CN114724710A

CN114724710A - Emergency scheme recommendation method and device for emergency events and storage medium

Info

Publication number: CN114724710A
Application number: CN202210649826.5A
Authority: CN
Inventors: 计虹; 王梦莹; 孙震; 贾末; 朱声荣; 谷今一
Original assignee: Peking University Third Hospital Peking University Third Clinical Medical College
Current assignee: Peking University Third Hospital Peking University Third Clinical Medical College
Priority date: 2022-06-10
Filing date: 2022-06-10
Publication date: 2022-07-08

Abstract

The invention relates to the technical field of data processing, in particular to an emergency scheme recommendation method and device for an emergency, wherein the method comprises the following steps: acquiring electronic medical record information of an emergency wounded person in the process of carrying out emergency treatment on the emergency wounded person; determining a target emergency type corresponding to the first-aid wounded person and a target level under the target emergency type according to the electronic medical record information and a pre-trained emergency classification model; and outputting a corresponding target emergency scheme according to the type of the target emergency and the target level. Through this technical scheme, promote emergency's treatment effect, realize giving treatment scheme as soon as possible.

Description

Emergency scheme recommendation method and device for emergency events and storage medium

Technical Field

The invention relates to the technical field of data processing, in particular to an emergency scheme recommendation method and device for an emergency event and a storage medium.

Background

With the rapid development of internet information technology and the construction of regional intelligent medical service platforms, the intelligent medical treatment can really realize the interconnection and intercommunication of medical information. The intelligent medical engineering is a multi-level data processing platform, and the fusion of intelligent medical information is finally realized by comprehensively processing and cooperatively utilizing various systems and multi-element data related information of the Internet of things by associating, estimating and combining data of a plurality of information sources. In the process of medical first aid, because the condition of a patient is critical, how to provide a diagnosis and treatment scheme for the patient timely and reasonably according to the condition of the patient becomes a problem to be solved urgently.

Disclosure of Invention

In order to overcome the problems in the related art, the invention provides an emergency scheme recommendation method and device for an emergency.

According to a first aspect of an embodiment of the present invention, there is provided a method for recommending an emergency scenario of an emergency event, the method including:

acquiring electronic medical record information of an emergency wounded person in the process of carrying out emergency treatment on the emergency wounded person;

determining a target emergency type corresponding to the first-aid wounded person and a target level under the target emergency type according to the electronic medical record information and a pre-trained emergency classification model;

and outputting a corresponding target emergency scheme according to the type of the target emergency and the target level.

In one embodiment, preferably, the emergency event includes a nuclear biochemical emergency event, and the target emergency event type includes any one of the following: highly infectious viruses, biological toxins, biological pathogens, nerve agents, asphyxiating stimulants, erosive agents, systemic toxicants, intrinsic and exposure nuclear events, extrinsic exposure nuclear events and extrinsic exposure nuclear events, the target grade comprising: light grade, medium grade and heavy grade.

In one embodiment, preferably, the electronic medical record information includes: gender, age, medical history, exam report information, heart rate information, blood oxygen saturation information, respiratory rate, blood pressure, and body temperature.

In one embodiment, preferably, the emergency classification model includes a first-layer classification model and a second-layer classification model;

predicting a target emergency type corresponding to the first-aid wounded person by using the first-layer classification model and the electronic medical record information;

predicting a target level of the emergency responder under the target incident type using the two-layer classification model and the electronic medical record information.

In one embodiment, preferably, the training process of the emergency classification model includes:

acquiring medical record information and diagnosis information of patients with different historical emergencies;

identifying entities and entity relations in the medical record information by using an NLP information extraction method so as to perform structural processing on the medical record information to obtain the medical record information after the structural processing;

extracting all medical record features from the medical record information after the structured processing, processing abnormal values and missing values, selecting target medical record features with probability values smaller than a preset value as training features by using single factor analysis of statistical analysis, and putting the training features into a training feature set;

normalizing each training feature in the training feature set, and splicing into sample input data;

performing model training according to the sample input data, the emergency type in the diagnostic information and a convolutional neural network model to obtain the first-layer classification model;

performing model training according to the sample input data, the emergency type in the diagnosis information and a plurality of level classification models to obtain the two-layer classification model;

in one embodiment, preferably, the plurality of level classification models includes an XGBoost model, a random forest model, a support vector machine model, and a logistic regression model, and the method further includes:

calculating the model accuracy rate predicted by each level classification model, and determining the target level classification model with the highest model accuracy rate as an optimal model;

and taking the prediction result of the optimal model as the final prediction result of the two-layer classification model.

In one embodiment, preferably, performing model training according to the sample input data, the emergency event type in the diagnostic information, and a convolutional neural network model to obtain the top-layer classification model includes:

inputting the sample input data into a convolutional layer to obtain a first output result;

inputting the first output result into a pooling layer to obtain a second output result;

inputting the second output result to a full connection layer to obtain a third output result;

outputting the third output result to an integrated classifier to obtain an output result, wherein the output result comprises probability values of various emergency events;

and the output layer outputs the emergency type with the highest probability value, wherein the emergency type with the highest probability value is the target emergency type.

According to a second aspect of the embodiments of the present invention, there is provided an emergency scenario recommendation apparatus for an emergency, the apparatus including:

the acquisition module is used for acquiring the electronic medical record information of the first-aid wounded person in the process of carrying out first aid on the first-aid wounded person;

the determining module is used for determining a target emergency type corresponding to the first-aid wounded person and a target level under the target emergency type according to the electronic medical record information and a pre-trained emergency classification model;

and the output module is used for outputting a corresponding target emergency scheme according to the target emergency type and the target level.

in one embodiment, preferably, the plurality of level classification models include an XGBoost model, a random forest model, a support vector machine model, and a logistic regression model, and the apparatus further includes:

the calculation module is used for calculating the model accuracy rate predicted by each level classification model and determining the target level classification model with the highest model accuracy rate as the optimal model;

and the result determining module is used for taking the prediction result of the optimal model as the final prediction result of the two-layer classification model.

According to a third aspect of the embodiments of the present invention, there is provided an emergency scenario recommendation apparatus for an emergency, the apparatus including:

a processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to:

and outputting a corresponding target emergency scheme according to the target emergency type and the target level.

According to a fourth aspect of embodiments of the present invention, there is provided a computer-readable storage medium having stored thereon computer instructions which, when executed by a processor, implement the steps of the method of any one of the first aspects.

The technical scheme provided by the embodiment of the invention can have the following beneficial effects:

although the occurrence probability of the emergency is small, the emergency has great influence on the society and the rehabilitation of the patient, and in the process of transferring any disease, the transferring emergency doctor needs to record the state of an illness and vital signs of the patient, so that the transferring emergency doctor needs to obtain the optimal diagnosis and treatment scheme recommendation timely according to the real-time life improvement and the change of the state of the illness of the patient, further diffusion of the event is avoided, and the patient is helped to recover as soon as possible. In the embodiment, the optimal diagnosis and treatment scheme recommendation method for the emergency is provided, so that learning according to the historical optimal diagnosis and treatment scheme is realized in the emergency process, the recommendation of the emergency treatment scheme is obtained, and the treatment effect of the emergency is improved.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.

Fig. 1 is a flow chart illustrating a method for emergency scenario recommendation in an emergency event according to an exemplary embodiment.

FIG. 2 is a flow diagram illustrating a training process for an incident classification model according to an example embodiment.

FIG. 3 is a diagram illustrating a structure of a top-level classification model in accordance with an exemplary embodiment.

FIG. 4 is a block diagram illustrating an emergency scenario recommendation device in accordance with an exemplary embodiment.

Detailed Description

Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present invention.

Fig. 1 is a flowchart illustrating an emergency scenario recommendation method for an emergency event according to an exemplary embodiment, as shown in fig. 1, the method including:

step S101, acquiring electronic medical record information of an injured person in emergency in the process of carrying out emergency treatment on the injured person in emergency;

Step S102, determining a target emergency type corresponding to the first-aid wounded person and a target level under the target emergency type according to the electronic medical record information and a pre-trained emergency classification model;

In the embodiment, the complexity of the model can be reduced through layered modeling, the emergency is more in types and can be regarded as a multi-classification problem, the multi-classification problem can be converted into a two-classification problem whether the emergency is a certain definite emergency (known emergency) or not through the first layer of the layered modeling, and the early warning classification (light, medium and heavy and historical treatment schemes) according to historical learning results is recommended through the second layer, so that the complexity of the model is reduced.

And step S103, outputting a corresponding target emergency scheme according to the target emergency type and the target level.

step S201, acquiring medical record information and diagnosis information of patients with different historical emergencies;

medical record information is the most comprehensive and detailed record that describes the patient. Therefore, the patient condition information can be better reflected. The feature set of learning comes from basic information such as sex, age and the like of a patient, data such as medical history documents, examination and examination reports and the like, data such as heart rate, blood oxygen saturation, respiratory rate, blood pressure, body temperature and the like of an electrocardiogram monitor, and labels come from diagnostic information of the patient.

Step S202, an entity and an entity relation in the medical record information are identified by using an NLP information extraction method, so that the medical record information is subjected to structural processing, and the medical record information subjected to structural processing is obtained;

entities (symptoms, diseases, time, etc.) and entity relationships (accompanying time, occurrence location, etc.) in medical records are identified using NLP information extraction methods, enabling post-structuring of data.

Step S203, extracting all medical record features from the medical record information after the structured processing, processing abnormal values and missing values, selecting target medical record features with probability values smaller than a preset value as training features by using single factor analysis of statistical analysis, and putting the training features into a training feature set;

based on the structured data extraction features, because different doctors have differences in description of information such as symptoms and signs, synonym information in a knowledge base needs to be introduced to perform normalization operation on the symptoms, diseases, signs and the like, and results which are different in description but actually refer to the same object are normalized to be standard names. For example, the standard names of 'respiratory limitation', 'respiratory laboriousness', 'respiratory disorder' and the like are replaced by 'respiratory difficulty'. In addition, in order to more comprehensively and accurately express the patient's condition, the combination of characteristics is realized according to entity relations, such as that the main complaint ' cough 3 days, accompanied by expectoration 2 days ' contains two symptom entities of cough and expectoration, and two time entities of 3 days and 2 days. Wherein the physical relationship of cough and expectoration is concomitant, the physical relationship of 3 days to cough is duration, and the physical relationship of 2 days to expectoration is duration. Therefore, four symptoms of cough, 3 days of cough, expectoration and 2 days of expectoration are obtained.

All information of the whole patient record is used as input features, so the feature dimension is large and sparse, and the model training is slow. For example, the number of the variables initially included in organophosphorus is 69, and after the treatment of abnormal values and missing values of data, 14 feature sets with a p value less than 0.05 are included as final feature sets through single factor analysis of statistical analysis, which is specifically shown in table 1. The number of the variables originally included in the new crown is 83, 20 in the deletion proportion of more than 50%, 10 in the deletion proportion of 30% -49%, 9 in the deletion proportion of 20% -30% and 36 in the deletion proportion of less than 20%, the characteristics with the deletion rate of more than 50% are removed, and then 24 characteristics with the p value of less than 0.05 are selected as a final characteristic set by using single-factor analysis of statistical analysis after the abnormal values and the deletion values of the data are processed, as shown in the following table 2.

TABLE 1

TABLE 2

Step S204, each training feature in the training feature set is subjected to normalization processing and spliced into sample input data;

wherein discrete training features and continuous numerical features are processed separately.

And (3) for the continuous numerical characteristic, processing abnormal values, and filtering the interval which is obviously deviated from the normal numerical value. Then, in order to eliminate the adverse effect of different dimensions of different features on model training, continuous features need to be normalized, and the numerical values are normalized to be within a range of [0, 1], specifically, the used normalization method can be MinMaxScalter, the main idea is that after each continuous numerical value X is centralized according to the minimum value, the numerical value X is scaled according to the range of range (maximum value-minimum value), and the normalized specific formula is as follows:

for example, the continuous data processing of the electrocardiographic monitor such as body temperature, systolic pressure, diastolic pressure, etc. is shown in the following table 3.

TABLE 3

And performing One-Hot coding processing on the discrete features, and performing binary representation on the original type features in a high-dimensional space by 0/1. A1 indicates that the patient has this feature and a 0 indicates that the patient does not. As shown in table 4 below.

TABLE 4

And S205, performing model training according to the sample input data, the emergency type in the diagnostic information and a convolutional neural network model to obtain the first-layer classification model.

The classification model adopted in the first-layer classification model is a convolutional neural network model, as shown in fig. 3, firstly, all entities and entity relations are extracted from a database storing structured case information as input characteristics of the first-layer classification model, after the maximum word number of a sample is set, Zero-Padding is used for filling so as to ensure that the data length is consistent, and finally, the maximum sequence length is L. And forming a Word Embedding matrix through Word Embedding convolutional layers. The word vector model used in the invention is obtained by training a public open source tool Gensim module based on a plurality of hospital real Chinese electronic case data as a corpus, and each word is expressed by a 100-dimensional word vector.

A one-dimensional convolution mode is adopted in the convolution neural network. In the model structure selection, multiple factors such as the number of samples, the performance of hardware equipment, the complexity of the model, the characteristics of case data and the like are considered, and according to past experimental experience, a Grid Search (Grid Search) method is adopted to set multiple values of the same parameter in different value ranges and magnitude ranges in a descending order. Finally, 256 convolution kernels (filters) with the sizes of (L-1) × 100, (L-2) × 100, (L-3) × 100 are set by comparing the accuracy of the trained models on the test set. After the convolution layers are formed, 256 feature surfaces of 2 x 1, 3 x 1 and 4 x 1 are respectively obtained, and then a pooling layer (max-circulation) is added to perform dimensionality reduction operation on the features of the Filter layer. And splicing the pooled vectors through a full Connection (full Connection) layer to be used as the input of a Softmax layer, thereby realizing multi-classification prediction of various emergency events.

In the first layer model, the distribution of the medical records related to the emergency events and the common medical records is very unbalanced, the sample size of the non-emergency events is large, the output of the model is prone to the non-emergency events under the condition, and in order to reduce the influence of the unbalanced sample size among the classes on the classification result, an integrated idea can be adopted: and when the training set is generated every time, small sample quantities in all the classifications are used, and data are randomly extracted from large sample quantities in the classifications to be combined with the small sample quantities to form the training set, so that a plurality of training sets and training models can be obtained after the data are repeated for many times. Finally, when applied, a combination method (e.g., voting, weighted voting, etc.) is used to generate the classification prediction results.

Step S206, model training is carried out according to the sample input data, the emergency type in the diagnosis information and a plurality of level classification models to obtain the two-layer classification model.

calculating the model accuracy rate predicted by each level classification model, and determining the target level classification model with the highest model accuracy rate as the optimal model;

The main function of the two-layer classification model is to predict the probability that a patient belongs to a certain grade of an emergency, which is a multi-classification problem, the emergency is obtained by the first-layer model, and the classification model adopted by the two-layer model is tig.

1）XGBoost（Extreme Gradient Boosting）

Training a base learner from an initial training set by integrating the learning ideas of experts and Boosting; adjusting sample distribution according to the performance of the base learner, so that the samples with wrong classification get more attention in subsequent training; training the next base learner based on the adjusted sample adjustment: the above steps are repeated until the base learner reaches a specified number. Finally, when a new case is decided, the new case is decided by adopting a voting method or an average score method.

The advantages are that: the method has great advantages over other algorithms in many current data sets, supports high-speed parallel computation, and can automatically learn the splitting direction of a sample with missing characteristic values.

2) Random Forest (Random Forest, RF)

On the basis of building Bagging integration by taking a decision tree as a base learner, random attribute selection is further mapped in the training process of the decision tree by the RF, namely for each node of the base decision tree, a subset containing k attributes is randomly selected from an attribute set of the node, and then an optimal attribute is selected from the subset for division. Thus further improving the generalization performance.

The advantages are that: the method can process a large number of high-dimensional features, does not need dimension reduction, can evaluate the importance of each feature on the classification problem, and is insensitive to abnormal values and missing values.

3) Support Vector Machine (SVM)

The SVM is a generalized linear classifier for binary classification of data in a supervised learning mode, and a decision boundary of the SVM is a maximum margin hyperplane for solving learning samples. After reconstruction, the SVM algorithm can be used for multi-classification problems. One of the solutions is to implement the construction of a multi-classifier by combining a plurality of two classifiers, and the common methods include one-against-one and one-against-all.

4) Logistic Regression (LR)

The logistic regression model is a binary algorithm for processing binary labels based on the combination of a linear regression model and a Sigmoid activation function. The model is simple in structure, the speed block is trained, and the weight of the relatively deep neural network can be explained to be strong because the relatively deep neural network only has a single layer of weight. The range of values output by the model is in [0, 1], and can be considered as a probability of belonging to a certain class. In multi-classification research, strategies such as One VS One or One VS Rest are needed to convert a two-classification model into a multi-classification prediction framework.

The confusion matrix for multi-classification model prediction is shown in table 5 below (with three classes as an example):

TABLE 5

WhereinTP _kRepresenting a real objectSigning a class k, and predicting the model as the number of samples of the class k;E _k,irepresenting that the real label is k type and the model is predicted to be the sample number of i type; c denotes the total number of classes of the multi-classification.

According to the definition of the multi-classification confusion matrix, the Accuracy is used for evaluating the overall prediction performance of the model, and the specific formula is as follows:

in order to more fully embody the prediction performance of the multi-classification model, Precision (Precision) and Recall (Recall) are used for evaluation.

The formula for calculating the precision rate of the category k is as follows:

the recall ratio for category k is calculated as:

analyzing a model result:

in the experimental process, 4-fold cross validation is adopted for each model, in order to randomly divide experimental data into 4 parts, 3 parts of the data are used for model training, and the rest 1 part of the data are used for testing the model. After repeating 4 times, we obtained 4 models and its evaluation results.

The model accuracy rate accurve is used for evaluating the performance of the model, and the model with the highest accuracy rate is regarded as the optimal model. The new crown grade prediction test results are shown in table 6, wherein the accure of LR is 0.668, which is higher than the other three models, so LR is the best model. The organophosphorus hierarchical prediction test result is shown in table 7, and the accuracy of XGBOOST is 0.702, which is higher than the other three models, so XGBOOST is the best model.

TABLE 6

TABLE 7

an obtaining module 41, configured to obtain electronic medical record information of an injured person in first aid in a process of performing first aid by the injured person in first aid;

a determining module 42, configured to determine a target emergency type and a target level under the target emergency type corresponding to the emergency victim according to the electronic medical record information and a pre-trained emergency classification model;

and an output module 43, configured to output a corresponding target emergency scheme according to the target emergency type and the target level.

a processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to:

It is further understood that the term "plurality" means two or more, and other terms are analogous. "and/or" describes the association relationship of the associated objects, meaning that there may be three relationships, e.g., a and/or B, which may mean: a exists alone, A and B exist simultaneously, and B exists alone. The character "/" generally indicates that the former and latter associated objects are in an "or" relationship. The singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.

It will be further understood that the terms "first," "second," and the like are used to describe various information and that such information should not be limited by these terms. These terms are only used to distinguish one type of information from another, and do not indicate a particular order or degree of importance. Indeed, the terms "first," "second," and the like are fully interchangeable. For example, first information may also be referred to as second information, and similarly, second information may also be referred to as first information, without departing from the scope of the present invention.

It will be further appreciated that while operations are depicted in the drawings in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in serial order, or that all illustrated operations be performed, to achieve desirable results. In certain environments, multitasking and parallel processing may be advantageous.

Other embodiments of the invention will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention following, in general, the principles of the invention and including such departures from the present disclosure as come within known or customary practice within the art to which the invention pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the invention being indicated by the following claims.

It will be understood that the invention is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the invention is limited only by the appended claims.

Claims

1. An emergency scenario recommendation method for an emergency event, the method comprising:

2. The method of claim 1, wherein the emergency event comprises a nuclear biochemical emergency event, and the target emergency event type comprises any one of: highly infectious viruses, biological toxins, biological pathogens, nerve agents, asphyxiating stimulants, erosive agents, systemic toxicants, intrinsic and exposure nuclear events, extrinsic exposure nuclear events and extrinsic exposure nuclear events, the target grade comprising: light grade, medium grade and heavy grade.

3. The method of claim 1, wherein the electronic medical record information comprises: gender, age, medical history, exam report information, heart rate information, blood oxygen saturation information, respiratory rate, blood pressure, and body temperature.

4. The method of claim 1, wherein the emergency classification model comprises a first-level classification model and a second-level classification model;

5. The method of claim 4, wherein the training process of the emergency classification model comprises:

identifying entities and entity relations in the medical record information by using an NLP information extraction method so as to carry out structural processing on the medical record information to obtain the medical record information after the structural processing;

and performing model training according to the sample input data, the emergency type in the diagnosis information and a plurality of level classification models to obtain the two-layer classification model.

6. The method of claim 5, wherein the plurality of level classification models comprises an XGboost model, a random forest model, a support vector machine model, and a logistic regression model, the method further comprising:

7. The method of claim 5, wherein performing model training based on the sample input data, the type of emergency in the diagnostic information, and a convolutional neural network model to obtain the top-level classification model comprises:

8. An emergency scenario recommendation apparatus for an emergency event, the apparatus comprising:

9. An emergency scenario recommendation apparatus for an emergency event, the apparatus comprising:

a processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to:

10. A computer-readable storage medium having stored thereon computer instructions, which when executed by a processor, perform the steps of the method of any one of claims 1 to 7.