Respiration and exhalation conversion method of breathing machine

Characteristic extraction and prediction of non-invasive ventilator air flow data through the CNN-LSTM network model solves the problems of poor synchronization and false triggering of non-invasive ventilator, and achieves high-accurate breathing and breathing conversion control, improving the patient's user experience.

CN118925007APending Publication Date: 2024-11-12NANJING SUPERSTAR MEDICAL EQUIP
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310529317.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-11
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The existing non-invasive ventilators have poor synchronization during use and are easily disturbed by air leakage and noise, resulting in mistriggering, and the patient is intolerant. In actual applications, the sensitivity of the algorithm based on flow graph is affected by the waveform shape and is frequently misjudged.

Method used

The CNN-LSTM network model is used to sample and extract the non-invasive ventilator air flow data through sliding windows, build training sets, verification sets and test sets, predict the breathing and ventilator conversion points through feature engineering and model, and set thresholds to control the air supply of the ventilator and Deflate switching to improve prediction accuracy.

Benefits of technology

The accurate prediction of the ventilator's breathing and breathing conversion is achieved, with an accuracy rate of 0.95 and an error trigger rate of less than 3%, improving the patient's user experience and the synchronization of the equipment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118925007A_ABST
    Figure CN118925007A_ABST
Patent Text Reader

Abstract

The invention discloses an inspiration and expiration conversion method of a breathing machine, which comprises the following steps: step 1) collecting the air flow of a noninvasive breathing machine to obtain time sequence data; 2) judging the inspiration and expiration state of the time sequence data, and marking according to the inspiration and expiration state; 3) performing sliding window sampling and feature extraction on the time series data to obtain new sliding window data, retaining corresponding marks, and creating a new data set; 4) dividing the new data set to obtain a training set, a verification set and a test set; 5) predicting an inspiration and call conversion point by constructing a CNN-LSTM network model; and 6) setting a threshold value for state classification of the breathing machine inhalation and exhalation switching point, and controlling the breathing machine to switch between air supply and air deflation. According to the method, on the basis of feature engineering and a CNN-LSTM network, inspiration and call conversion prediction is carried out, meanwhile, a new data set is generated through feature screening, a verification set and a test set are selected according to a certain proportion, and the model can better predict inspiration and call conversion points.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of ventilators, and in particular relates to a breathing-exhalation conversion method of a ventilator. Background Art

[0002] Non-invasive ventilator refers to effective mechanical ventilation that does not require tracheal intubation or incision, and can be connected through a nasal mask, face mask or laryngeal mask. When using a non-invasive ventilator, the airflow passes through the mouth or nasal cavity, then through the pharynx, and finally reaches the lungs through the trachea. Non-invasive breathing is currently widely used in clinical practice and has become an important means of treating respiratory failure, especially early acute respiratory failure and chronic respiratory failure. Many patients have benefited greatly from early intervention treatment with non-invasive ventilators, which not only improves their condition and saves costs, but also protects patients from damage caused by invasive mechanical ventilation and various possible complications. In general, it is a device that can play the main function of the lungs, while assisting normal physiological breathing of people, increasing lung ventilation, thereby improving respiratory function, reducing respiratory work consumption, and saving heart reserve capacity. It is equivalent to a "lung" in the patient's external world.

[0003] The mainstream algorithms related to respiratory triggering can be divided into three categories, among which zero-crossing detection and threshold-based flow triggering detect through thresholds, and the threshold size (also known as sensitivity) can be changed to advance or delay the trigger point. Triggering based on flow graphs is more flexible than the first two fixed threshold setting methods. The trigger point can change with the breathing process and can adapt to different people.

[0004] At present, the areas that need to be improved include: 1. During use, the synchronization of the machine is not good, which makes the patient more laborious and thus causes intolerance. 2. The disadvantage of the fixed threshold algorithm used in the current mainstream respiratory trigger algorithm is that it is easily affected by air leakage and noise interference, and is prone to false triggering; while the algorithm based on flow graph requires the flow waveform to be relatively regular and smooth, and the trigger sensitivity is affected by the shape of the waveform. In actual application, the patient's coughing, swallowing, sobbing and other behaviors may cause the flow to drop or rise suddenly, and respiratory trigger misjudgment caused by the waveform shape often occurs. Summary of the invention

[0005] In order to solve the problems existing in the prior art, the present invention provides a breathing conversion method of a ventilator to solve the above technical problems, which is specifically implemented by the following technical solutions: The breathing conversion method of the ventilator comprises the following steps: Step 1) collecting the air flow of the non-invasive ventilator to obtain time series data including the respiratory air flow; Step 2) determining the breathing state of the time series data and marking it according to the breathing state; Step 3) Perform sliding window sampling and feature extraction on the time series data to obtain new sliding window data and retain the corresponding labels to create a new data set; Step 4) Divide the new data set into a training set, a validation set, and a test set; Step 5) Predict the inhalation-exhalation transition point by constructing a CNN-LSTM network model; Step 6) Setting a threshold for state classification of the ventilator's inhalation-exhalation switching point to control the ventilator to switch between air supply and air release.

[0006] The further design of the breathing machine's breathing-exhalation conversion method is that the step 2) specifically includes the following steps: Step 2-1) According to the clinical judgment standard of non-invasive ventilator respiratory airflow, the data is judged and marked as: inspiration, exhalation and intermediate state; Step 2-2) Integrate the obtained labeling results to obtain an airflow time series data set containing three breathing states.

[0007] The further design of the breathing machine's breathing-exhalation conversion method is that the specific process of step 3) is as follows: Step 3-1) In combination with the actual scenario, set the specific sliding window width and the interval between each sliding window, and extract four time series features by moving the sliding window in the marked respiratory airflow time series data set: the average value of the sliding window data, the maximum value of the sliding window data, the minimum value of the sliding window data, and the difference between the maximum and minimum values ​​in the sliding window data; Step 3-2) At the same time, the obtained sliding window data is re-labeled according to the labeled data set results: if there is an inhalation conversion point in the sliding window data, the four time series feature data states obtained by the corresponding sliding window data are labeled as the inhalation conversion point; if there is an exhalation conversion point in the sliding window data, the four time series feature data states obtained by the corresponding sliding window data are labeled as the exhalation conversion point; if there is neither an inhalation conversion point nor an exhalation conversion point in the sliding window data, the four time series feature data states obtained by the corresponding sliding window data are labeled as the intermediate conversion point; Step 3-3) Integrate the above results to obtain a new labeled feature data set.

[0008] The further design of the breathing machine's breathing-exhalation conversion method is that the specific implementation method of step 4) is: Step 4-1) Perform Min-Max normalization on the labeled data set; Step 4-2) Split the new dataset into training set, validation set, and test set.

[0009] The further design of the breathing machine's breathing-exhalation conversion method is that the steps of step 5) are: Step 5-1) Input the training set into the model for predicting inhalation and exhalation transitions based on feature engineering and CNN-LSTM network; Step 5-2) The model for predicting breathing conversion based on feature engineering and CNN-LSTM network is set to include three convolution layers conv1, LSTM layer, maximum pooling layer max_pooling and dense layer. The number of convolution layers conv1_0, conv1_1, conv1_2 is 1, the length of the kernel size is 2, the width is the width of the input data, and the span is 1; the window size of the maximum pooling sampling layer maxpool is 2; the number of nodes in the LSTM network layer is 8; the output dimension of the dense layer is 1, and the activation function is linear; Step 5-3) Set the loss function of the model to the absolute mean error, set the optimizer to the Adam algorithm, reversely calculate the gradient of each parameter, and continuously update the parameters in the network; Step 5-4) Use the number of training times as the horizontal axis and the values ​​of the loss function for training and validation as the vertical axis to display the loss function values ​​during the training process; Step 5-5) Obtain the trained model according to the above method and visualize the model through the netron visualization tool.

[0010] The further design of the inhalation-exhalation conversion method of the ventilator is that the step 6) is specifically as follows: inputting the test set into the trained model, setting the threshold according to the output result of the model, including the inhalation state threshold and the exhalation state threshold to achieve a model with the highest accuracy; when the output result of the model reaches the inhalation state threshold, the inhalation conversion point is reached, and the ventilator will cooperate to perform the air supply operation; when the output result of the model reaches the exhalation state threshold, the exhalation conversion point is reached, and the ventilator will cooperate to perform the deflation operation; when the output results of the model do not meet the inhalation state threshold and the inhalation state threshold, this is an intermediate state, and the ventilator will maintain the current state.

[0011] The present invention also provides an electronic device, comprising a memory, a processor and a computer program, wherein the computer program is stored in the memory and is configured to be executed by the processor to implement the breathing-exhalation conversion method of the ventilator.

[0012] The present invention also provides a computer-readable storage medium, wherein the storage medium stores a computer program, and the computer program is executed by a processor to implement the breathing-exhalation conversion method of the ventilator.

[0013] Beneficial effects of the present invention: The inhalation-exhalation conversion method of the ventilator of the present invention predicts the inhalation-exhalation conversion based on air flow data based on feature engineering and CNN-LSTM network. At the same time, the respiratory air flow time series data is collected for feature screening to generate a new data set. The validation set and the test set are selected according to a certain ratio, so that the model can better predict the inhalation-exhalation conversion point. The final prediction accuracy reaches 0.95, and the false trigger rate is less than 3%. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flow chart of the present invention; Figure 2 It is the overall structure diagram of the neural network model of the present invention; Figure 3 It is a detailed diagram of the dimension of the neural network of the present invention; Figure 4 It is a prediction effect diagram of the present invention. Implementation

[0015] The technical solutions in the embodiments of the present invention are clearly and completely described below in conjunction with the accompanying drawings.

[0016] like Figure 1 , the breathing and exhalation conversion method of the ventilator includes the following steps: Step 1) Collect the air flow of the non-invasive ventilator to obtain time series data including the respiratory air flow.

[0017] Step 2) determining the breathing state of the time series data and marking it according to the breathing state.

[0018] Step 3) Perform sliding window sampling and feature extraction on the time series data to obtain new sliding window data and retain the corresponding labels to create a new data set.

[0019] Step 4) Divide the new data set into a training set, a validation set, and a test set.

[0020] Step 5) Predict the inhalation-exhalation transition point by constructing a CNN-LSTM network model.

[0021] Step 6) Setting a threshold for state classification of the ventilator's inhalation-exhalation switching point to control the ventilator to switch between air supply and air release.

[0022] Step 2) of this embodiment specifically includes the following steps: Step 2-1) According to the clinical judgment standard of the respiratory airflow of the non-invasive ventilator, the data is judged and marked as: inspiration, exhalation and intermediate state.

[0023] Step 2-2) Integrate the obtained labeling results to obtain an airflow time series data set containing three breathing states.

[0024] The specific process of step 3) in this embodiment is: Step 3-1) In combination with the actual scenario, set the specific sliding window width and the interval between each sliding window, and move the sliding window in the marked respiratory airflow time series data set to extract four time series features: the average value of the sliding window data, the maximum value of the sliding window data, the minimum value of the sliding window data, and the difference between the maximum and minimum values ​​in the sliding window data.

[0025] Step 3-2) At the same time, the obtained sliding window data are re-labeled according to the labeled data set results: if there is an inhalation transition point in the sliding window data, the four time series feature data states obtained by the corresponding sliding window data are labeled as inhalation transition points; if there is an exhalation transition point in the sliding window data, the four time series feature data states obtained by the corresponding sliding window data are labeled as exhalation transition points; if there is neither an inhalation transition point nor an exhalation transition point in the sliding window data, the four time series feature data states obtained by the corresponding sliding window data are labeled as intermediate transition points.

[0026] Step 3-3) Integrate the above results to obtain a new labeled feature data set.

[0027] The specific implementation method of step 4) of this embodiment is: Step 4-1) Perform Min-Max normalization on the labeled data set; Step 4-2) Split the new dataset into training set, validation set, and test set.

[0028] The steps of step 5) of this embodiment are: Step 5-1) Input the training set into the model for predicting inhalation and exhalation transitions based on feature engineering and CNN-LSTM network; Step 5-2) The model for predicting breathing conversion based on feature engineering and CNN-LSTM network is set to include three convolution layers conv1, LSTM layer, maximum pooling layer max_pooling and dense layer. The number of convolution layers conv1_0, conv1_1, conv1_2 is 1, the length of the kernel size is 2, the width is the width of the input data, and the span is 1; the window size of the maximum pooling sampling layer maxpool is 2; the number of nodes in the LSTM network layer is 8; the output dimension of the dense layer is 1, and the activation function is linear; Step 5-3) Set the loss function of the model to the absolute mean error (L1 loss), set the optimizer to the Adam algorithm, reversely calculate the gradient of each parameter, and continuously update the parameters in the network; Step 5-4) Use the number of training times as the horizontal axis and the values ​​of the loss function for training and validation as the vertical axis to display the loss function values ​​during the training process; Step 5-5) Obtain the trained model according to the above method and visualize the model through the netron visualization tool.

[0029] Step 6) of this embodiment is specifically as follows: input the test set into the trained model, and set the threshold according to the output of the model, including the inhalation state threshold and the exhalation state threshold, so as to achieve the maximum accuracy of the model; when the output of the model reaches the inhalation state threshold, the inhalation conversion point is reached, and the ventilator will cooperate to perform the air supply operation; when the output of the model reaches the exhalation state threshold, the exhalation conversion point is reached, and the ventilator will cooperate to perform the deflation operation; when the output of the model does not meet the inhalation state threshold and the inhalation state threshold, it is an intermediate state, and the ventilator will maintain the current state. In this step, the output of the model based on feature engineering and CNN-LSTM network for inhalation and exhalation conversion prediction can also be compared with the true value according to the needs, and the sample labels of the test set are used as the horizontal coordinates, and the predicted values ​​and the true values ​​are used as the vertical coordinates to display the prediction results, which is more intuitive.

[0030] This embodiment also provides an electronic device, including a memory, a processor, and a computer program, wherein the computer program is stored in the memory and is configured to be executed by the processor to implement the breathing-exhalation conversion method of the ventilator.

[0031] The present invention also provides a computer-readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the breathing-exhalation conversion method of the ventilator.

[0032] The technical means disclosed in the scheme of the present invention are not limited to the technical means disclosed in the above-mentioned implementation mode, but also include technical schemes composed of any combination of the above-mentioned technical features. 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, and these improvements and modifications are also regarded as the protection scope of the present invention.

Claims

1. A method for converting breathing into breathing of a ventilator, characterized in that The steps include: Step 1) collecting the air flow of the non-invasive ventilator to obtain time series data including the respiratory air flow; Step 2) determining the breathing state of the time series data and marking it according to the breathing state; Step 3) Perform sliding window sampling and feature extraction on the time series data to obtain new sliding window data and retain the corresponding labels to create a new data set; Step 4) Divide the new data set into a training set, a validation set, and a test set; Step 5) Predict the inhalation-exhalation transition point by constructing a CNN-LSTM network model; Step 6) Setting a threshold for state classification of the ventilator's inhalation-exhalation switching point to control the ventilator to switch between air supply and air release.

2. The breathing-exhalation conversion method of a ventilator according to claim 1, characterized in that: The step 2) specifically includes the following steps: Step 2-1) According to the clinical judgment standard of non-invasive ventilator respiratory airflow, the data is judged and marked as: inspiration, exhalation and intermediate state; Step 2-2) Integrate the obtained labeling results to obtain an airflow time series data set containing three breathing states.

3. The breathing-exhalation conversion method of a ventilator according to claim 2, characterized in that: The specific process of step 3) is as follows: Step 3-1) In combination with the actual scenario, set the specific sliding window width and the interval between each sliding window, and extract four time series features by moving the sliding window in the marked respiratory airflow time series data set: the average value of the sliding window data, the maximum value of the sliding window data, the minimum value of the sliding window data, and the difference between the maximum and minimum values ​​in the sliding window data; Step 3-2) At the same time, the obtained sliding window data is re-labeled according to the labeled data set results: if there is an inhalation conversion point in the sliding window data, the four time series feature data states obtained by the corresponding sliding window data are labeled as the inhalation conversion point; if there is an exhalation conversion point in the sliding window data, the four time series feature data states obtained by the corresponding sliding window data are labeled as the exhalation conversion point; if there is neither an inhalation conversion point nor an exhalation conversion point in the sliding window data, the four time series feature data states obtained by the corresponding sliding window data are labeled as the intermediate conversion point; Step 3-3) Integrate the above results to obtain a new labeled feature data set.

4. The breathing-exhalation conversion method of a ventilator according to claim 1, characterized in that: The specific implementation method of step 4) is: Step 4-1) Perform Min-Max normalization on the labeled data set; Step 4-2) Split the new dataset into training set, validation set, and test set.

5. The breathing-exhalation conversion method of a ventilator according to claim 1, characterized in that: The steps of step 5) are: Step 5-1) Input the training set into the model for predicting inhalation and exhalation transitions based on feature engineering and CNN-LSTM network; Step 5-2) The model for predicting breathing conversion based on feature engineering and CNN-LSTM network is set to include three convolution layers conv1, LSTM layer, maximum pooling layer max_pooling and dense layer. The number of convolution layers conv1_0, conv1_1 and conv1_2 is 1, the length of the kernel size is 2, the width is the width of the input data, and the span is 1; the window size of the maximum pooling sampling layer maxpool is 2; The number of nodes in the LSTM network layer is 8; the output dimension of the dense layer is 1 and the activation function is linear; Step 5-3) Set the loss function of the model to the absolute mean error, set the optimizer to the Adam algorithm, reversely calculate the gradient of each parameter, and continuously update the parameters in the network; Step 5-4) Use the number of training times as the horizontal axis and the values ​​of the loss function for training and validation as the vertical axis to display the loss function values ​​during the training process; Step 5-5) Obtain the trained model according to the above method and visualize the model through the netron visualization tool.

6. The breathing-exhalation conversion method of a ventilator according to claim 1, characterized in that: The step 6) is specifically as follows: inputting the test set into the trained model, setting the threshold according to the output of the model, including the inhalation state threshold and the exhalation state threshold to achieve a model with the highest accuracy; when the output result of the model reaches the inhalation state threshold, the inhalation switching point is reached, and the ventilator will cooperate to perform the air supply operation; when the output result of the model reaches the exhalation state threshold, the exhalation switching point is reached, and the ventilator will cooperate to perform the deflation operation; when the output results of the model do not meet the inhalation state threshold and the inhalation state threshold, this is an intermediate state, and the ventilator will maintain the current state.

7. An electronic device, characterized in that: The invention comprises a memory, a processor and a computer program, wherein the computer program is stored in the memory and is configured to be executed by the processor to implement the breathing-exhalation conversion method of the ventilator as described in any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and the computer program is executed by a processor to implement the breathing-exhalation conversion method of the ventilator according to any one of claims 1-6.