respiratory waveform classification models, comprises the following steps:

[0039] Collect respiratory data of continuous respiratory cycles, including airway pressure data;

[0040] Extracting the collected airway pressure data, sorting the airway pressure data in ascending order, and extracting characteristic data of the airway pressure data in groups;

[0041] The extracted feature data is used as the feature value of the corresponding respiratory cycle and input into the trained respiratory waveform classification model;

[0042] According to the input feature value, the respiratory waveform classification model determines the respiratory waveform type of this breath, and the respiratory waveform types include normal respiratory waveform, double-triggered inspiratory waveform and ineffective inspiratory effort waveform;

[0043] If the output breathing type is double-trigger breathing or ineffective inspiratory effort, it is judged as human-machine asynchrony phenomenon, and the breathing waveform type output will show a prompt message and indicate the type of human-machine asynchrony;

[0044] If the output is normal breathing, it is judged as non-human-machine asynchrony phenomenon.

[0045] A computer-readable medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in any of the above-mentioned methods for training a respiratory waveform classification model for detecting human-machine ventilation asynchrony.

[0046] A terminal device comprises: a processor, a memory and a communication bus; the memory stores a computer-readable program executable by the processor;

[0047] The communication bus realizes the connection and communication between the processor and the memory;

[0048] When the processor executes the computer-readable program, the processor implements the steps in any of the above-mentioned methods for training a respiratory waveform classification model for detecting human-machine ventilation asynchrony.

[0049] The respiratory waveform classification model training method and device for detecting human-machine ventilation asynchrony in the embodiment of the present invention, the method of the present invention observes various waveforms, and after arranging the airway pressure data of each respiratory cycle in ascending order, the data distribution characteristics of different respiratory types have obvious differences; therefore, it is very effective to arrange the airway pressure data in ascending order and then extract features, which can very effectively distinguish the respiratory types according to the waveform; and the most basic machine learning classification algorithm can be applied for rapid classification training, and the respiratory waveform classification model trained by the method of the present application is very effective in detecting respiratory types. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0051] Figure 1 A flow chart of a respiratory waveform classification model training method for detecting human-machine ventilation asynchrony according to the present invention;

[0052] Figure 2 This is a schematic diagram of a respiratory waveform classification model training device for detecting human-machine ventilation asynchrony according to the present invention;

[0053] Figure 3 A flow chart of a method for real-time classification of respiratory waveform data of the present invention;

[0054] Figure 4 A normal breathing airway pressure waveform diagram of a complete breathing cycle of the present invention;

[0055] Figure 5 It is a double-triggered respiratory airway pressure waveform diagram of a complete respiratory cycle of the present invention;

[0056] Figure 6 It is a waveform diagram of the airway pressure of the ineffective inspiratory effort of a complete breathing cycle of the present invention;

[0057] Figure 7 It is a complete arrangement waveform diagram of normal breathing airway pressure after being arranged in ascending order according to the present invention;

[0058] Figure 8 It is a double-trigger breathing waveform diagram of the present invention after being arranged in ascending order;

[0059] Fig. 9 This is a schematic diagram of the terminal device of the present invention. DETAILED DESCRIPTION

[0060] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0061] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0062] See also Figure 1 According to an embodiment of the present invention, a respiratory waveform classification model training method for detecting human-machine ventilation asynchrony is provided, comprising the following steps:

[0063] S101: Collecting respiratory data of continuous respiratory cycles, where the respiratory data includes airway pressure data.

[0064] Specifically, the experimental platform was built with the Test simulated lung and the Mindray SV300 ventilator for data collection. The experimental data here can also collect the respiratory data of real patients as a substitute; the breathing mode of the simulated lung is set to acute respiratory distress syndrome (ARDS); the sampling frequency is 50HZ, and the breathing rate is 21 times per minute; the ventilation mode of the ventilator is set to CPAP / PSV mode; the Test simulated lung can output data from 15 channels, such as flow, airway pressure, tidal volume, alveolar pressure, intrapleural pressure, heart pressure, etc.

[0065] The present invention uses airway pressure data for example analysis because the waveform characteristics of airway pressure data are obvious, it is easy to distinguish different types of waveforms, and the values ​​are all positive; the present invention chooses to simulate the breathing of patients with acute respiratory distress syndrome as experimental data, but the method and principle are not limited to this type of disease.

[0066] In terms of data collection, the present invention collects simulated breathing data of patients with acute respiratory distress syndrome and selects certain parameters; however, experimental data collected by selecting other parameters or through other channels can also apply the same feature extraction principles, classifier selection principles, practical application principles, etc., so they need to be protected.

[0067] In terms of the selection of respiratory data, the respiratory data selected in the present invention is airway pressure data; although data such as tidal volume is different from the concept of airway pressure data, the data distribution characteristics are similar to those of airway pressure data, and the method of the present invention can also be used to extract the characteristics of the wave.

[0068] S102: Segment the airway pressure data into a plurality of signal segments, wherein each signal segment corresponds to a complete breathing cycle.

[0069] Specifically, the airway pressure data of the collected continuous respiratory signal is segmented; the segmentation principle is that every two troughs are divided into a complete respiratory cycle. After the segmentation, the airway pressure data of each respiratory cycle needs to be saved separately as a table file for subsequent use; when the respiratory cycle is divided, the number of data points of each respiratory data segment can be different, but preferably the number should not be too different. For example, in the experimental data used in the present invention, the number of data points of each divided respiratory cycle varies between 80 and 90.

[0070] S103: performing multiple different types of breathing type annotations according to the waveform characteristics of the airway pressure data, the breathing type annotations including a normal breathing waveform, a double-triggered inspiratory waveform, and an ineffective inspiratory effort waveform.

[0071] In the embodiment, each divided respiratory cycle is classified according to its waveform characteristics; the present invention mainly classifies it into normal respiratory waveform (such as Figure 4 ), double-trigger inspiratory waveform (such as Figure 5 ), ineffective inspiratory effort waveform (such as Figure 6 );

[0072] Specifically, in this step, a total of 1,200 normal breathing waveforms, 1,500 double-triggered inspiratory waveforms, and 1,400 ineffective inspiratory effort waveforms of airway pressure data were collected; the data of the three types of breathing were stored in three different folders, and the breathing types were marked on the folders to complete the classification of breathing types.

[0073] S104: Read the airway pressure data of each respiratory cycle, and arrange the airway pressure data of each respiratory cycle in ascending order.

[0074] Step S104 includes:

[0075] Extract feature data according to feature calculation rules;

[0076] The characteristic calculation rule is:

[0077] S1041: Divide each respiratory cycle arranged in ascending order into 10 equal number of characteristic data points;

[0078] S1042: The first feature data point is the minimum value of the entire set of sorted data, and 10 other feature points are taken, and the maximum value of each data point is taken in turn, adding up to a total of 11 feature points;

[0079] S1043: The 11 calculated characteristic data are combined as characteristic values ​​of each corresponding respiratory cycle.

[0080] Specifically, programming software is used to read the airway pressure data of each respiratory cycle in each folder, and the data of each respiratory cycle is sorted in ascending order while reading, and 11 feature data are extracted in total; the calculation rule of the feature data is: divide each respiratory cycle after ascending order into 10 equal parts. The first feature data point takes the minimum value of the entire group of data after sorting, and the other 10 feature points take the maximum value of each data point in turn, adding up to a total of 11 feature points; the 11 feature points are respectively recorded as: fea1, fea2, fea3, fea4, fea5, fea6, fea7, fea8, fea9, fea10, fea11. Among them, fea1 is the minimum value of the entire group of data, and fea11 is the maximum value of the entire group of data.

[0081] In terms of feature extraction, the feature extraction of the present invention is a key part of classification using machine learning; there are many methods of machine learning, not only simple machine learning classification algorithms, but also neural network algorithms, deep learning algorithms, etc.; the feature extraction method of the present invention, the extracted features can be used for effective classification by any machine learning algorithm.

[0082] The feature calculation rule used in the present invention to extract feature numbers is not only applicable to the feature extraction of normal breathing data, double-triggered breathing data, and invalid inspiratory effort breathing data; other types of respiratory waveform data can also be used for feature extraction using the same principle and method, and ultimately achieve the classification effect.

[0083] In this embodiment, each breathing cycle is divided into 10 parts for feature extraction; in other embodiments, each breathing cycle can also be evenly divided into other numbers of parts and feature extraction can be performed; in addition, the present invention arranges the data in ascending order, but it is also feasible to arrange the data in descending order and then use the same feature extraction method, which is not limited here.

[0084] like Figure 4 As shown, it is a normal breathing airway pressure waveform of a complete breathing cycle; the ordinate is the value of the airway pressure (unit:), and the abscissa is the serial number of the data point in this cycle.

[0085] like Figure 5 As shown, the double-triggered breathing airway pressure waveform of a complete breathing cycle; the ordinate is the value of the airway pressure (unit:), and the abscissa is the serial number of the data point in this cycle.

[0086] like Figure 6 As shown, the airway pressure waveform of the ineffective inspiratory effort of a complete respiratory cycle; the ordinate is the value of the airway pressure (unit:), and the abscissa is the serial number of the data point in this cycle.

[0087] like Figure 7As shown in the figure, it is a complete normal breathing airway pressure arrangement waveform after being arranged in ascending order. Figure 8 As shown, the double-trigger breathing waveform is arranged in ascending order.

[0088] from Figures 4 to 8 It can be seen that the shape of the graph after the normal breathing is arranged is different from that before the arrangement.

[0089] S105: Extract characteristic data of each airway pressure data, and use each characteristic data as a characteristic value corresponding to each respiratory cycle.

[0090] Specifically, the 11 characteristic data obtained by calculation are combined as the characteristic values ​​of the corresponding breathing cycle, namely [fea1, fea2, fea3, fea4, fea5, fea6, fea7, fea8, fea9, fea10, fea11].

[0091] The present invention arranges in ascending order in each respiratory cycle and then extracts features in segments. The present embodiment uses a method of taking the maximum value by grouping after arrangement. In other embodiments, the minimum value can be taken by grouping, the median can be taken by grouping, or statistical features can be extracted by grouping after sorting, etc., which can also achieve effective feature extraction.

[0092] S106: All eigenvalues ​​are stored in a data table to obtain eigenvalue data sets of all respiratory cycles, and respiratory type annotations are stored in the data table in the order in which the eigenvalues ​​are stored in the data table to obtain an annotation list corresponding to the eigenvalue data set.

[0093] Specifically, the characteristic values ​​of all respiratory cycles are stored in a data table according to the file reading order, so that the characteristic value data set of all respiratory cycles is obtained;

[0094] Create a list to store the labels of the breathing waveform types. When storing the feature values ​​one by one in the feature data table, the labels of the breathing types are also stored in this list in the same order; for example, the normal breathing waveform is marked with 0, the double-triggered breathing waveform is marked with 1, and the ineffective inspiratory effort waveform is marked with 2; at this time, a feature data set containing all the breathing cycle features and a label list with the same order are obtained.

[0095] S107: Divide the eigenvalue data set into a training set and a test set.

[0096] Specifically, 70% of the data in the feature value data set is set as the training set, and 30% of the data is set as the test set; the ratio of the training set to the test set can be selected from other ratios and can be adjusted based on experience.

[0097] S108: Based on the machine learning classification algorithm, the training set and the test set are used to perform model training to obtain a respiratory waveform classification model.

[0098] Specifically, the training set and the test set are input into the machine learning classification algorithm for model training and testing; wherein the support vector machine algorithm, the nearest neighbor node algorithm, the logistic regression algorithm, the decision tree algorithm, the random forest algorithm and the naive Bayes algorithm of the present invention are used for classification model training;

[0099] In the test results of the present invention: the classification accuracy of the vector machine algorithm, the nearest node algorithm, the logistic regression algorithm, the decision tree algorithm, and the random forest algorithm is as high as 100.00%, all 1230 classification tasks are classified correctly, and the accuracy of the naive Bayes algorithm is 99.92%; after obtaining the classification accuracy results, the classifier model is selected and saved.

[0100] The respiratory waveform classification model training method and device for detecting human-machine ventilation asynchrony in the embodiment of the present invention, the method of the present invention observes various waveforms, and after arranging the airway pressure data of each respiratory cycle in ascending order, the data distribution characteristics of different respiratory types have obvious differences; therefore, it is very effective to arrange the airway pressure data in ascending order and then extract features, which can very effectively distinguish the respiratory types according to the waveform; and the most basic machine learning classification algorithm can be applied for rapid classification training, and the respiratory waveform classification model trained by the method of the present application is very effective in detecting respiratory types.

[0101] In addition, the principle of the method of the present application is simple and easy to apply. It only needs to arrange in ascending order and then extract the maximum value in segments to very effectively extract the characteristics of each segment of waveform data; the most basic machine learning classification algorithm can be used, which is very convenient to operate; the number of data points in each group of the present invention can be different, which greatly improves the feasibility in practical applications, and high-dimensional data is processed by dimensionality reduction, which improves computing efficiency.

[0102] The present invention provides an effective method for extracting respiratory data features and a set of detection and classification algorithms for human-machine asynchrony, which solve the problem of real-time detection and classification of human-machine asynchrony. The present invention mainly classifies normal respiratory waveforms, double-triggered inspiratory waveforms, and invalid inspiratory effort waveforms. In addition, other types of human-machine asynchrony waveforms can also be classified using the same principle.

[0103] See also Figure 2 According to another embodiment of the present invention, a respiratory waveform classification model training device for detecting human-machine ventilation asynchrony is provided, comprising:

[0104] The data acquisition module 100 is used to acquire respiratory data of continuous respiratory cycles, where the respiratory data includes airway pressure data;

[0105] A data segmentation module 200, for segmenting the airway pressure data into a plurality of signal segments, wherein each signal segment corresponds to a complete breathing cycle;

[0106] A data labeling module 300, for labeling a plurality of different types of breathing types according to the waveform characteristics of the airway pressure data;

[0107] The data sorting module 400 is used to read the airway pressure data of each respiratory cycle and arrange the airway pressure data of each respiratory cycle in ascending order;

[0108] A data extraction module 500 is used to extract characteristic data of each airway pressure data, and use each characteristic data as a characteristic value corresponding to each respiratory cycle;

[0109] The data set generation module 600 is used to store all the eigenvalues ​​into a data table to obtain the eigenvalue data set of all the respiratory cycles, and store the respiratory type annotations into the data table in the order in which the eigenvalues ​​are stored into the data table to obtain an annotation list corresponding to the eigenvalue data set;

[0110] A data classification module 700 is used to divide the feature value data set into a training set and a test set;

[0111] The model training module 800 is used to perform model training based on a machine learning classification algorithm, using a training set and a test set to obtain a respiratory waveform classification model.

[0112] The respiratory waveform classification model training method and device for detecting human-machine ventilation asynchrony in the embodiment of the present invention, the device of the present invention observes various waveforms, and after arranging the airway pressure data of each respiratory cycle in ascending order, the data distribution characteristics of different respiratory types have obvious differences; therefore, it is very effective to arrange the airway pressure data in ascending order and then extract features, which can very effectively distinguish the respiratory types according to the waveform; and the most basic machine learning classification algorithm can be applied for rapid classification training, and the respiratory waveform classification model trained by the method of the present application is very effective in detecting respiratory types.

[0113] See also Figure 3 According to another embodiment of the present invention, a method for real-time classification of respiratory waveform data is provided, which uses any one of the above respiratory waveform classification models and includes the following steps:

[0114] S201: Collecting respiratory data of continuous respiratory cycles, where the respiratory data includes airway pressure data;

[0115] S202: extracting the collected airway pressure data, sorting the airway pressure data in ascending order, and extracting characteristic data of the airway pressure data in groups;

[0116] S203: using the extracted characteristic data as characteristic values ​​of the corresponding respiratory cycle and inputting them into a trained respiratory waveform classification model;

[0117] S204: The respiratory waveform classification model determines the respiratory waveform type of this breathing according to the input feature value, and the respiratory waveform type includes a normal respiratory waveform, a double-triggered inspiratory waveform, and an ineffective inspiratory effort waveform;

[0118] S205: If the output breathing type is double-trigger breathing and ineffective inspiratory effort, it is determined to be a phenomenon of human-machine asynchrony, and a prompt message is output for the breathing waveform type, indicating the type of human-machine asynchrony;

[0119] S206: If the output is normal breathing, it is determined to be a non-human-machine asynchrony phenomenon.

[0120] A method for instant classification of respiratory waveform data in an embodiment of the present invention can instantly detect the occurrence of human-machine ventilation asynchrony. The present invention can automatically and instantly classify and output the patient's breathing type after each breath of the patient.

[0121] Based on the above-mentioned breathing waveform classification model training method for detecting human-machine ventilation asynchrony, the present embodiment provides a computer-readable storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the steps in the breathing waveform classification model training method for detecting human-machine ventilation asynchrony as in the above-mentioned embodiment.

[0122] A terminal device includes: a processor, a memory and a communication bus; the memory stores a computer-readable program that can be executed by the processor; the communication bus realizes the connection and communication between the processor and the memory; when the processor executes the computer-readable program, the steps in the above-mentioned breathing waveform classification model training method for detecting human-machine ventilation asynchrony are realized.

[0123] Based on the above-mentioned respiratory waveform classification model training method for detecting human-machine ventilation asynchrony, the present application provides a terminal device, such as Fig. 9 As shown, it includes at least one processor (processor) 20; display screen 21; and memory (memory) 22, and may also include a communications interface (Communications Interface) 23 and a bus 24. Among them, the processor 20, the display screen 21, the memory 22 and the communication interface 23 can communicate with each other through the bus 24. The display screen 21 is configured to display a preset user guide interface in the initial setting mode. The communication interface 23 can transmit information. The processor 20 can call the logic instructions in the memory 22 to execute the method in the above embodiment.

[0124] In addition, the logic instructions in the memory 22 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product.

[0125] The memory 22 is a computer-readable storage medium that can be configured to store software programs, computer executable programs, such as program instructions or modules corresponding to the methods in the embodiments of the present disclosure. The processor 20 executes functional applications and data processing by running the software programs, instructions or modules stored in the memory 22, that is, implementing the methods in the above embodiments.

[0126] The memory 22 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the terminal device, etc. In addition, the memory 22 may include a high-speed random access memory and may also include a non-volatile memory. For example, a variety of media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, may also be a transient storage medium.

[0127] In addition, the specific process of loading and executing the multiple instruction processors in the above-mentioned storage medium and the terminal device has been described in detail in the above-mentioned method, and will not be described one by one here.

[0128] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A respiratory waveform classification model training method for detecting human-machine ventilation asynchrony, characterized in that: The following steps are involved: Collecting respiratory data of continuous respiratory cycles, wherein the respiratory data includes airway pressure data; Segmenting the airway pressure data into a plurality of signal segments, wherein each of the signal segments corresponds to a complete breathing cycle; Performing multiple different types of breathing type annotations according to the waveform characteristics of the airway pressure data, wherein the breathing type annotations include a normal breathing waveform, a double-triggered inspiratory waveform, and an ineffective inspiratory effort waveform; Reading the airway pressure data of each of the respiratory cycles, and arranging the airway pressure data of each of the respiratory cycles in ascending order; Extracting characteristic data of each of the airway pressure data, and using each of the characteristic data as a characteristic value corresponding to each of the respiratory cycles; All the characteristic values ​​are stored in a data table to obtain a characteristic value data set of all respiratory cycles, and the respiratory type annotations are stored in the data table in the order in which the characteristic values ​​are stored in the data table to obtain a annotation list corresponding to the characteristic value data set; Dividing the eigenvalue data set into a training set and a test set; Based on a machine learning classification algorithm, the training set and the test set are used to perform model training to obtain a respiratory waveform classification model; The extracting of characteristic data of each of the airway pressure data and taking each of the characteristic data as the characteristic value corresponding to each of the respiratory cycles is specifically as follows: Extracting the feature data according to a feature calculation rule; The characteristic calculation rule is: Divide each of the respiratory cycles arranged in ascending order into 10 equal number of characteristic data points; The first feature data point is the minimum value of the entire set of sorted data, and 10 other feature points are taken, and the maximum value of each data point is taken in turn, adding up to a total of 11 feature points; The 11 characteristic data obtained by calculation are combined as characteristic values ​​of each corresponding respiratory cycle.

2. The respiratory waveform classification model training method for detecting human-machine ventilation asynchrony according to claim 1, characterized in that: The respiratory data collected during continuous respiratory cycles are specifically: The Test simulated lung and Mindray SV300 ventilator were used to build an experimental platform for data collection.

3. The respiratory waveform classification model training method for detecting human-machine ventilation asynchrony according to claim 1, characterized in that: The airway pressure data is divided into a plurality of signal segments, wherein each signal segment corresponds to a complete breath: Every two troughs constitute a complete breathing cycle.

4. The respiratory waveform classification model training method for detecting human-machine ventilation asynchrony according to claim 3, characterized in that: The step of dividing the airway pressure data into a plurality of signal segments, wherein each of the signal segments corresponds to a complete breath and includes: After the airway pressure data is segmented, the airway pressure data of each respiratory cycle needs to be saved separately as a table file.

5. The respiratory waveform classification model training method for detecting human-machine ventilation asynchrony according to claim 1, characterized in that: The machine learning classification algorithm is based on the training set and the test set to perform model training to obtain the respiratory waveform classification model: Support vector machine algorithm, nearest neighbor algorithm, logistic regression algorithm, decision tree algorithm, random forest algorithm and naive Bayes algorithm are used for classification model training.

6. A respiratory waveform classification model training device for detecting human-machine ventilation asynchrony, characterized in that: include: A data acquisition module, used for acquiring respiratory data of continuous respiratory cycles, wherein the respiratory data includes airway pressure data; A data segmentation module, used for segmenting the airway pressure data into a plurality of signal segments, wherein each of the signal segments corresponds to a complete breathing cycle; A data annotation module, used for annotating a plurality of different types of breathing types according to the waveform characteristics of the airway pressure data; A data sorting module, used for reading the airway pressure data of each respiratory cycle, and arranging the airway pressure data of each respiratory cycle in ascending order; A data extraction module is used to extract characteristic data of each of the airway pressure data and use each of the characteristic data as a characteristic value corresponding to each of the respiratory cycles; specifically: The extracting of characteristic data of each of the airway pressure data and taking each of the characteristic data as the characteristic value corresponding to each of the respiratory cycles is specifically as follows: Extracting the feature data according to a feature calculation rule; The characteristic calculation rule is: Divide each of the respiratory cycles arranged in ascending order into 10 equal number of characteristic data points; The first feature data point is the minimum value of the entire set of sorted data, and 10 other feature points are taken, and the maximum value of each data point is taken in turn, adding up to a total of 11 feature points; The 11 characteristic data obtained by calculation are combined as characteristic values ​​of each corresponding respiratory cycle; A data set generation module, used to store all the characteristic values ​​into a data table to obtain a characteristic value data set of all respiratory cycles, and store the respiratory type annotations into the data table in the order in which the characteristic values ​​are stored in the data table to obtain a annotation list corresponding to the characteristic value data set; A data classification module, used for dividing the feature value data set into a training set and a test set; The model training module is used to perform model training based on a machine learning classification algorithm, apply the training set and the test set, and obtain a respiratory waveform classification model.

7. A method for real-time classification of respiratory waveform data, characterized in that: The respiratory waveform classification model training method for detecting human-machine ventilation asynchrony according to any one of claims 1 to 5 comprises the following steps: Collecting respiratory data of continuous respiratory cycles, wherein the respiratory data includes airway pressure data; Extracting the collected airway pressure data, sorting the airway pressure data in ascending order, and extracting characteristic data of the airway pressure data in groups; The extracted characteristic data is used as the characteristic value of the corresponding respiratory cycle and input into the trained respiratory waveform classification model; According to the inputted characteristic value, the respiratory waveform classification model determines the respiratory waveform type of this breathing, wherein the respiratory waveform type includes a normal respiratory waveform, a double-triggered inspiratory waveform, and an ineffective inspiratory effort waveform; If the output breathing type is double-trigger breathing or ineffective inspiratory effort, it is judged as a phenomenon of human-machine asynchrony, and a prompt message is output for the breathing waveform type, indicating the type of human-machine asynchrony; If the output is normal breathing, it is judged as non-human-machine asynchrony phenomenon; The extracting of characteristic data of each of the airway pressure data and taking each of the characteristic data as the characteristic value corresponding to each of the respiratory cycles is specifically as follows: Extracting the feature data according to a feature calculation rule; The characteristic calculation rule is: Divide each of the respiratory cycles arranged in ascending order into 10 equal number of characteristic data points; The first feature data point is the minimum value of the entire set of sorted data, and 10 other feature points are taken, and the maximum value of each data point is taken in turn, adding up to a total of 11 feature points; The 11 characteristic data obtained by calculation are combined as characteristic values ​​of each corresponding respiratory cycle.

8. A computer-readable medium, characterized in that The computer-readable medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in the respiratory waveform classification model training method for detecting human-machine ventilation asynchrony as described in any one of claims 1-5.

9. A terminal device, characterized in that: include: Processor, memory and communication bus; The memory stores a computer-readable program executable by the processor; The communication bus realizes the connection and communication between the processor and the memory; When the processor executes the computer-readable program, the steps in the respiratory waveform classification model training method for detecting human-machine ventilation asynchrony as described in any one of claims 1-5 are implemented.