CN112354042A

CN112354042A - Analgesia pump flow control method and device

Info

Publication number: CN112354042A
Application number: CN202011389430.9A
Authority: CN
Inventors: 许红亮; 程世易; 许容芳; 倪红霞; 王萍
Original assignee: Nantong Tumor Hospital
Current assignee: Nantong Tumor Hospital
Priority date: 2020-12-01
Filing date: 2020-12-01
Publication date: 2021-02-12

Abstract

The present invention provides a method and device for controlling the flow of an analgesic pump, which are applied to a processor and include the following steps for controlling the analgesic pump: S1, establishing an analgesic pump database; S2, training through a stacking model structure data; S3, input medical data into XGBoost model, LightGBM model and Random Foreast model, perform weighted calculation on the results of the three models to give the final result, and save the trained model; S4, deploy the trained model to the cloud on the server. The beneficial effects of the present invention are as follows: the present invention avoids outputting results with large differences when the training of a single model fails, improves the safety of the analgesic pump, the required database can be continuously expanded according to clinical experience and patient feedback, and at the same time new training Data augmentation can also improve the accuracy of algorithmic output metering.

Description

Analgesia pump flow control method and device

Technical Field

The invention relates to the technical field of analgesia pump fluid control, in particular to a method and a device for controlling the flow of an analgesia pump.

Background

The analgesia pump is a common analgesia mode after the operation of a patient, can effectively relieve the pain of the patient after the operation and improves the comfort level of the patient after the operation. The analgesic pump has certain side effects in the using process, such as symptoms of dizziness, nausea, vomiting and the like, and even respiratory depression. The cause of such symptoms is often due to overdosing. If the dose is reduced, no analgesic effect is achieved.

Because the setting of the flow of the analgesia pump is subjective, senior doctors have rich experience of medicine taking and accurate flow setting, and young doctors usually face the difficult problem of inaccurate holding, for example, if the flow setting of the analgesia pump is too large, certain side effects, such as symptoms of dizziness, nausea, vomiting and the like, even respiratory depression, can be brought. If the dosage is reduced, the stress response of the organism is enhanced, the immunity is reduced, the wound healing is delayed, the body and mind are seriously injured, and even psychological diseases are generated to cause suicide.

The flow of the existing analgesia pump is set by a doctor in an upper limit mode, and a patient can adjust the flow autonomously. However, the method does not consider the surgical characteristics, physiological characteristics, living conditions and the like of patients, so that all people share one set of analgesic method.

How to solve the above technical problems is the subject of the present invention.

Disclosure of Invention

The invention aims to provide an analgesia pump flow control method and device, which are used for outputting individually customized analgesia pump flow control aiming at each patient by using an artificial intelligence algorithm, can generate analgesia most suitable for the patient according to the self condition of each patient, can improve the postoperative life quality under the condition of ensuring safety, do not depend on the level of a doctor, and are suitable for large-area popularization.

The invention is realized by the following measures: a method for controlling the flow of an analgesic pump, wherein the method is applied to a processor and comprises the following steps of:

s1, establishing an analgesia pump database, recording the analgesia information such as sex, age, operation duration, operation position, operation mode, past history, smoking and alcoholism history and the like of the patient in the database, and simultaneously recording the standard analgesia dosage of the corresponding patient;

s2, training data through a stacking model structure, inputting model characteristics as database recording data, outputting results as appropriate analgesic measures, and performing model training;

s3, inputting medical data into the XGboost model, the LightGBM model and the Random Foreast model, outputting a measurement judgment result by each model, inputting all the results of the three models into a neural network model, performing weighted calculation on the results of the three models respectively to give a final result, and storing the trained models;

s4, deploying the trained model to a cloud server, inputting patient information into a stacking model structure after a new patient appears, and outputting a metering calculation result of the analgesic pump by the model.

In order to better achieve the above object, the present invention further provides an apparatus for implementing a method for controlling a flow rate of an analgesic pump, wherein the apparatus is applied to a processor and comprises a primary model and a secondary model for controlling the analgesic pump: the first-level model comprises an XGboost model, a LightGBM model and a Random Foreast model; the secondary model comprises a neural network model;

the apparatus for controlling an analgesic pump comprises the following:

s1, establishing an analgesia pump database, recording analgesia information such as sex, age, operation duration, operation position, operation mode, past history, smoking and alcoholism history and the like of a patient in the database, and simultaneously recording a standard analgesia metering device corresponding to the patient;

s2, training data through a stacking model structure, inputting model characteristics as database recording data, outputting results as appropriate analgesic measurement, and performing model training;

and S4, deploying the trained model to a cloud server, inputting the patient information into a stacking model structure after a new patient appears, and outputting the metering calculation result of the analgesic pump by the model.

The invention comprises a training stage and an application stage in practical use:

a training stage: the method comprises the steps that 3 primary models are trained by utilizing patient physiological data and standard analgesic flow in a database respectively, then 3 models can all output analgesic flow results, then a neural network model is utilized to train the results of the 3 primary models, the analgesic flow given by the 3 models is input, the standard flow results are output, the input characteristics of the 3 primary models are the patient physiological data, the input label is the patient standard analgesic flow, the output result is algorithm calculation analgesic flow, the input characteristics of the secondary model are the analgesic flow calculated by the 3 primary models, the input label is the standard analgesic flow, and the output is the final calculation result of the algorithm.

An application stage: the physiological data of the patient is input into 3 primary models, the 3 models respectively give corresponding calculation results, the 3 results are input into a secondary model, and the secondary model integrates the results of the 3 primary models and outputs the final result to guide a doctor to take medicine.

Compared with the prior art, the invention has the beneficial effects that:

1. the invention can achieve the advantages of reasonable medication and different treatment schemes according to the advantages of analgesia pump flow under the condition of the patient self condition.

2. The stability of the algorithm is improved by combining 3 models in the algorithm model, the result with larger difference is prevented from being output when the training of a single model fails, and the safety of the analgesia pump is improved.

3. The database required by the model can be continuously expanded according to clinical experience and patient feedback, and meanwhile, the accuracy of algorithm output measurement can be improved due to the addition of new training data.

4. The flow recommendation of the invention is generated by model calculation, does not need to depend on manpower, achieves the purpose of popularizing the experience of experts, can share the trained model to hospitals with weak medical resources, and improves the anesthesia level of the hospitals in the laggard areas.

5. When the hospital algorithm model is deployed at the cloud end, the hospital algorithm model can be shared by other hospitals, and the hospital business income is improved by selling the model using permission.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention.

FIG. 1 is a schematic diagram of the finishing process of the present invention.

FIG. 2 is a schematic diagram of a database and stacking model structure according to the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. Of course, the specific embodiments described herein are merely illustrative of the invention and are not intended to be limiting.

Example 1

Referring to fig. 1 to 2, the present invention provides a method for controlling the flow rate of an analgesic pump, wherein the method is applied to a processor and comprises the following steps for controlling the analgesic pump:

the apparatus for controlling an analgesic pump comprises the following:

The following are examples of methods and apparatus for implementing the analgesia pump flow control of the present invention:

(1) establishing databases in a plurality of hospitals for collecting safe and comfortable values of analgesic flow of each patient after operation and recording physiological characteristic data of the patient, such as sex, age, operation duration, operation position, operation mode, past history and history of smoking and alcoholism;

(2) establishing an algorithm model, wherein the model is divided into 2 layers, the first layer is 3 XGboost models, LightGBM models and Random Foreast models, the second layer is a neural network model, all the models are regression models, the input of the first layer XGboost model, LightGBM models and Random Foreast models is set as physiological characteristic data of a patient, the output is set as analgesic flow of the patient, the input of the second layer neural network model is set as the output result of the 3 models of the first layer, and the output is set as analgesic flow of the patient;

(3) utilizing a database training model, substituting patient data stored in a database into the model, and carrying out iterative training for a plurality of times until the loss function of the model is reduced to 0.001;

(4) deploying the obtained model to a cloud server;

(5) when a new case occurs, inputting the physiological data of the patient into a cloud model, and outputting the analgesic flow result obtained by calculation by the model;

(6) the doctor sets the analgesia flow for the patient according to the output result of the model.

The following is another example of an implementation of the analgesia pump flow control method and apparatus of the present invention:

1) establishing a database in a plurality of hospitals, wherein the analgesic pump is sufentanil, the database is used for collecting safe and comfortable analgesic flow values of each patient after operation, and simultaneously recording physiological characteristic data of the patient, such as sex, age, operation duration, operation position, operation mode, past history and history of smoking and alcoholism;

the database structure is as follows, 30000 samples are collected from a plurality of hospitals;

2) establishing an algorithm model, wherein the model is divided into 2 layers, the first layer is 3 XGboost models, LightGBM models and Random Foreast models, the second layer is a neural network model, all the models are regression models, the input of the first layer XGboost model, LightGBM models and Random Foreast models is set as physiological characteristic data of a patient, the output is set as analgesic flow of the patient, the input of the second layer neural network model is set as the output result of the 3 models of the first layer, and the output is set as analgesic flow of the patient;

model parameter setting

Leading in a Random Foreast model;

n _ estimators (maximum number of iterations): 500

criterion (evaluation criterion): mse

max _ depth (maximum depth): 20

The other parameters are model default values;

introducing an XGboost model;

the XGboost model type: gbtree

eta (learning Rate): 0.01

max _ depth (maximum depth of tree): 15

LightGBM model (L2 regularization): 0.01

The other parameters are model default values;

importing LightGBM model

boosting _ type (tree structure type): gbdt

object type regression

learning rate 0.03

num _ leaves (maximum number of leaves) 40

max _ depth 12

subsample 0.8

Importing a neural network model

The model input is the result of the first 3 models, so a 30000 × 3 matrix is input, and a 1 × 1 matrix result is output.

The hidden layer is 1 layer, and the number of nodes is 16.

The optimizer is sgd and the learning rate is 0.03.

3) Training the model by using the database;

taking out data in a database, wherein the data are characterized by the first 9 data, one-hot codes are sampled from the past history, so that the total number of data columns is 12, the characteristic data in the database can show a matrix of 30000 multiplied by 12, the data labels are the last 1 data, namely the flow rate of the labor pain, and the label data in the database can show a matrix of 30000 multiplied by 1;

normalizing the data in the matrix, and scaling the maximum and minimum values of each column of data to the range of [ -1,1 ];

in 3 primary models, a 30000 × 12 matrix (patient physiological characteristics) is input and a 30000 × 1 matrix (analgesic pump flow) is output by training the data. Completing the first-level model training;

the data is input into 3 primary models again to obtain a primary prediction result of 30000 multiplied by 3, the data is used as the input of a neural network of a secondary model, and the output is still a matrix of 30000 multiplied by 1 (analgesic pump flow rate). After the iteration is carried out for 800 times, the loss value of the neural network is reduced to 1e-4, and the training of the secondary model is completed;

4) deploying the obtained model to a cloud server;

5) when a new case occurs, the physiological data of the patient is pre-processed by installing the normalized scaling rules in the database. Inputting the preprocessed data into a cloud model, wherein 3 primary models respectively give calculation results and input into a secondary model, and the secondary model integrates the first 3 results to give final flow data;

6) the doctor sets the analgesia flow for the patient according to the output result of the model.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims

1. an analgesic pump flow control method, is characterized in that, is applied in the processor, comprises the steps for controlling the analgesic pump as follows:

S1. Establish an analgesic pump database, record the patient's gender, age, operation time, operation site, operation method, past history, smoking and drinking history and other analgesic information in the database, and at the same time record the standard analgesic measurement of the corresponding patient;

S2. Train data through a stacking model structure, input model features as database record data, and output results as appropriate analgesic measurement, and perform model training;

S3. Input the medical data into the XGBoost model, the LightGBM model and the Random Forest model, each model outputs the measurement and judgment results, the results of the three models are all input into a neural network model, and the results of the three models are weighted respectively. Give the final result and save the trained model;

S4. The trained model is deployed on the cloud server. When a new patient appears, the patient information is input into the stacking model structure, and the model outputs the measurement and calculation result of the analgesic pump.

2. a device for realizing an analgesic pump flow control method, is characterized in that, is applied in the processor, comprises the primary model and the secondary model for controlling the analgesic pump: the primary model comprises XGBoost model, LightGBM model and RandomForeast model; the secondary model includes a neural network model;

Devices used to control an analgesic pump include the following:

S1. Establish an analgesic pump database, record the patient's gender, age, operation time, operation site, operation method, past history, smoking and drinking history and other analgesic information in the database, and record the standard analgesic metering device corresponding to the patient at the same time;

S2, train data through a stacking model structure, input model features as database record data, and output results as appropriate analgesic measurement, a device for model training;

S3. Input the medical data into the XGBoost model, the LightGBM model and the Random Foreast model, each model outputs the measurement judgment result, the results of the three models are all input into a neural network model, and the results of the three models are weighted respectively. A device that gives the final result and saves the trained model;

S4. The trained model is deployed on the cloud server. When a new patient appears, the patient information is input into the stacking model structure, and the model outputs the device for calculating the measurement result of the analgesic pump.

3. an analgesic pump flow control method, is characterized in that, is applied in the processor, comprises the steps for controlling the analgesic pump as follows:

1) A database is established in several hospitals, and the analgesic pump is sufentanil. The database is used to collect the safe and comfortable analgesic flow value of each patient after surgery, and record the patient's physiological characteristics data, such as patient gender, Age, operation time, operation site, operation method, past history, history of smoking and drinking;

2) Establish an algorithm model. The model is divided into two layers. The first layer is XGBoost model, LightGBM model and RandomForeast model. The second layer is neural network model. All models are regression models. The first layer is XGBoost model, LightGBM model and RandomForeast model. The input of the Foreast model is set as the patient's physiological characteristic data, the output is set as the patient's analgesic flow, the input of the second layer of neural network model is set as the output results of the three models in the first layer, and the output is set as the patient's analgesic flow;

3) Use the database to train the model;

Take out the data in the database, the data features are the first 9, of which the past sampling one-hot encoding, the total number of data columns is 12, the feature data in the database can be seen as a 30000×12 matrix, and the data label is the last 1 1, namely the labor flow, the label data in the database can be seen as a 30000×1 matrix;

Normalize the data in the matrix, and scale the maximum and minimum values of each column of data to the range of [-1,1];

Use data to train three first-level models, input a 30000×12 matrix, and output a 30000×1 matrix to complete the first-level model training;

Input the data into the three first-level models again, and obtain the first-level prediction result of 30000×3. This data is used as the input of the neural network of the second-level model, and the output is still a matrix of 30000×1. After 800 iterations, the neural network’s The loss value drops to 1e-4, and the training of the secondary model is completed;

4) Deploy the obtained model to the cloud server;

5) When a new case occurs, the physiological data of the patient is preprocessed by installing the normalized scaling rules in the database, and the preprocessed data is input into the cloud model, and the three first-level models are given respectively. Calculate the results and input them into the second-level model. The second-level model combines the first three results to give the final flow data;

6) The doctor sets the analgesic flow rate for the patient according to the model output result.

4. A device for realizing an analgesic pump flow control method, characterized in that, applied in the processor, including a first-level model and a second-level model for controlling an analgesic pump: the first-level model includes XGBoost model, LightGBM model and RandomForeast model; the secondary model includes a neural network model;

Devices used to control an analgesic pump include the following:

1) A database is established in several hospitals, and the drug for the analgesic pump is sufentanil, and the database is used to collect the numerical value of the safe and comfortable analgesic flow of each patient after the operation, and simultaneously record the device of the patient's physiological characteristic data;

2) Establish an algorithm model. The model is divided into two layers. The first layer is XGBoost model, LightGBM model and RandomForeast model. The second layer is neural network model. All models are regression models. The first layer is XGBoost model, LightGBM model and RandomForeast model. The input of the Foreast model is set as the patient's physiological characteristic data, the output is set as the patient's analgesic flow, the input of the second layer of neural network model is set as the output results of the three models in the first layer, and the output is set as the device for the patient's analgesic flow; 3) a device for training a model using a database;

4) a device for deploying the obtained model to a cloud server;

5) When a new case occurs, the physiological data of the patient is preprocessed by installing the normalized scaling rules in the database, and the preprocessed data is input into the cloud model, and the three first-level models are given respectively. Calculate the results and input them into the secondary model. The secondary model combines the first 3 results to give the final flow data device;

6) The doctor sets an analgesic flow device for the patient according to the model output result.