CN113782210A

CN113782210A - Method for predicting treatment failure probability of noninvasive ventilator

Info

Publication number: CN113782210A
Application number: CN202111072623.6A
Authority: CN
Inventors: 戴征; 罗恢育
Original assignee: Hunan Micomme Zhongjin Medical Technology Development Co Ltd
Current assignee: Hunan Micomme Zhongjin Medical Technology Development Co Ltd
Priority date: 2021-09-14
Filing date: 2021-09-14
Publication date: 2021-12-10
Anticipated expiration: 2041-09-14
Also published as: CN113782210B

Abstract

The invention is suitable for the technical field of medical instruments, and relates to a method for predicting the treatment failure probability of a noninvasive ventilator, which comprises the following steps: s10, collecting training data of the user and recording the final treatment result of the user; s20, inputting the training data into the time series model and the classification model and carrying out calibration labeling; s30, training the training data by using an LSTM model, taking a time sequence model with relatively small MSE as a time sequence mother model, and predicting future physiological parameters through the time sequence mother model; s40, training future physiological parameters by using a LightGBM model, analyzing the final treatment effect of noninvasive respiration according to the distribution of training samples, adjusting parameters to determine an optimal classification model, and predicting the probability of treatment failure by using the optimal classification model; and S50, distributing the training data of the user and the weight of the training data in the time series mother model, and retraining the time series mother model. The invention can predict future physiological parameters and predict the probability of treatment failure according to the future physiological parameters.

Description

Method for predicting treatment failure probability of noninvasive ventilator

Technical Field

The invention belongs to the technical field of medical instruments, and particularly relates to a method for predicting the treatment failure probability of a noninvasive ventilator.

Background

Noninvasive ventilator although the invention is relatively late in invasive ventilators, treatment using a noninvasive ventilator is a preferred method for treating subjects with less symptoms of respiratory disease. However, when the symptoms of patients get worse, how to switch to invasive treatment at proper time is always a key topic of clinical attention. At present, medical staff analyze the concerned indexes according to past experience based on related monitoring indexes (such as PaCO2 indexes) so as to judge the invasive treatment time. The prior art has some defects: the subjectivity is too strong, the factors of non-invasive treatment failure are many, the age, basic diseases, clinical monitoring indexes and the like of a patient can influence the non-invasive treatment result, medical workers analyze the results according to the past experience, the effect is difficult to judge due to the common influence of a plurality of factors, and the judgment results are possibly different if the indexes of subjective attention of the medical workers are different; the accumulation time is long, and the method needs years of clinical experience accumulation; and the used physiological parameters are current, the physiological parameter change of the patient at the future time cannot be obtained, and the failure probability of the patient to use the noninvasive ventilation at the future time cannot be predicted.

Patent document CN202010961481.8 discloses a method for determining expected effect of a respiratory support apparatus and a respiratory support apparatus, which adopts a technical solution comprising the steps of: s1, obtaining basic sign data of a user, wherein the basic sign data at least comprises basic characteristics, medical history characteristics and detection data; and S2, comprehensively judging the basic sign data by using a judgment network model, and determining the expected effect of the non-invasive treatment. This application also fails to address the problems of the prior art.

Therefore, how to provide a method capable of predicting physiological parameters of a patient at a future time and predicting the failure probability of a noninvasive ventilator based on the future physiological parameters is a problem to be solved by those skilled in the art.

Disclosure of Invention

Aiming at the defects of the prior art, the invention aims to provide a method for predicting the treatment failure probability of a noninvasive ventilator so as to solve the problems that the future physiological parameters cannot be predicted and the treatment failure probability cannot be predicted through the future physiological parameters in the prior art.

In order to solve the technical problems, the invention adopts the following technical scheme:

the invention provides a method for predicting the treatment failure probability of a noninvasive ventilator, which comprises the following steps:

s10, collecting training data of the user and recording the final treatment result of the user, wherein the training data comprises basic characteristics, disease conditions and monitoring data of the user;

s20, inputting the training data into the time series model and the classification model and carrying out calibration labeling;

s30, selecting an LSTM model to train the training data, taking a time sequence model with relatively small MSE as a time sequence mother model, and predicting future physiological parameters through the time sequence mother model;

s40, selecting a LightGBM model to train the future physiological parameters, analyzing the final treatment effect of noninvasive respiration according to the distribution of training samples, adjusting parameters to determine an optimal classification model, and predicting the probability of treatment failure through the optimal classification model;

and S50, distributing the training data of the user and the weight of the training data in the time series mother model, and retraining the time series mother model.

Further, in step S40, the training data is divided into a training set and a test set according to a predetermined ratio, where the training set is used for training the LightGBM model, and the test set is used for test evaluation.

Further, the ratio of the training set to the test set is 4: 1.

Further, in step S50, the weight ratio of the training data of the user to the training data in the time-series mother model is 9:1 or 4: 1.

Further, in step S30, a machine learning training frame adopted by the LSTM model training is TensorFlow, and the related parameters include training batch number, learning rate, neuron number, output result number, and iteration number.

Further, in step S40, the model algorithm includes a logistic regression model, a support vector classification model, a neural network model, and a decision tree model.

Further, in step S40, the LightGBM model related parameters include a learning rate, a tree depth, a number of leaf nodes, and a number of toolboxes.

Further, in step S10, the basic characteristics include gender and age, the disease condition includes diagnosis of disease, hypertension, diabetes and cerebrovascular disease, and the monitoring data includes body temperature, consciousness index GCS, diastolic blood pressure, systolic blood pressure, respiratory rate, heart rate, CO₂Partial pressure, O₂Partial pressure, blood pH, blood oxygen concentration, arterial blood partial pressure of carbon dioxide PaCO₂Partial pressure of arterial blood oxygen PaO₂Oxygen concentration fraction FiO in inhalation gas₂。

Further, in the step S10, the final treatment effect includes non-invasive treatment failure, non-invasive treatment success and inconclusive.

Further, in step S20, the user characteristics described in the foregoing text need to be subjected to numerical conversion.

Compared with the prior art, the method for predicting the treatment failure probability of the noninvasive ventilator provided by the invention at least has the following beneficial effects:

the invention can predict the future physiological parameters of the patient through the time series model, and predict the future physical condition of the patient; the probability of non-invasive respiratory therapy failure can be predicted on the basis of future physiological parameters through the classification model; and the time series model is retrained by distributing the data weight of the patient data and the time series mother model which are in use so as to improve the accuracy of predicting the future physiological parameters.

Drawings

In order to illustrate the solution of the invention more clearly, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are some embodiments of the invention, and that other drawings may be derived from these drawings by a person skilled in the art without inventive effort.

Fig. 1 is a flowchart of a method for predicting the probability of a therapy failure of a noninvasive ventilator according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a conventional RNN model;

fig. 3 is a schematic diagram of a first unit structure of an LSTM model of a method for predicting the probability of therapy failure of a noninvasive ventilator according to an embodiment of the present invention.

Detailed Description

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs; the terminology used in the description presented herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention, e.g., the terms "length," "width," "upper," "lower," "left," "right," "front," "rear," "vertical," "horizontal," "top," "bottom," "inner," "outer," etc., refer to an orientation or position based on that shown in the drawings, are for convenience of description only and are not to be construed as limiting of the present disclosure.

The terms "including" and "having," and any variations thereof, in the description and claims of this invention and the description of the above figures are intended to cover non-exclusive inclusions; the terms "first," "second," and the like in the description and in the claims, or in the drawings, are used for distinguishing between different objects and not necessarily for describing a particular sequential order. In the description and claims of the present invention and in the description of the above figures, when an element is referred to as being "fixed" or "mounted" or "disposed" or "connected" to another element, it may be directly or indirectly located on the other element. For example, when an element is referred to as being "connected to" another element, it can be directly or indirectly connected to the other element.

Furthermore, reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the invention. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It is explicitly and implicitly understood by one skilled in the art that the embodiments described herein can be combined with other embodiments.

The invention provides a method for predicting the treatment failure probability of a noninvasive ventilator, which is applied to the noninvasive ventilator and comprises the following steps: s10, collecting training data of the user and recording the final treatment result of the user, wherein the training data comprises the basic characteristics, the diseased condition and the monitoring data of the user; s20, inputting the training data into the time series model and the classification model and carrying out calibration labeling; s30, selecting an LSTM model to train training data, taking a time sequence model with relatively small MSE as a time sequence mother model, and predicting future physiological parameters through the time sequence mother model; s40, selecting a LightGBM model to train future physiological parameters, analyzing the final treatment effect of noninvasive respiration according to the distribution of training samples, adjusting parameters to determine an optimal classification model, and predicting the probability of treatment failure through the optimal classification model; and S50, distributing the training data of the user and the weight of the training data in the time series mother model, and retraining the time series mother model.

In order to make the technical solutions of the present invention better understood by those skilled in the art, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

The invention provides a method for predicting the treatment failure probability of a noninvasive ventilator, which is applied to the noninvasive ventilator and comprises the following steps as shown in figure 1:

s10, collecting training data of the user and recording the final treatment result of the user, wherein the training data comprises the basic characteristics, the diseased condition and the monitoring data of the user;

in particular, training data from a clinically collectable user includes, but is not limited to: basic characteristics include gender, age, etc.; the disease condition comprises diagnosis of diseases, whether hypertension, diabetes, cerebrovascular diseases and the like exist once; the monitoring data includes body temperature, consciousness index GCS, diastolic pressure, systolic pressure, respiratory rate, heart rate, and CO₂Partial pressure, O₂Partial pressure, blood pH, blood oxygen concentration, arterial blood partial pressure of carbon dioxide PaCO₂Partial pressure of arterial blood oxygen PaO₂Oxygen concentration fraction FiO in inhalation gas₂And the training data are collected every 1 hour (preferably 0.5-2 hours); the final treatment effects of the user comprise non-invasive treatment failure (the patient directly dies, the intubation tube is used after the patient's disease condition deteriorates, the patient is discharged but the prognosis is poor and the like), non-invasive treatment success (the patient is discharged smoothly and the like), non-judgment (the patient forcibly discharges but medical staff cannot know follow-up conditions which are few and can be deleted in follow-up analysis), recording the final treatment effect of the patient, and performing numerical conversion according to the final treatment effect, wherein the non-invasive treatment failure can be 0, the non-invasive treatment success can be 1, and the non-invasive treatment success can be 2;

specifically, the characteristics described in the chinese language need to be labeled, such as gender (male is 1, female is 0), the time series model uses every 12 sets of parameters as a time series training unit, the next set of parameters of each time series training unit is used as a time series training label, and so on, so that one patient can obtain multiple sets of time series training units; the classification model takes the parameter of the last moment as a classification training unit, and the final treatment result is a classification training label;

s30, selecting an LSTM model to train training data, taking a time sequence model with relatively small MSE as a time sequence mother model, and predicting future physiological parameters through the time sequence mother model;

specifically, as shown in FIGS. 2 and 3, an LSTM model (or related optimization model such as GRU, etc.) is selected, which is improved by an RNN model, where h is_N＝σ(W_h*h_N-1+W_I*x_N) W denotes the parameters of the desired training, σ denotes the sigmoid activation function, y_N＝σ(W_o*h_N) Combining fig. 2, fig. 3 and the above formula, we can obtain: h is_NCan be formed by_N-1、x_NThe precondition obtained is that h is considered_NAnd h_N-1、x_NThere is correlation, and the correlation will continue all the time in the transmission process, when some parameter in the transmission process is always greater than 1 or less than 1, and thus it is easy to accumulate in the transmission process, and the parameter value is exponential change or always tends to 0 (i.e. gradient explosion or disappearance), so the LSTM model is selected to select h in the RNN model_NIs divided into h_NAnd c_NWherein c is_NIs a variable that changes relatively quickly, h_NIs a variable with slow change, and the specific formula is c_N＝z_f×c_N-1+z_i×z，h_N＝z_o×tanh(c_N-1)，y_N＝σ(W*h_N) Where "x" denotes multiplication of element matrices, "+" element matrices are added, and z is all represented by h_N-1Multiplied by x, only provided with different activation functions and parameters, where z_f、z_i、z_oRespectively showing a forgetting gate, an input gate and an output gate, respectively having different functions, and not specifically unfolded here; the LSTM model uses the mean square error as an evaluation criterion,

wherein y is_iIn the form of an actual value of the value,

for predictive value, a common machine learning training framework (e.g., TensorFl) can be usedow, etc.), relevant parameters such as training batch number, learning rate, neuron number, output result number and iteration number are adjusted in different modes, and then a time sequence model with relatively small MSE is obtained and is used as a mother model for predicting future physiological parameters;

s40, selecting a LightGBM model to train future physiological parameters, analyzing the final treatment effect of noninvasive respiration according to the distribution of training samples, adjusting parameters to determine an optimal classification model, and predicting the probability of treatment failure through the optimal classification model;

specifically, in the embodiment, a LightGBM model is preferably selected to train data, a LightGBM model core algorithm is developed from a decision tree, and compared with other decision tree (such as GBDT) algorithms, the LightGBM model core algorithm has the characteristics of low memory, high operating efficiency, higher accuracy, parallelization operation and the like, future physiological parameters are divided into a training set and a test set according to the proportion of 4:1, the training set is used for training the LightGBM model, and the test set is used for test evaluation; the model algorithm comprises a logistic regression model, a support vector classification model, a neural network model and a decision tree model, the relevant parameters of the LightGBM model comprise learning rate, tree depth, the number of leaf nodes and the number of toolboxes, and in the embodiment, the key parameters can be initialized: the learning rate can be set to 0.1, the tree depth can be set to 4, the number of leaf nodes is set to 8, the number of toolboxes is set to 50, other parameters can be modified and perfected according to actual conditions, the proportion of success and failure of noninvasive respiratory therapy is analyzed according to the distribution of training samples, one index can be selected as a measurement standard adjustment parameter, such as the recall rate (the proportion correctly predicted in the failed samples) or the accuracy rate (the proportion correctly predicted in all samples), and therefore an optimal model is determined;

s50, distributing the training data of the user and the weight of the training data in the time sequence mother model, and retraining the time sequence mother model;

specifically, based on a time series mother model, targeted retraining can be performed on a patient who is using the model, when the model is retrained, a database needs to be built, data and labels thereof are sequentially stored according to a mode of collecting training data in the time series model, in the retraining process, the calculation of MSE is different from that in the primary training process, a weight ratio of the patient data which is using the model to the data in the mother model can be given to be 9:1 or 4:1 and the like, namely the MSE of the patient data which is using the model is calculated and multiplied by 0.9 or 0.8 and the like, the MSE of the data in the mother model is calculated and multiplied by 0.1 or 0.2 and the like, and other processes are consistent with the training process.

The method for predicting the treatment failure probability of the noninvasive ventilator can predict future physiological parameters of the patient through the time series model, and predict the future physical condition of the patient; the probability of non-invasive respiratory therapy failure can be predicted on the basis of future physiological parameters through the classification model; the time series model is retrained by distributing the data weight of the patient data and the time series mother model which are in use so as to improve the accuracy of predicting the future physiological parameters; in addition, the non-invasive respiratory failure probability is predicted, and meanwhile, the future physiological parameters are predicted in an intermediate way, so that other purposes can be realized, such as whether related medicines need to be used when the blood PH value is too low or too high is predicted, and the like.

It is to be understood that the above-described embodiments are merely preferred embodiments of the present invention, and not all embodiments are shown in the drawings, which are set forth to limit the scope of the invention. This invention may be embodied in many different forms and, on the contrary, these embodiments are provided so that this disclosure will be thorough and complete. Although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that various changes in the embodiments and modifications can be made, and equivalents may be substituted for elements thereof. All equivalent structures made by using the contents of the specification and the attached drawings of the invention can be directly or indirectly applied to other related technical fields, and are also within the protection scope of the patent of the invention.

Claims

1. A method of predicting the probability of a noninvasive ventilator treatment failure, comprising the steps of:

2. The method of claim 1, wherein in step S40, the training data is divided into a training set and a test set according to a predetermined ratio, the training set is used for training the LightGBM model, and the test set is used for test evaluation.

3. The method of claim 2, wherein the ratio of the training set to the test set is 4: 1.

4. The method of claim 3, wherein in step S50, the weight ratio of the training data of the user to the training data in the time-series mother model is 9:1 or 4: 1.

5. The method according to claim 1, wherein in step S30, the machine learning training framework used in the LSTM model training is tensrflow, and the related parameters include training batch number, learning rate, neuron number, output result number, and iteration number.

6. The method according to claim 5, wherein in the step S40, the model algorithm includes a logistic regression model, a support vector classification model, a neural network model, and a decision tree model.

7. The method of claim 6, wherein in the step S40, the LightGBM model-related parameters include learning rate, tree depth, number of leaf nodes, and number of toolboxes.

8. The method according to claim 1, wherein the basic characteristics include sex and age in step S10, the disease condition includes diagnosis of disease, hypertension, diabetes and cerebrovascular disease, the monitoring data includes body temperature, consciousness index GCS, diastolic blood pressure, systolic blood pressure, respiratory rate, heart rate, CO, and the like₂Partial pressure, O₂Partial pressure, blood pH, blood oxygen concentration, arterial blood partial pressure of carbon dioxide PaCO₂Partial pressure of arterial blood oxygen PaO₂Oxygen concentration fraction FiO in inhalation gas₂。

9. The method of claim 8, wherein in step S10, the final treatment effect includes noninvasive treatment failure, noninvasive treatment success and no judgment.

10. The method of claim 9, wherein the step S20 is performed by numerical conversion of the user characteristics described in the text.