CN117059250B - Method, system and prediction device for constructing respiratory flow prediction model - Google Patents

Method, system and prediction device for constructing respiratory flow prediction model Download PDF

Info

Publication number
CN117059250B
CN117059250B CN202311162078.9A CN202311162078A CN117059250B CN 117059250 B CN117059250 B CN 117059250B CN 202311162078 A CN202311162078 A CN 202311162078A CN 117059250 B CN117059250 B CN 117059250B
Authority
CN
China
Prior art keywords
data
pressure
data acquisition
respiratory flow
time
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202311162078.9A
Other languages
Chinese (zh)
Other versions
CN117059250A (en
Inventor
邹云霄
潘建
王建良
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Joymed Technology (shanghai) Ltd
Original Assignee
Joymed Technology (shanghai) Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Joymed Technology (shanghai) Ltd filed Critical Joymed Technology (shanghai) Ltd
Priority to CN202311162078.9A priority Critical patent/CN117059250B/en
Publication of CN117059250A publication Critical patent/CN117059250A/en
Application granted granted Critical
Publication of CN117059250B publication Critical patent/CN117059250B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
    • G16H40/20—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the management or administration of healthcare resources or facilities, e.g. managing hospital staff or surgery rooms
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
    • G16H40/60—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices
    • G16H40/63—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices for local operation

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • General Business, Economics & Management (AREA)
  • Business, Economics & Management (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Primary Health Care (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Medical Informatics (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Public Health (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Epidemiology (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Abstract

The invention discloses a method, a system and a device for constructing a respiratory flow prediction model, comprising the following steps: acquiring first pressure data and first pressure difference data of a breathing machine at a plurality of data acquisition moments and simulating first breathing flow of a lung at each data acquisition moment, and preprocessing all the first pressure data, all the first pressure difference data and all the first breathing flow to obtain a sample data set; constructing an artificial intelligent model comprising a plurality of linear perceptron layers and a plurality of activation function layers, and optimizing parameter values of the artificial intelligent model by utilizing a sample data set; and fusing each plurality of adjacent linear perceptron layers in the artificial intelligent model with optimized parameter values into one linear perceptron layer to obtain a respiratory flow prediction model. According to the invention, the nonlinear relation is introduced into the artificial intelligent model, so that the expression capacity of the model is improved, the gradient disappearance problem is relieved, and then the multi-layer linear perceptron layer of the model is simplified, so that the parameter quantity of the model is reduced.

Description

Method, system and prediction device for constructing respiratory flow prediction model
Technical Field
The invention relates to the field of respiratory flow evaluation based on respirators, in particular to a method, a system and a device for constructing a respiratory flow prediction model.
Background
In the process of breathing work of the breathing machine, the breathing machine continuously outputs air through the fan so as to help a user to breathe and ventilate. In the process, the ventilator often collects the pressure difference at the inlet of the internal fan as the basis for predicting/evaluating the respiration related data of the user, and through the prediction or evaluation of the respiration flow, the ventilator can help medical staff evaluate the respiration function or the respiration treatment effect of the user of the ventilator or discover the abnormal respiratory distress or the symptoms of dyspnea as soon as possible, so that the ventilator setting parameters or the respiration treatment scheme can be adjusted in time, and the proper ventilation support or the respiration treatment can be provided for the user of the ventilator.
At present, a method for predicting the respiratory flow of a user based on the internal pressure difference of a breathing machine acquired in real time mainly uses a flow value converted by the pressure difference as a respiratory flow value, and ignores the leakage air quantity of an air leakage port on a mask. In addition, there is a method that the flow of inlet air is calculated by using pressure difference data in the breathing machine, then the flow of the air leakage port on the breathing mask is set to be a fixed value or the average air leakage amount in a period of time is calculated to be used as the fixed air leakage flow of the air leakage port, and finally the difference value between the flow of the inlet air and the fixed air leakage flow of the air leakage port is used as the breathing flow prediction result of the breathing machine user.
Disclosure of Invention
The embodiment of the invention provides a method, a system and a device for constructing a respiratory flow prediction model, which are used for introducing a nonlinear relation into the model, improving the expression capacity of the model, relieving the gradient disappearance problem, simplifying a plurality of linear perception machine layers of the model and reducing the parameter quantity of the model.
In order to solve the above technical problems, an embodiment of the present invention provides a method for constructing a respiratory flow prediction model, including:
Acquiring first pressure data and first pressure difference data of a breathing machine at a plurality of data acquisition moments and simulating first breathing flow of a lung at each data acquisition moment, and preprocessing all the first pressure data, all the first pressure difference data and all the first breathing flow to obtain a sample data set; wherein the first pressure data is a real-time pressure value of a leak on a conduit of the ventilator, and the first pressure difference data is a real-time pressure value of an interior of the ventilator;
building an artificial intelligent model, and optimizing parameter values of the artificial intelligent model by utilizing the sample data set; wherein the artificial intelligent model comprises a plurality of linear perceptron layers and a plurality of activation function layers;
And fusing each plurality of adjacent linear perceptron layers in the artificial intelligent model with optimized parameter values into one linear perceptron layer to obtain a respiratory flow prediction model.
According to the embodiment of the invention, based on a sample data set consisting of real-time pressure values of an air leakage port on a conduit of a breathing machine at a plurality of data acquisition moments, real-time pressure values of the inside of the breathing machine at the plurality of data acquisition moments and first respiratory flow of a simulated lung at the plurality of data acquisition moments, parameter value optimization is carried out on an artificial intelligent model comprising an input layer, a plurality of linear perception machine layers, a plurality of activation function layers and an output layer, and after the parameter value optimization is completed, each plurality of adjacent linear perception machine layers in the current artificial intelligent model are fused into one linear perception machine layer, so that the number of parameters in the model can be reduced, the model can be lighter due to fewer parameter numbers, the calculation complexity of the model is reduced, the occupation of a storage space is reduced, and the prediction speed of the model can be accelerated. In addition, because the complex nonlinear relation cannot be processed by simple linear transformation, an activation function layer is additionally arranged on the basis of a multi-layer linear perceptron layer, the nonlinear relation is introduced into an artificial intelligent model, the expression capacity of the model is improved, the model can better adapt to complex data distribution and class boundaries, the problem that gradient vanishes can be caused by stacking a plurality of linear layers together, the nonlinear gradient can be increased by introducing the activation function layer, and the gradient can be better transferred in the counter propagation process, so that the gradient vanishing problem is relieved.
As a preferred scheme, the artificial intelligent model is built, and the sample data set is utilized to optimize the parameter value of the artificial intelligent model, specifically:
constructing an artificial intelligent model comprising an input layer, a plurality of linear perceptron layers, a plurality of activation function layers and an output layer;
Performing iterative training on the artificial intelligent model by using the sample data set, selecting the last data acquisition time in every N continuous data acquisition times from the sample data set as a time to be predicted corresponding to every N continuous data acquisition times during each iterative processing, then respectively inputting the first pressure data and the first pressure difference data corresponding to every N continuous data acquisition times in the sample data set into the current artificial intelligent model so that the current artificial intelligent model performs classified prediction on input data and outputs second respiratory flow of the time to be predicted corresponding to every N continuous data acquisition times as respiratory flow prediction results of every time to be predicted, then calculating to obtain an error value of the current artificial intelligent model according to the comparison of the second respiratory flow of all the time to be predicted and the first respiratory flow, and adjusting the parameter value of the current artificial intelligent model until the error value of the current artificial intelligent model is smaller than a first threshold value, and stopping iterative intelligent model to finish parameter value optimization of the artificial intelligent model;
The linear perceptron layer is used for carrying out linear transformation on input data, and the activation function layer is used for carrying out nonlinear transformation on the input data, wherein N is more than or equal to 4.
By implementing the preferred scheme of the embodiment of the invention, the artificial intelligent model is iteratively trained by utilizing a sample data set consisting of the real-time pressure values of the air leakage opening on the conduit of the breathing machine at a plurality of data acquisition moments, the real-time pressure difference values of the inside of the breathing machine at a plurality of data acquisition moments and the first respiration flow of the simulated lung at a plurality of data acquisition moments, and the artificial intelligent model is fitted with the actual respiration flow data through the iterative training, so that the model can more accurately predict the current respiration flow level according to the pressure data and the pressure difference data measured in real time, and precious reference information is provided for medical staff. In addition, in the iterative training process, the respiratory flow prediction result output by the model is compared with the first respiratory flow in the sample data set, and the error value of the current artificial intelligent model is calculated, so that the accuracy and reliability of the model are effectively evaluated, the parameter value of the current artificial intelligent model is adjusted, iteration is stopped until the error value of the current artificial intelligent model is smaller than a first threshold value, and therefore the prediction performance of the model is improved and the prediction error of the model is further reduced through the iterative training process.
As a preferred solution, the acquiring first pressure data and first pressure difference data of the ventilator at a plurality of data acquisition moments and simulating a first respiratory flow of the lung at each data acquisition moment specifically includes:
Adjusting the ventilation pressure of the respirator and/or the breathing mode of the simulated lung for a plurality of times, and controlling the simulated lung to perform simulated breathing according to the current breathing mode after each adjustment;
Detecting the pressure of an air leakage port on a conduit of the breathing machine in real time at a plurality of data acquisition moments preset in the process of simulating breathing to obtain first pressure data of the breathing machine at each data acquisition moment, and detecting the pressure difference in the breathing machine in real time through a pressure difference sensor arranged in the breathing machine to obtain first pressure difference data of the breathing machine at each data acquisition moment; the air leakage port on the conduit is used for simulating a small hole on the breathing mask, and the breathing machine is communicated with the conduit through a corrugated pipe;
And detecting flow data of the inhaled air of the simulated lung in real time through a flow sensor arranged in the simulated lung at a plurality of data acquisition moments preset in the process of simulating the breath, so as to obtain the first respiratory flow of the simulated lung at each data acquisition moment.
When the air flows into the breathing mask from the internal fan of the breathing machine or the simulated lung needs to pass through a corrugated pipe, when the air passes through the corrugated pipe, one part of air flows to the outside through the air leakage port, and the other part of air is inhaled by a user wearing the breathing mask or the simulated lung, because the air resistance and certain compliance of the pipeline exist, the real-time pressure data of the air leakage port on the conduit of the breathing machine and the real-time pressure difference data inside the breathing machine are collected and used as the basis of the breathing flow prediction, the breathing flow prediction is not carried out according to the pressure and the pressure difference at the inlet of the fan, and the pressure difference data is prevented from being influenced by the corrugated pipe to change when the corrugated pipe expands or contracts, so that the accuracy of the breathing flow prediction result is improved.
As a preferred solution, the preprocessing is performed on all the first pressure data, all the first pressure difference data and all the first respiratory flow to obtain a sample data set, specifically:
Data cleaning and format unification are respectively and sequentially carried out on all the first pressure data, all the first pressure difference data and all the first respiratory flow;
Screening an initial time from all the data acquisition time according to all the first pressure data, all the first pressure difference data and all the first respiratory flow which are subjected to data cleaning and uniform in format; wherein a difference between the first pressure data at the initial time and the first pressure data at a last data acquisition time of the initial time is greater than a second threshold, a difference between the first pressure difference data at the initial time and the first pressure difference data at a last data acquisition time of the initial time is greater than a third threshold, a difference between the first respiratory flow at the initial time and the first respiratory flow at a last data acquisition time of the initial time is greater than a fourth threshold, the second threshold, the third threshold and the fourth threshold are all greater than zero, the first respiratory flow at the initial time is greater than zero and the first respiratory flow at a last data acquisition time of the initial time is less than or equal to zero;
And constructing the sample data set according to the first pressure data, the first pressure difference data and the first respiration flow at the initial moment and the first pressure data, the first pressure difference data and the first respiration flow at all the data acquisition moments after the initial moment.
By implementing the preferred scheme of the embodiment of the invention, the collected pressure data and pressure difference data are sequentially subjected to data cleaning and format unification so as to remove invalid data and abnormal data and ensure consistency and comparability of model input data. Then, when the first pressure difference data and the first pressure difference data change and the first respiration flow just crosses the zero point, taking the current moment as the initial moment, and starting from the initial moment, constructing a sample data set according to the first pressure difference data, the first pressure difference data and the first respiration flow at the initial moment and the first pressure difference data, the first pressure difference data and the first respiration flow at all data acquisition moments after the initial moment, so as to screen out invalid sample data under the conditions that the first pressure difference data or the first pressure difference data do not change obviously and the like, thereby improving the model training effect.
In order to solve the same technical problem, the embodiment of the invention further provides a system for constructing a respiratory flow prediction model, which comprises:
The data acquisition module is used for acquiring first pressure data and first pressure difference data of the breathing machine at a plurality of data acquisition moments and simulating first breathing flow of the lung at each data acquisition moment;
The preprocessing module is used for preprocessing all the first pressure data, all the first pressure difference data and all the first respiratory flow to obtain a sample data set; wherein the first pressure data is a real-time pressure value of a leak on a conduit of the ventilator, and the first pressure difference data is a real-time pressure value of an interior of the ventilator;
the parameter optimization module is used for constructing an artificial intelligent model and optimizing the parameter value of the artificial intelligent model by utilizing the sample data set; wherein the artificial intelligent model comprises a plurality of linear perceptron layers and a plurality of activation function layers;
and the fusion module is used for fusing each plurality of adjacent linear perception machine layers in the artificial intelligent model with optimized parameter values into one linear perception machine layer to obtain a respiratory flow prediction model.
Preferably, the parameter optimization module specifically includes:
the model building unit is used for building an artificial intelligent model comprising an input layer, a plurality of linear perceptron layers, a plurality of activation function layers and an output layer; the linear perceptron layer is used for carrying out linear transformation on input data, and the activation function layer is used for carrying out nonlinear transformation on the input data;
The iterative training unit is used for carrying out iterative training on the artificial intelligent model by utilizing the sample data set, selecting the last data acquisition time in every N continuous data acquisition times from the sample data set as a time to be predicted corresponding to every N continuous data acquisition times when carrying out iterative processing, then respectively inputting the first pressure data and the first pressure difference data corresponding to every N continuous data acquisition times in the sample data set into the current artificial intelligent model so as to enable the current artificial intelligent model to carry out classified prediction on input data and output the second respiratory flow at the time to be predicted corresponding to every N continuous data acquisition times as a respiratory flow prediction result at each time to be predicted, then calculating to obtain an error value of the current artificial intelligent model according to the comparison of the second respiratory flow at all the time to be predicted and the first respiratory flow, and adjusting the parameter value of the current artificial intelligent model until the error value of the current artificial intelligent model is smaller than a threshold value, and stopping iteration until the parameter value of the current artificial intelligent model is smaller than a threshold value, so as to finish parameter value optimization on the artificial intelligent model; wherein N is more than or equal to 4.
As a preferred solution, the data acquisition module specifically includes:
The adjusting unit is used for adjusting the ventilation pressure of the respirator and/or the breathing mode of the simulated lung for a plurality of times, and controlling the simulated lung to perform simulated breathing according to the current breathing mode after each adjustment;
The first acquisition unit is used for detecting the pressure intensity of an air leakage port on a guide pipe of the breathing machine in real time at a plurality of data acquisition moments preset in the breathing simulation process to obtain first pressure intensity data of the breathing machine at each data acquisition moment, and meanwhile, detecting the pressure difference inside the breathing machine in real time through a pressure difference sensor arranged inside the breathing machine to obtain first pressure difference data of the breathing machine at each data acquisition moment; the air leakage port on the conduit is used for simulating a small hole on the breathing mask, and the breathing machine is communicated with the conduit through a corrugated pipe;
The second acquisition unit is used for detecting flow data of the inhaled air of the simulated lung in real time through a flow sensor arranged in the simulated lung at a plurality of data acquisition moments preset in the process of simulating the breath, and obtaining the first respiratory flow of the simulated lung at each data acquisition moment.
As a preferred solution, the preprocessing module specifically includes:
The first preprocessing unit is used for sequentially carrying out data cleaning and format unification on all the first pressure data, all the first pressure difference data and all the first respiratory flow respectively;
The second preprocessing unit is used for screening initial time from all the data acquisition time according to all the first pressure data, all the first pressure difference data and all the first respiratory flow which are subjected to data cleaning and uniform format; wherein a difference between the first pressure data at the initial time and the first pressure data at a last data acquisition time of the initial time is greater than a second threshold, a difference between the first pressure difference data at the initial time and the first pressure difference data at a last data acquisition time of the initial time is greater than a third threshold, a difference between the first respiratory flow at the initial time and the first respiratory flow at a last data acquisition time of the initial time is greater than a fourth threshold, the second threshold, the third threshold and the fourth threshold are all greater than zero, the first respiratory flow at the initial time is greater than zero and the first respiratory flow at a last data acquisition time of the initial time is less than or equal to zero; and constructing the sample data set according to the first pressure data, the first pressure difference data and the first respiration flow at the initial moment and the first pressure data, the first pressure difference data and the first respiration flow at all the data acquisition moments after the initial moment.
In order to solve the same technical problem, an embodiment of the present invention further provides a respiratory flow prediction apparatus, including:
the acquisition module is used for acquiring second pressure data and second pressure difference data of the breathing machine at a plurality of data acquisition moments in real time; the breathing machine is communicated with the breathing mask with the small holes through a corrugated pipe;
The prediction module is used for inputting all the second pressure data and all the second pressure difference data into a respiratory flow prediction model, and outputting corresponding third respiratory flow as a respiratory flow prediction result at the moment to be predicted; the time to be predicted is the last data acquisition time in the data acquisition time corresponding to all the second pressure data of the respiratory flow prediction model, and the respiratory flow prediction model is constructed by using the method for constructing the respiratory flow prediction model.
By implementing the embodiment of the invention, the second pressure data and the second pressure difference data of the breathing machine at a plurality of data acquisition moments are obtained in real time, the second pressure data and the second pressure difference data at the plurality of data acquisition moments are input into the breathing flow prediction model constructed by the method for constructing the breathing flow prediction model in the embodiment of the invention, the real-time monitoring of the breathing flow and the related state of the user using the breathing machine is realized, the dependence on the experience of medical staff is reduced, and the real-time and accurate breathing flow prediction result is provided for the medical staff, so that the medical staff can be assisted in timely and adaptively adjusting the setting parameters of the breathing machine, and the experience and the treatment effect of the user are improved.
Preferably, the acquiring module specifically includes:
the first detection unit is used for detecting the pressure intensity of the small holes on the breathing mask in real time at a plurality of data acquisition moments to obtain the second pressure intensity data of the breathing machine at each data acquisition moment;
The second detection unit is used for detecting the internal pressure difference of the breathing machine in real time through a pressure difference sensor arranged in the breathing machine at a plurality of data acquisition moments to obtain second pressure difference data of the breathing machine at each data acquisition moment.
Drawings
Fig. 1: a schematic flow chart of a method for constructing a respiratory flow prediction model according to the first embodiment of the present invention;
fig. 2: a schematic structural diagram of an artificial intelligence model according to a first embodiment of the present invention;
Fig. 3: a schematic structural diagram of a system for constructing a respiratory flow prediction model according to a first embodiment of the present invention;
fig. 4: a schematic structural diagram of a respiratory flow prediction device is provided in a second embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Embodiment one:
Referring to fig. 1, a method for constructing a respiratory flow prediction model according to an embodiment of the present invention includes steps S1 to S4, where each step is specifically as follows:
Step S1, first pressure data and first pressure difference data of the breathing machine at a plurality of data acquisition moments are obtained, and first breathing flow of the lung at each data acquisition moment is simulated.
The first pressure data is a real-time pressure value of a gas leakage port on a conduit of the breathing machine, and the first pressure difference data is a real-time pressure value of the inside of the breathing machine.
The ventilator transmits external air to the corrugated pipe through the internal fan, and in the transmission process, the pressure difference sensor and the pressure sensor monitor pressure difference and pressure data when the air flows; when air passes through the corrugated pipe, the air passes through a conduit with an air leakage port (the conduit is used for simulating a small hole on a breathing mask), the air leakage port is connected with the external atmosphere, when air passes through the conduit, part of air can flow out to the outside through the air leakage port, and the other part of air can be inhaled by a simulated lung (or a user wearing the breathing mask), at the moment, a flow sensor in the simulated lung can record flow data of air inhaled by the simulated lung (the flow data is positive when the air is inhaled and the flow data is negative when the air is exhaled).
In this embodiment, the model of the ventilator is a cart noninvasive bi-level ventilator, the bellows is a 1.5m medical bellows, the conduit with the leakage orifice is a plastic conduit with a pipe diameter of 1cm, a length of 4cm, and a leakage orifice diameter of 2mm, and the simulated lung is an ASL5000 simulated lung for simulating the breathing state of different users.
Preferably, step S1 includes steps S11 to S13, and each step is specifically as follows:
step S11, the ventilation pressure of the respirator and/or the breathing mode of the simulated lung are adjusted for a plurality of times, and the simulated lung is controlled to perform simulated breathing according to the current breathing mode after each adjustment.
In this embodiment, the ventilation pressure of the ventilator may be any one of 4cm·h2o, 6cm·h2o, 8cm·h2o, 10cm·h2o, and 12cm·h2o, and the breathing pattern of the simulated lung may be any one of 18 breathing patterns preset; wherein the preset 18 breathing modes are respectively adult_normal、adult_ARDS、adult_apnea、adult_asthma、adult_CF、adult_chbronchitis、adult_COPD、adult_COPD_unssisted、adult_emphysema、adult_Normal_unassisted、adolescent_normal、neonate_normal、neonate_BPD、neonate_RDS、Ped_10kg_normal、pediatric5yo_normal、pediatric12yo_normal、Toddler_normal.
Step S12, detecting the pressure of an air leakage port on a conduit of the breathing machine in real time at a plurality of preset data acquisition moments in the breathing simulation process to obtain first pressure data of the breathing machine at each data acquisition moment, and detecting the pressure difference in the breathing machine in real time through a pressure difference sensor arranged in the breathing machine to obtain first pressure difference data of the breathing machine at each data acquisition moment.
Wherein, the gas leakage mouth on the pipe is used for simulating the aperture on the breathing mask, and the breathing machine is communicated with the pipe through the bellows.
And step S13, detecting flow data of the inhaled air of the simulated lung in real time by a flow sensor arranged in the simulated lung at a plurality of preset data acquisition moments in the process of simulating the breath, and obtaining the first respiratory flow of the simulated lung at each data acquisition moment.
In this embodiment, when the ventilator is kept operating at different ventilation pressures, the simulated lung is controlled to perform simulated breathing according to different breathing modes, and during the simulated breathing, pressure data at the leak port on the conduit of the ventilator (as first pressure data of the ventilator at the respective data acquisition times), pressure difference data inside the ventilator (as first pressure difference data of the ventilator at the respective data acquisition times), and flow data of the "chamber" in the simulated lung (as first breathing flow of the simulated lung at the respective data acquisition times) are acquired at the respective data acquisition times from the start to the duration.
Step S2, preprocessing all the first pressure data, all the first pressure difference data and all the first respiratory flow to obtain a sample data set.
Preferably, step S2 includes steps S21 to S23, and each step is specifically as follows:
step S21, data cleaning and format unification are sequentially carried out on all the first pressure data, all the first pressure difference data and all the first respiratory flow respectively.
It should be noted that, because there may be other characters, such as "section", in addition to the acquired pressure difference and pressure data when acquiring data from the ventilator through the serial port, and there may be a missing value in the acquired data, the acquired data may be cleaned, so that the invalid value, the abnormal value and the missing value in the acquired data may be cleaned. After the data cleaning is completed, the first pressure data and the first pressure difference data may still be characters, for example, the data acquired in this embodiment is [ Diff:78, P:488 ], where Diff represents the first pressure difference data, P represents the first pressure difference data, and at this time, the pressure difference and the pressure need to be subjected to data segmentation and ordered arrangement, so as to obtain "pressure difference data" respectively: 78, and pressure data: 488 "format unifies the results.
Step S22, screening out initial time from all data acquisition time according to all first pressure data, all first pressure difference data and all first respiration flow which are subjected to data cleaning and uniform in format; the difference value between the first pressure data at the initial time and the first pressure data at the last data acquisition time at the initial time is larger than a second threshold value, the difference value between the first pressure difference data at the initial time and the first pressure difference data at the last data acquisition time at the initial time is larger than a third threshold value, the difference value between the first respiration flow at the initial time and the first respiration flow at the last data acquisition time at the initial time is larger than a fourth threshold value, the second threshold value, the third threshold value and the fourth threshold value are all larger than zero, the first respiration flow at the initial time is larger than zero, and the first respiration flow at the last data acquisition time at the initial time is smaller than or equal to zero.
Step S23, a sample data set is constructed according to the first pressure data, the first pressure difference data and the first respiration flow at the initial moment and the first pressure data, the first pressure difference data and the first respiration flow at all data acquisition moments after the initial moment.
It should be noted that, since the sampling interval of the ventilator is 20ms and the sampling interval of the simulated lung is 2ms, and the synchronicity of the data of the ventilator and the simulated lung cannot be determined (that is, the time interval between the differential pressure data and the simulated lung flow output at the same time cannot be determined), in order to synchronize the data of the ventilator and the simulated lung, the initial starting time point is compared, for example, at a certain time point, the ventilator starts to operate and the simulated lung starts to simulate respiration, the first pressure data and the first differential pressure data start to change significantly (that is, the second threshold value and the third threshold value are both greater than a certain preset positive value), the time point is taken as the initial time point when the first respiratory flow just starts to cross the zero point, then the first respiratory flow of the simulated lung is acquired at intervals of 20ms from the initial time point, and the acquired first respiratory flow and the first pressure data and the first differential pressure data are in one-to-one correspondence according to the data acquisition time.
And S3, building an artificial intelligent model, and optimizing parameter values of the artificial intelligent model by using a sample data set.
The artificial intelligent model comprises a plurality of linear perceptron layers and a plurality of activation function layers, and can be built by utilizing pytorch libraries.
Preferably, step S3 includes steps S31 to S32, and each step is specifically as follows:
and S31, constructing an artificial intelligent model comprising an input layer, a plurality of linear perceptron layers, a plurality of activation function layers and an output layer.
The linear perceptron layer is used for carrying out linear transformation on input data, and the activation function layer is used for carrying out nonlinear transformation on the input data.
In this embodiment, referring to fig. 2, the input of the artificial intelligence model is 8 floating point numbers, which respectively represent the first differential pressure data and the first pressure data at 4 continuous data acquisition moments, the middle fc1 (with dimension 16), fc2 (with dimension 16), fc3 (with dimension 32), fc4 (with dimension 32), fc5 (with dimension 8), and fc6 (with dimension 1) layers represent the linear perceptron layer (or the fully connected layer) in the artificial intelligence model, and relu represents an activation function, which transforms the output of the upper neural network layer, specifically, the number smaller than zero is directly set to zero, while the number greater than or equal to zero is unchanged; the fc1 and fc2 layers output 16 floating point numbers, fc3 and fc4 output 32 floating point numbers, fc5 outputs 8 floating point numbers, and f6 outputs 1 floating point number, and finally outputs the floating point numbers as user respiratory flow data. Wherein every 20ms is taken as a data acquisition time.
As an example, the input of the artificial intelligence model is x= [ Δp1, P1, Δp2, P2, Δp3, P3, Δp4, P4]; wherein DeltaPn is first pressure difference data, pn is first pressure difference data, n is serial number of the first pressure difference data/the first pressure difference data, X is a one-dimensional array formed by the first pressure difference data and the first pressure difference data at four continuous data acquisition moments, and the first pressure difference data/the first pressure difference data in the array X are ordered according to the acquisition sequence. When passing through the linear sensing layer fc1, performing calculation processing of y=x×w+b, where W is a matrix with a size of 8×16, B is a matrix with a size of 1×16, and Y is an array with a size of 1×16; and the model outputs a number Y after passing through a plurality of linear perception layers and an activation function, wherein the number is the respiratory flow prediction result of the user at the last data acquisition time.
And S32, carrying out iterative training on the artificial intelligent model by using a sample data set, selecting the last data acquisition time in every N continuous data acquisition times from the sample data set as a time to be predicted corresponding to every N continuous data acquisition times during each iterative processing, then respectively inputting first pressure data and first pressure difference data corresponding to every N continuous data acquisition times in the sample data set into the current artificial intelligent model so as to enable the current artificial intelligent model to carry out classified prediction on input data and output second respiratory flow of the time to be predicted corresponding to every N continuous data acquisition times as respiratory flow prediction results of every time to be predicted, then calculating to obtain an error value of the current artificial intelligent model according to comparison of the second respiratory flow of all the time to be predicted and the first respiratory flow, and adjusting the parameter value of the current artificial intelligent model until the error value of the current artificial intelligent model is smaller than a first threshold value, and stopping iteration to finish parameter value optimization of the artificial intelligent model. Wherein N is more than or equal to 4.
In this embodiment, N takes a value of 4. That is, it is indicated that the input data of the artificial intelligence model and the respiratory flow prediction model in the present embodiment should be pressure difference data and pressure data at 4 consecutive data acquisition times, for a total of 8 data.
It should be noted that, by monitoring the error value of the model in real time, the working state of the ventilator and the respiration condition of the patient can be estimated. If the error value exceeds the threshold value, this may mean that the ventilator is not set properly or that there is an abnormal situation, requiring further inspection and adjustment. Thus, monitoring the error value may help the healthcare worker to find problems in time and take corresponding action.
And S4, fusing each plurality of adjacent linear perceptron layers in the artificial intelligent model with optimized parameter values into one linear perceptron layer to obtain a respiratory flow prediction model.
In this embodiment, the multi-layer linear perceptron layer is used to perform continuous multiplication matrix operation on the input data. The parameter matrices are fused in advance through matrix operation, so that two adjacent linear perceptron layers in reference to fig. 2 are fused into one linear perceptron layer.
As an example, two adjacent linear perceptron layers have a calculation formula of (x×wa+ba) ×wb+bb, where X is input data, wa (size is 8×16), ba (size is 1×16) is a parameter matrix of the first layer fc1, wb (size is 16×16), bb (size is 16) is a parameter matrix of the second layer fc2, and the parameters of the two linear perceptron layers are 8×16+16+16×16+16=416. After matrix operation, the two linear perceptron layers are fused into one linear perceptron layer, and the calculation formula of the fused linear perceptron layer is X (Wa) Wb) + (Ba) wb+bb, wherein the size of (Wa) Wb is 8×16, the size of (Ba) wb+bb is 16, and the parameter number is 8×16+16=144. Through the operation, the parameter optimization of the artificial intelligent model is realized, so that the artificial intelligent model is lighter and more efficient, and is more suitable for being used in environments with limited resources such as embedded systems.
Referring to fig. 3, a schematic structural diagram of a respiratory flow prediction model building system provided by an embodiment of the present invention includes a data acquisition module M1, a preprocessing module M2, a parameter optimization module M3, and a fusion module M4, where each module is specifically as follows:
the data acquisition module M1 is used for acquiring first pressure data and first pressure difference data of the breathing machine at a plurality of data acquisition moments and simulating first breathing flow of the lung at each data acquisition moment;
the preprocessing module M2 is used for preprocessing all the first pressure data, all the first pressure difference data and all the first respiratory flow to obtain a sample data set; the first pressure data is a real-time pressure value of a gas leakage port on a conduit of the breathing machine, and the first pressure difference data is a real-time pressure value of the inside of the breathing machine;
the parameter optimization module M3 is used for constructing an artificial intelligent model and optimizing the parameter value of the artificial intelligent model by utilizing the sample data set; the artificial intelligent model comprises a plurality of linear perceptron layers and a plurality of activation function layers;
And the fusion module M4 is used for fusing each plurality of adjacent linear perceptron layers in the artificial intelligent model with optimized parameter values into one linear perceptron layer to obtain a respiratory flow prediction model.
As a preferred solution, the parameter optimization module M3 specifically includes a model building unit 31 and an iterative training unit 32, where each unit specifically includes:
A model building unit 31 for building an artificial intelligent model including an input layer, a plurality of linear perceptron layers, a plurality of activation function layers, and an output layer; the linear perceptron layer is used for carrying out linear transformation on input data, and the activation function layer is used for carrying out nonlinear transformation on the input data;
The iterative training unit 32 is configured to perform iterative training on the artificial intelligent model by using a sample data set, select, from the sample data set, a last data acquisition time of every N consecutive data acquisition times as a time to be predicted corresponding to every N consecutive data acquisition times, and then input first pressure data and first pressure difference data corresponding to every N consecutive data acquisition times in the sample data set to the current artificial intelligent model respectively, so that the current artificial intelligent model performs classification prediction on the input data and outputs a second respiratory flow of the time to be predicted corresponding to every N consecutive data acquisition times as a respiratory flow prediction result of each time to be predicted, calculate to obtain an error value of the current artificial intelligent model according to a comparison of the second respiratory flow of all the time to be predicted and the first respiratory flow, and adjust the parameter value of the current artificial intelligent model until the error value of the current artificial intelligent model is smaller than a first threshold value, so as to complete optimization of the parameter value of the artificial intelligent model; wherein N is more than or equal to 4.
As a preferred solution, the data acquisition module M1 specifically includes an adjustment unit 11, a first acquisition unit 12, and a second acquisition unit 13, where each unit specifically includes:
An adjusting unit 11, configured to adjust the ventilation pressure of the ventilator and/or the breathing pattern of the simulated lung for a plurality of times, and control the simulated lung to perform simulated breathing according to the current breathing pattern after each adjustment;
The first collecting unit 12 is configured to detect, in real time, a pressure of a leakage port on a conduit of the ventilator at a plurality of data collection moments preset in a respiratory simulating process, to obtain first pressure data of the ventilator at each data collection moment, and simultaneously, detect, in real time, a pressure difference inside the ventilator through a pressure difference sensor disposed inside the ventilator, to obtain first pressure difference data of the ventilator at each data collection moment; the air leakage port on the conduit is used for simulating a small hole on the breathing mask, and the breathing machine is communicated with the conduit through the corrugated pipe;
the second collecting unit 13 is configured to detect, in real time, flow data of air inhaled by the simulated lung through a flow sensor disposed in the simulated lung at a plurality of data collecting moments preset in the process of simulating respiration, so as to obtain a first respiratory flow of the simulated lung at each data collecting moment.
As a preferred solution, the preprocessing module M2 specifically includes a first preprocessing unit 21 and a second preprocessing unit 22, where each unit specifically includes:
A first preprocessing unit 21, configured to sequentially perform data cleansing and format unification on all the first pressure data, all the first differential pressure data, and all the first respiratory flow, respectively;
the second preprocessing unit 22 is configured to screen out an initial time from all data acquisition times according to all first pressure data, all first pressure difference data and all first respiratory flow rate which are subjected to data cleaning and form unification; the difference value between the first pressure data at the initial time and the first pressure data at the last data acquisition time at the initial time is larger than a second threshold value, the difference value between the first pressure difference data at the initial time and the first pressure difference data at the last data acquisition time at the initial time is larger than a third threshold value, the difference value between the first respiration flow at the initial time and the first respiration flow at the last data acquisition time at the initial time is larger than a fourth threshold value, the second threshold value, the third threshold value and the fourth threshold value are all larger than zero, the first respiration flow at the initial time is larger than zero, and the first respiration flow at the last data acquisition time at the initial time is smaller than or equal to zero; and constructing a sample data set according to the first pressure data, the first pressure difference data and the first respiratory flow at the initial moment and the first pressure data, the first pressure difference data and the first respiratory flow at all data acquisition moments after the initial moment.
It will be clear to those skilled in the art that, for convenience and brevity of description, reference may be made to the corresponding process in the foregoing method embodiment for the specific working process of the above-described system, which is not described herein again.
Compared with the prior art, the embodiment of the invention has the following beneficial effects:
The invention provides a method and a system for constructing a respiratory flow prediction model, which are used for optimizing parameter values of an artificial intelligent model comprising an input layer, a plurality of linear perceptron layers, a plurality of activation function layers and an output layer based on a sample data set consisting of real-time pressure values of an air leakage port on a conduit of a breathing machine at a plurality of data acquisition moments, real-time pressure values of the inside of the breathing machine at the plurality of data acquisition moments and first respiratory flow of a simulated lung at the plurality of data acquisition moments, and fusing each plurality of adjacent linear perceptron layers in the current artificial intelligent model into one linear perceptron layer after the parameter value optimization is completed, so that the number of parameters in the model can be reduced, the model can be lighter, the calculation complexity of the model is reduced, the occupation of a storage space is reduced, and the prediction speed of the model can be accelerated. In addition, because the complex nonlinear relation cannot be processed by simple linear transformation, an activation function layer is additionally arranged on the basis of a multi-layer linear perceptron layer, the nonlinear relation is introduced into an artificial intelligent model, the expression capacity of the model is improved, the model can better adapt to complex data distribution and class boundaries, the problem that gradient vanishes can be caused by stacking a plurality of linear layers together, the nonlinear gradient can be increased by introducing the activation function layer, and the gradient can be better transferred in the counter propagation process, so that the gradient vanishing problem is relieved.
Embodiment two:
Referring to fig. 4, a schematic structural diagram of a respiratory flow prediction device provided by an embodiment of the present invention is shown, where the prediction device includes an acquisition module M5 and a prediction module M6, and each module is specifically as follows:
The acquisition module M5 is used for acquiring second pressure data and second pressure difference data of the breathing machine at a plurality of data acquisition moments in real time; the breathing machine is communicated with the breathing mask with the small holes through a corrugated pipe;
The prediction module M6 is configured to input all the second pressure data and all the second pressure difference data to the respiratory flow prediction model, and output a corresponding third respiratory flow as a respiratory flow prediction result at the moment to be predicted; the time to be predicted is the last data acquisition time in the data acquisition time corresponding to all the second pressure data of the input respiratory flow prediction model, and the respiratory flow prediction model is constructed by using the construction method of the respiratory flow prediction model in the first embodiment.
As a preferred solution, the acquiring module M5 specifically includes a first detecting unit 51 and a second detecting unit 52, where each unit specifically includes:
The first detecting unit 51 is configured to detect pressure of a small hole on the breathing mask in real time at a plurality of data acquisition moments, so as to obtain second pressure data of the breathing machine at each data acquisition moment;
The second detecting unit 52 is configured to detect, in real time, an internal pressure difference of the ventilator at a plurality of data acquisition moments by using a pressure difference sensor disposed inside the ventilator, so as to obtain second pressure difference data of the ventilator at each data acquisition moment.
Compared with the prior art, the embodiment of the invention has the following beneficial effects:
The invention provides a respiratory flow prediction device, which is used for acquiring second pressure data and second pressure difference data of a breathing machine at a plurality of data acquisition moments in real time, inputting the second pressure data and the second pressure difference data at the plurality of data acquisition moments into a respiratory flow prediction model constructed by using the respiratory flow prediction model construction method according to the first embodiment of the invention, realizing real-time monitoring of respiratory flow and related states of a user using the breathing machine, reducing dependence on experience of medical staff, providing real-time and accurate respiratory flow prediction results for the medical staff, thereby assisting the medical staff in timely and adaptively adjusting setting parameters of the breathing machine, and improving experience and treatment effects of the user.
The foregoing embodiments have been provided for the purpose of illustrating the general principles of the present invention, and are not to be construed as limiting the scope of the invention. It should be noted that any modifications, equivalent substitutions, improvements, etc. made by those skilled in the art without departing from the spirit and principles of the present invention are intended to be included in the scope of the present invention.

Claims (8)

1. A method of constructing a respiratory flow prediction model, comprising:
Acquiring first pressure data and first pressure difference data of a breathing machine at a plurality of data acquisition moments and simulating first breathing flow of a lung at each data acquisition moment, and preprocessing all the first pressure data, all the first pressure difference data and all the first breathing flow to obtain a sample data set; wherein the first pressure data is a real-time pressure value of a leak on a conduit of the ventilator, and the first pressure difference data is a real-time pressure value of an interior of the ventilator;
building an artificial intelligent model, and optimizing parameter values of the artificial intelligent model by utilizing the sample data set; wherein the artificial intelligent model comprises a plurality of linear perceptron layers and a plurality of activation function layers;
fusing each plurality of adjacent linear perceptron layers in the artificial intelligent model with optimized parameter values into a linear perceptron layer to obtain a respiratory flow prediction model;
The method comprises the steps of constructing an artificial intelligent model, optimizing parameter values of the artificial intelligent model by utilizing the sample data set, and specifically comprises the following steps:
constructing an artificial intelligent model comprising an input layer, a plurality of linear perceptron layers, a plurality of activation function layers and an output layer;
Performing iterative training on the artificial intelligent model by using the sample data set, selecting the last data acquisition time in every N continuous data acquisition times from the sample data set as a time to be predicted corresponding to every N continuous data acquisition times during each iterative processing, then respectively inputting the first pressure data and the first pressure difference data corresponding to every N continuous data acquisition times in the sample data set into the current artificial intelligent model so that the current artificial intelligent model performs classified prediction on input data and outputs second respiratory flow of the time to be predicted corresponding to every N continuous data acquisition times as respiratory flow prediction results of every time to be predicted, then calculating to obtain an error value of the current artificial intelligent model according to the comparison of the second respiratory flow of all the time to be predicted and the first respiratory flow, and adjusting the parameter value of the current artificial intelligent model until the error value of the current artificial intelligent model is smaller than a first threshold value, and stopping iterative intelligent model to finish parameter value optimization of the artificial intelligent model;
The linear perceptron layer is used for carrying out linear transformation on input data, and the activation function layer is used for carrying out nonlinear transformation on the input data, wherein N is more than or equal to 4.
2. The method for constructing a respiratory flow prediction model according to claim 1, wherein the acquiring the first pressure data and the first pressure difference data of the ventilator at a plurality of data acquisition moments and simulating the first respiratory flow of the lung at each data acquisition moment specifically comprises:
Adjusting the ventilation pressure of the respirator and/or the breathing mode of the simulated lung for a plurality of times, and controlling the simulated lung to perform simulated breathing according to the current breathing mode after each adjustment;
Detecting the pressure of an air leakage port on a conduit of the breathing machine in real time at a plurality of data acquisition moments preset in the process of simulating breathing to obtain first pressure data of the breathing machine at each data acquisition moment, and detecting the pressure difference in the breathing machine in real time through a pressure difference sensor arranged in the breathing machine to obtain first pressure difference data of the breathing machine at each data acquisition moment; the air leakage port on the conduit is used for simulating a small hole on the breathing mask, and the breathing machine is communicated with the conduit through a corrugated pipe;
And detecting flow data of the inhaled air of the simulated lung in real time through a flow sensor arranged in the simulated lung at a plurality of data acquisition moments preset in the process of simulating the breath, so as to obtain the first respiratory flow of the simulated lung at each data acquisition moment.
3. The method for constructing a respiratory flow prediction model according to claim 1, wherein the preprocessing is performed on all the first pressure data, all the first differential pressure data and all the first respiratory flow to obtain a sample data set, specifically:
Data cleaning and format unification are respectively and sequentially carried out on all the first pressure data, all the first pressure difference data and all the first respiratory flow;
Screening an initial time from all the data acquisition time according to all the first pressure data, all the first pressure difference data and all the first respiratory flow which are subjected to data cleaning and uniform in format; wherein a difference between the first pressure data at the initial time and the first pressure data at a last data acquisition time of the initial time is greater than a second threshold, a difference between the first pressure difference data at the initial time and the first pressure difference data at a last data acquisition time of the initial time is greater than a third threshold, a difference between the first respiratory flow at the initial time and the first respiratory flow at a last data acquisition time of the initial time is greater than a fourth threshold, the second threshold, the third threshold and the fourth threshold are all greater than zero, the first respiratory flow at the initial time is greater than zero and the first respiratory flow at a last data acquisition time of the initial time is less than or equal to zero;
And constructing the sample data set according to the first pressure data, the first pressure difference data and the first respiration flow at the initial moment and the first pressure data, the first pressure difference data and the first respiration flow at all the data acquisition moments after the initial moment.
4. A system for constructing a respiratory flow prediction model, comprising:
The data acquisition module is used for acquiring first pressure data and first pressure difference data of the breathing machine at a plurality of data acquisition moments and simulating first breathing flow of the lung at each data acquisition moment;
The preprocessing module is used for preprocessing all the first pressure data, all the first pressure difference data and all the first respiratory flow to obtain a sample data set; wherein the first pressure data is a real-time pressure value of a leak on a conduit of the ventilator, and the first pressure difference data is a real-time pressure value of an interior of the ventilator;
the parameter optimization module is used for constructing an artificial intelligent model and optimizing the parameter value of the artificial intelligent model by utilizing the sample data set; wherein the artificial intelligent model comprises a plurality of linear perceptron layers and a plurality of activation function layers;
the fusion module is used for fusing each plurality of adjacent linear perceptron layers in the artificial intelligent model with optimized parameter values into one linear perceptron layer to obtain a respiratory flow prediction model;
the parameter optimization module specifically comprises:
the model building unit is used for building an artificial intelligent model comprising an input layer, a plurality of linear perceptron layers, a plurality of activation function layers and an output layer; the linear perceptron layer is used for carrying out linear transformation on input data, and the activation function layer is used for carrying out nonlinear transformation on the input data;
The iterative training unit is used for carrying out iterative training on the artificial intelligent model by utilizing the sample data set, selecting the last data acquisition time in every N continuous data acquisition times from the sample data set as a time to be predicted corresponding to every N continuous data acquisition times when carrying out iterative processing, then respectively inputting the first pressure data and the first pressure difference data corresponding to every N continuous data acquisition times in the sample data set into the current artificial intelligent model so as to enable the current artificial intelligent model to carry out classified prediction on input data and output the second respiratory flow at the time to be predicted corresponding to every N continuous data acquisition times as a respiratory flow prediction result at each time to be predicted, then calculating to obtain an error value of the current artificial intelligent model according to the comparison of the second respiratory flow at all the time to be predicted and the first respiratory flow, and adjusting the parameter value of the current artificial intelligent model until the error value of the current artificial intelligent model is smaller than a threshold value, and stopping iteration until the parameter value of the current artificial intelligent model is smaller than a threshold value, so as to finish parameter value optimization on the artificial intelligent model; wherein N is more than or equal to 4.
5. The system for constructing a respiratory flow prediction model according to claim 4, wherein the data acquisition module specifically comprises:
The adjusting unit is used for adjusting the ventilation pressure of the respirator and/or the breathing mode of the simulated lung for a plurality of times, and controlling the simulated lung to perform simulated breathing according to the current breathing mode after each adjustment;
The first acquisition unit is used for detecting the pressure intensity of an air leakage port on a guide pipe of the breathing machine in real time at a plurality of data acquisition moments preset in the breathing simulation process to obtain first pressure intensity data of the breathing machine at each data acquisition moment, and meanwhile, detecting the pressure difference inside the breathing machine in real time through a pressure difference sensor arranged inside the breathing machine to obtain first pressure difference data of the breathing machine at each data acquisition moment; the air leakage port on the conduit is used for simulating a small hole on the breathing mask, and the breathing machine is communicated with the conduit through a corrugated pipe;
The second acquisition unit is used for detecting flow data of the inhaled air of the simulated lung in real time through a flow sensor arranged in the simulated lung at a plurality of data acquisition moments preset in the process of simulating the breath, and obtaining the first respiratory flow of the simulated lung at each data acquisition moment.
6. The system for constructing a respiratory flow prediction model according to claim 4, wherein the preprocessing module specifically comprises:
The first preprocessing unit is used for sequentially carrying out data cleaning and format unification on all the first pressure data, all the first pressure difference data and all the first respiratory flow respectively;
The second preprocessing unit is used for screening initial time from all the data acquisition time according to all the first pressure data, all the first pressure difference data and all the first respiratory flow which are subjected to data cleaning and uniform format; wherein a difference between the first pressure data at the initial time and the first pressure data at a last data acquisition time of the initial time is greater than a second threshold, a difference between the first pressure difference data at the initial time and the first pressure difference data at a last data acquisition time of the initial time is greater than a third threshold, a difference between the first respiratory flow at the initial time and the first respiratory flow at a last data acquisition time of the initial time is greater than a fourth threshold, the second threshold, the third threshold and the fourth threshold are all greater than zero, the first respiratory flow at the initial time is greater than zero and the first respiratory flow at a last data acquisition time of the initial time is less than or equal to zero; and constructing the sample data set according to the first pressure data, the first pressure difference data and the first respiration flow at the initial moment and the first pressure data, the first pressure difference data and the first respiration flow at all the data acquisition moments after the initial moment.
7. A respiratory flow prediction apparatus, comprising:
the acquisition module is used for acquiring second pressure data and second pressure difference data of the breathing machine at a plurality of data acquisition moments in real time; the breathing machine is communicated with the breathing mask with the small holes through a corrugated pipe;
The prediction module is used for inputting all the second pressure data and all the second pressure difference data into a respiratory flow prediction model, and outputting corresponding third respiratory flow as a respiratory flow prediction result at the moment to be predicted; the time to be predicted is the last data acquisition time of the data acquisition time corresponding to all the second pressure data input to the respiratory flow prediction model, and the respiratory flow prediction model is constructed by using the construction method of one of the respiratory flow prediction models in any one of claims 1 to 3.
8. The respiratory flow prediction apparatus according to claim 7, wherein the obtaining module specifically comprises:
the first detection unit is used for detecting the pressure intensity of the small holes on the breathing mask in real time at a plurality of data acquisition moments to obtain the second pressure intensity data of the breathing machine at each data acquisition moment;
The second detection unit is used for detecting the internal pressure difference of the breathing machine in real time through a pressure difference sensor arranged in the breathing machine at a plurality of data acquisition moments to obtain second pressure difference data of the breathing machine at each data acquisition moment.
CN202311162078.9A 2023-09-11 2023-09-11 Method, system and prediction device for constructing respiratory flow prediction model Active CN117059250B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311162078.9A CN117059250B (en) 2023-09-11 2023-09-11 Method, system and prediction device for constructing respiratory flow prediction model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311162078.9A CN117059250B (en) 2023-09-11 2023-09-11 Method, system and prediction device for constructing respiratory flow prediction model

Publications (2)

Publication Number Publication Date
CN117059250A CN117059250A (en) 2023-11-14
CN117059250B true CN117059250B (en) 2024-07-12

Family

ID=88659100

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311162078.9A Active CN117059250B (en) 2023-09-11 2023-09-11 Method, system and prediction device for constructing respiratory flow prediction model

Country Status (1)

Country Link
CN (1) CN117059250B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103182120A (en) * 2011-12-30 2013-07-03 北京谊安医疗系统股份有限公司 Method and device for man-machine synchronization of invasive respirator in noninvasive ventilation mode
CN108619597A (en) * 2017-03-17 2018-10-09 深圳市大雅医疗技术有限公司 A kind of pressure compensating method of lung ventilator runner
CN114177451A (en) * 2021-12-29 2022-03-15 杭州电子科技大学 Control method of single breathing cycle pressure-volume double control mode of breathing machine

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108135536B (en) * 2015-09-29 2021-07-20 皇家飞利浦有限公司 Simultaneous estimation of respiratory dynamics and patient effort via parameter optimization
CN115052648A (en) * 2020-02-26 2022-09-13 诺瓦雷斯普科技有限公司 Method and apparatus for determining and/or predicting sleep and respiratory behavior governing airway pressure
CN111529847A (en) * 2020-05-07 2020-08-14 山东凯鑫宏业生物科技有限公司 Breathing machine system with breathing phase identification control applied to intelligent medical system and control method
CN111657950A (en) * 2020-06-12 2020-09-15 山东凯鑫宏业生物科技有限公司 Breathing machine system applied to intelligent medical treatment and capable of predicting breathing state of patient and control method thereof
US20220168527A1 (en) * 2020-12-02 2022-06-02 Tiro Medical Limited Method and apparatus to guide mechanical ventilation

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103182120A (en) * 2011-12-30 2013-07-03 北京谊安医疗系统股份有限公司 Method and device for man-machine synchronization of invasive respirator in noninvasive ventilation mode
CN108619597A (en) * 2017-03-17 2018-10-09 深圳市大雅医疗技术有限公司 A kind of pressure compensating method of lung ventilator runner
CN114177451A (en) * 2021-12-29 2022-03-15 杭州电子科技大学 Control method of single breathing cycle pressure-volume double control mode of breathing machine

Also Published As

Publication number Publication date
CN117059250A (en) 2023-11-14

Similar Documents

Publication Publication Date Title
CN102187343B (en) For detecting the system and method for respiratory insufficiency in subject breathed
CN103169476B (en) Method and device for identification and early warning of respiration wave form image
CN102770069B (en) Nitric oxide measurement method and apparatus
CN102770070A (en) System and method for assessing real time pulmonary mechanics
CN102481119B (en) System and method for quantifying lung compliance in a self-ventilating subject
CN104334080B (en) Cough auxiliary and measuring system and method
CN102056539A (en) Systems and methods for ventilation in proportion to patient effort
CN102469954A (en) Method and apparatus of determining exhaled nitric oxide
CN105662417B (en) A kind of control method and device based on the pressure signal feature recognition sound of snoring
CN117059250B (en) Method, system and prediction device for constructing respiratory flow prediction model
CN107928645A (en) A kind of subject's respiratory capacity senses and intelligent control system and its implementation in real time
CA2066756A1 (en) Device and method for neural network breathing alarm
US20220249999A1 (en) Systems and methods for detecting the status of a particle filter
CN110786855B (en) Sputum induction device and control method thereof
CN111973188A (en) Method for estimating respiratory mechanics parameter based on neural network
CN110057733B (en) Experimental device for respiratory system of warm-up dummy
CN115944820A (en) Control method and device for accurately quantifying respirator parameters
Anitha et al. Controlled mechanical ventilation for enhanced measurement in pressure and flow sensors
US20220168527A1 (en) Method and apparatus to guide mechanical ventilation
Cabacungan et al. Design and development of A-vent: a low-cost ventilator with cost-effective mobile cloud caching and embedded machine learning
CN117180573B (en) Respiratory system mechanical ventilation method and system based on double-index model
CN116500903B (en) Breathing machine heating and humidifying control method and system based on artificial intelligence and breathing machine
US20230384289A9 (en) Determining composition of respiratory air
CN114894512A (en) Multifunctional simulated lung control device based on digital twins and working method thereof
KR102490153B1 (en) Apparatus and method for self-chosen intelligent respiratory disease intervention based on personal particulate matter exposure information

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant