First-aid patient risk level assessment method and system based on attention mechanism

Abstract

The invention belongs to the technical field of intelligent medical treatment, and provides an attention mechanism-based first-aid patient risk level assessment method and system, and the method comprises the steps: obtaining the basic information of a first-aid patient and the historical data of physiological related information, carrying out the preprocessing and dividing of the data, and dividing the data into a training set, a verification set and a test set; constructing an evaluation model which is a multi-head self-attention mechanism model, training the evaluation model by using the training set, evaluating the trained evaluation by using the verification set and performing parameter tuning, and testing the performance of the evaluation model by using the test set; and acquiring basic information and physiological related information data of the first-aid patient, and processing the acquired data by using the evaluation model passing the performance test to obtain a risk level evaluation result of the first-aid patient. According to the invention, emergency pre-examination triage can be effectively assisted, and the problem of emergency delay is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent healthcare, and particularly relates to a method and system for evaluating the risk level of emergency patients based on an attention mechanism. Background Art

[0002] The statements in this part only provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] Currently, the emergency department has a large amount of medical data, which easily leads to a lack of timely emergency care due to a shortage of emergency resources. In addition, overcrowding in the emergency department is a long-term and serious problem currently faced by emergency departments globally, resulting in an increasing negative impact on the provision of reliable healthcare services. Longer patient waiting times may lead to overcrowding in the emergency department, resulting in a patient surge and treatment delays, which are intolerable for emergency patients.

[0004] Some hospitals have designed and launched an emergency pre-triage information system in accordance with the industry norms of emergency pre-triage. Emergency nurses complete the measurement, assessment, recording, and triage processes of patients within 2 - 5 minutes after receiving patients, which poses high requirements for nurses' triage work. The problem of inaccurate triage caused by the lack of experience of junior nurses occurs from time to time. The analysis of the triage data of a certain hospital shows that the accuracy rate of emergency triage is 80.51%, and the accuracy rates of the relatively high-level third and fourth triage results are 78.12%. The correction rate of manual intervention for the third and fourth triage results is relatively high, and there is a potential risk of delaying the timely treatment of critically ill patients due to inaccurate grading. Therefore, exploring the establishment of a more intelligent emergency pre-triage assistance tool to provide accurate triage suggestions for the staff at the triage desk is of great significance for improving the emergency triage process. Summary of the Invention

[0005] In order to solve the above problems, the present invention proposes a method and system for evaluating the risk level of emergency patients based on an attention mechanism.

[0006] According to some embodiments, the present invention adopts the following technical solutions: A method for evaluating the risk level of emergency patients based on an attention mechanism, comprising the following steps: Obtain the basic information of emergency patients and historical data of physiological-related information, preprocess and partition the data into a training set, a validation set, and a test set; Construct an evaluation model, where the evaluation model is a multi-head self-attention mechanism model, train the evaluation model using the training set, tune the parameters of the trained model using the validation set, and test the performance of the evaluation model using the test set; Obtain the basic information and physiological-related information data of the first-aid patient, and use the evaluation model that has passed the performance test to process the obtained data to obtain the evaluation result of the first-aid patient risk level.

[0007] As an alternative implementation, the basic information of the first-aid patient includes age, gender, ethnicity, time from onset to admission, the ability of the patient and family members to communicate normally, and clinical data.

[0008] As an alternative implementation, the physiological-related information includes the patient's disease type, heart rate, diastolic blood pressure, systolic blood pressure, mental state, blood oxygen saturation, body temperature, pulse pressure, shock index, and mean blood pressure.

[0009] As an alternative implementation, the process of preprocessing the data includes: Clean and filter the data set, delete the incomplete data tuples in the data set, and delete the data in the data set that does not meet the standard data tuples; For the data in the data set where some attributes are continuous, calculate its mean, variance, maximum value, and minimum value, and perform standardization, dimensionality reduction, and normalization processing according to the calculation results; For the data in the data set where some attributes are discrete, calculate the number of discrete data, and perform classification processing based on the number of different types of discrete data.

[0010] As an alternative implementation, the specific process of constructing the evaluation model includes: initializing the multi-head attention model, and the main structure of the model includes three parts: an embedding layer, a multi-head attention pre-feedback layer, and an encoding layer connected in sequence.

[0011] As an alternative implementation, the specific process of training the evaluation model using the training set includes: the embedding layer is used to load the data set into the model, and the original baseline covariates are used to describe the characteristics of the patient. Let and be the number of categorical covariates and numerical covariates respectively. For discrete data, perform classification processing and normalization and standardization processing, and represent each variable in the dimensional space through the embedding matrix; The multi-head attention pre-feedback layer is used to make the variable parameters have sufficient interaction between embeddings using the multi-head self-attention method. The key-value attention mechanism allows the model to automatically learn the combined interaction and obtain the output embedding of the transformer layer; The encoder is used to generate the final representation of the patient.

[0012] Furthermore, represent the number of covariates as + Before entering the survival model, the categorical covariates are transformed into 0-1 vectors.

[0013] As a further step, each variable is represented in a -dimensional space through an embedding matrix: ; where is the 0-1 vector of the category field , and is the obtained embedding; To allow interactions between numerical covariates and categorical covariates, in the low-dimensional space, use: ; where is the scalar of the rd numerical value, is the embedding matrix of the numerical feature, is 's st row. For the above two types of embeddings, connect them: ; to obtain the representation of all the original input patient-related information parameters of the th patient .

[0014] As a further step, SELU represents the scaled exponential linear unit activation function, and the output embedding of the transformer layer is obtained through another -layer feed-forward network with residual connections: ; Convert the original embedding to the attention embedding ; Stack transformers: ; .

[0015] As an alternative implementation, the process of processing the acquired data using an evaluation model that has passed performance testing includes: (1) Save and export the trained attention mechanism model; (2) Build a first-aid patient risk level classification system based on the attention mechanism and provide an import channel for the patient's basic information data; (3) Store the model in the background server. The system can directly connect with the model by preloading the model from a file during startup, call the prediction function of the model, and be able to call the prediction classification interface of the model to output the risk level of the patient; (4) The system collects and obtains the relevant data of the first-aid patient, uses it as an influencing factor to input into the model, and calls the model for prediction.

[0016] A first aid patient risk level assessment system based on an attention mechanism, comprising: A data acquisition module configured to acquire basic information of a first aid patient and historical data of physiological-related information, preprocess and divide the data into a training set, a validation set, and a test set; A model construction and training module configured to construct an evaluation model, the evaluation model being a multi-head self-attention mechanism model, training the evaluation model using the training set, tuning parameters of the trained evaluation model using the validation set, and testing the performance of the evaluation model using the test set; An evaluation module configured to acquire basic information of a first aid patient and physiological-related information data, and process the acquired data using the evaluation model that has passed the performance test to obtain a first aid patient risk level assessment result.

[0017] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps in the above method.

[0018] An electronic device comprising a memory, a processor, and computer instructions stored on the memory and running on the processor, which, when run by the processor, complete the steps in the above method.

[0019] Compared with the prior art, the beneficial effects of the present invention are: The present invention divides the risk level of first aid patients based on an attention mechanism, and the division result is accurate, which can effectively assist in emergency pre-triage, provide accurate triage suggestions for the staff at the triage desk, and is of great significance for improving the emergency triage process.

[0020] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following specifically enumerates preferred embodiments and, in conjunction with the accompanying drawings, makes a detailed description as follows. Description of the Drawings

[0021] The specification drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention.

[0022] Figure 1 is a schematic flowchart of the method of this embodiment; Figure 2 is a model structure diagram of this embodiment. Detailed Embodiments

[0023] The present invention will be further described below in conjunction with the drawings and embodiments.

[0024] It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention pertains.

[0025] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0026] In the case of no conflict, the embodiments in this application and the features in the embodiments can be combined with each other.

[0027] Embodiment 1 A method for evaluating the risk level of emergency patients based on the attention mechanism, as Figure 1 shown, includes the following steps: Step 1: Obtain the emergency patient dataset and perform preprocessing; Step 2: Perform model pre-training on the emergency patient dataset; Step 3: Integrate and use the trained attention mechanism model; Step 4: Divide the risk levels of the model for emergency patients.

[0028] The specific process of Step 1 to obtain the emergency patient dataset and perform preprocessing is as follows: (1) Obtain the historical dataset of emergency patients, including basic information such as age, gender, ethnicity, time from onset to admission, ability of the patient and family members to communicate normally, and complete clinical data. In addition, other information includes relevant indicators such as the patient's disease type, heart rate, diastolic blood pressure, systolic blood pressure, mental state, blood oxygen saturation, body temperature, pulse pressure, shock index, and mean blood pressure as the influencing factors of the dataset model for emergency patients; (2) Clean and filter the dataset. First, delete the incomplete data tuples in the dataset and delete the data tuples that do not meet the standards in the dataset; (3) For some attributes in the dataset that are continuous data, calculate their inherent attributes such as mean, variance, maximum value, and minimum value, and on this basis, perform standardization, dimensionality reduction, and normalization processing; (4) For some attributes in the dataset that are discrete data, calculate the number of discrete data, and perform classification processing based on the number of different types of discrete data, and divide the original data into different datasets, which are used as the training set, validation set, and test set respectively.

[0029] The specific process of the above-mentioned step 2 for pre-training the model on the first-aid patient dataset is as follows: (1) Initialize the model, input the basic attribute variables of the model, set the network structure and loss function of the model, and set the training step size and learning rate of the model; The main structure of the model consists of three parts: an embedding layer, a multi-head attention feed-forward layer, and an encoding layer, which are connected in sequence.

[0030] (2) Input and embed the model, load the dataset into the model. As Figure 2 shown, the original baseline covariates describe the characteristics of the patients. In this embodiment, and are set as the numbers of categorical covariates and numerical covariates respectively. The categorical covariates and numerical covariates are set as patient-related information of discrete type and continuous type respectively, mainly for differentiating the data. For discrete data, that is, the data for classifying patients, for example, binary classification such as whether the patient has hypertension or diabetes, or the fracture grade of the patient for multi-classification. For continuous data, that is, the data for describing patient information, mainly for normalizing and standardizing the data, such as age, height, etc.

[0031] The number of covariates is expressed as + . Before entering the survival model, the categorical covariates are transformed into 0-1 vectors. In this embodiment, they are represented in the in dimensional space by the embedding matrix: ; Among them, is the 0-1 vector of the category field , and is the obtained embedding; (3) To allow the interaction between numerical covariates and categorical covariates, in this embodiment, in the low-dimensional space, use: ; Among them, is the scalar of the th numerical value, is the embedding matrix of the numerical features, is the embedding matrix 's th row. With these two types of embeddings, in this embodiment, they can be concatenated, ; to obtain the representation of all the original input covariates of the th patient ; (4) To enable sufficient interaction between covariate embeddings, this embodiment uses the multi-head self-attention method. The key-value attention mechanism allows the model to automatically learn combinatorial interactions. SELU represents the Scaled Exponential Linear Unit (SELU) activation function, and then through another feed-forward network layer (FFN) to obtain the output embedding of the transformer layer: ; In short, starting from the first transformer, this embodiment converts the original embedding into an attention embedding . To encourage further interaction between covariates to obtain high-order combinatorial embeddings, this embodiment can stack transformers: ; Therefore, is the final representation of the patient generated by stacking transformer encoders.

[0032] Covariates refer to the input parameters of the model, including two types: discrete and continuous. Discrete types are mainly used for classification, and continuous types are mainly standardized to improve the accuracy of the model.

[0033] The specific process of integrating and using the trained attention mechanism model in step 3 is as follows: (1) Save and export the trained attention mechanism model; (2) Build a first-aid patient risk level classification system based on the attention mechanism, and provide channels for importing, accessing, and inputting the patient's basic information data; (3) Store the model in the background server. The system can directly connect to the model by preloading the model during startup and can call the prediction and classification function of the model; (4) Collect and obtain relevant data of first-aid patients through the system, use it as an influencing factor to input into the model, and call the prediction method of the model to implement the prediction function of integrating and using the model.

[0034] The specific process of classifying the model risk level of first-aid patients in step 4 is as follows: (1) Collect and input the data of first-aid patients into the model by step 3, and output the risk probability and risk level of the patient through the prediction method of the model; (2) Visualize the patient's risk level and output the auxiliary recommended treatment plan for the patient according to the risk level to provide auxiliary triage assistance for emergency medical staff.

[0035] The following product embodiments are also provided: A first-aid patient risk level assessment system based on the attention mechanism, including: A data acquisition module, configured to acquire basic information of emergency patients and historical data of physiological-related information, preprocess and partition the data into a training set, a validation set, and a test set; A model construction and training module, configured to construct an evaluation model, where the evaluation model is a multi-head self-attention mechanism model, train the evaluation model using the training set, perform parameter tuning on the trained model using the validation set, and test the performance of the evaluation model using the test set; An evaluation module, configured to acquire basic information of emergency patients and physiological-related information data, and use the evaluation model that has passed the performance test to process the acquired data to obtain an evaluation result of the risk level of emergency patients.

[0036] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps in the above method.

[0037] An electronic device, including a memory, a processor, and computer instructions stored on the memory and running on the processor, which, when executed by the processor, complete the steps in the above method.

[0038] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD - ROM , optical memory, etc.).

[0039] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one or more flows or multiple flows and / or blocks Figure 1 one or more blocks or multiple blocks.

[0040] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction device that implements the functions specified in one or more of the procedures Figure 1 one or more procedures and / or blocks Figure 1 specified in the block or blocks.

[0041] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one or more of the procedures Figure 1 one or more procedures and / or blocks Figure 1 specified in the block or blocks.

[0042] The foregoing are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc. made by those skilled in the art without creative efforts within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for assessing the risk level of emergency patients based on an attention mechanism, characterized in that: The following steps are involved: Obtain basic information of emergency patients and historical data of physiological related information, pre-process and divide the data into training set, validation set and test set; Construct an evaluation model, wherein the evaluation model is a multi-head self-attention mechanism model, use a training set to train the evaluation model, use a validation set to evaluate the training and perform parameter tuning, and use a test set to test the performance of the evaluation model; The basic information and physiological related information data of the emergency patients are obtained, and the obtained data are processed using the evaluation model that has passed the performance test to obtain the risk level assessment results of the emergency patients.

2. A method for assessing the risk level of emergency patients based on an attention mechanism as claimed in claim 1, characterized in that: The basic information of emergency patients includes age, gender, ethnicity, time from onset to admission, whether the patient and family members can communicate normally, and clinical data.

3. The method for assessing the risk level of emergency patients based on the attention mechanism as claimed in claim 1, characterized in that: The physiological related information includes the patient's disease type, heart rate, diastolic blood pressure, systolic blood pressure, mental expression, blood oxygen saturation, body temperature, pulse pressure, shock index and average blood pressure.

4. The method for assessing the risk level of emergency patients based on the attention mechanism as claimed in claim 1, characterized in that: The process of preprocessing data includes: Clean and filter the data set, delete incomplete data tuples in the data set, and delete data that does not meet the standard data tuples in the data set; For data with continuous attributes in the data set, calculate the mean, variance, maximum and minimum values, and perform standardization, dimension reduction and normalization based on the calculation results; For data whose attributes are discrete in some data sets, the number of discrete data is calculated, and classification processing is performed based on the number of different types of discrete data.

5. The method for assessing the risk level of emergency patients based on the attention mechanism as claimed in claim 1, characterized in that: The specific process of building the evaluation model includes: initializing the multi-head attention model. The main structure of the model consists of three parts: an embedding layer, a multi-head attention feedback layer, and an encoding layer connected in sequence.

6. A method for assessing the risk level of emergency patients based on an attention mechanism as claimed in claim 1 or 5, characterized in that: The specific process of training the evaluation model using the training set includes: the embedding layer is used to load the data set into the model, the original baseline covariates are used to describe the characteristics of the patients, and Set as the number of categorical covariates and numerical covariates respectively. For discrete data, classification processing is performed and normalization and annotation processing is performed. dimensional space to represent each variable; The multi-head attention feedforward layer is used to make full interactions between the variable parameter embeddings using the multi-head self-attention method. The key value attention mechanism allows the model to automatically learn the combined interactions and obtain the output embedding of the transformer layer; The encoder is used to generate a final representation of the patient.

7. A method for assessing the risk level of emergency patients based on an attention mechanism as claimed in claim 6, characterized in that: Denote the number of covariates as + , categorical covariates are transformed into 0-1 vectors before entering the survival model; Dimensional space represents each variable: ; in, Is a category field A 0-1 vector, is the resulting embedding; To allow for interactions between numerical and categorical covariates, we use: ; in It is A scalar with a value, is the embedding matrix of numerical features, yes No. Line, the above two types of embedding, concatenate them: ； To obtain the Representation of all original input patient-related information parameters for a patient .

8. A method for assessing the risk level of emergency patients based on an attention mechanism as claimed in claim 7, characterized in that SE LU stands for the scaled exponential linear unit activation function, through another with a residual connection The layer feed-forward network obtains the output embedding of the transformer layer: ； Embed the original Convert to Attention Embedding ; Stack transformer: ; .

9. The method for assessing the risk level of emergency patients based on the attention mechanism as claimed in claim 1, characterized in that: Using the evaluation model that has passed the performance test, the process of processing the acquired data includes: (1) Save and export the trained attention mechanism model; (2) Build an attention-based system for grading the risk of emergency patients and provide a channel for importing basic patient information data; (3) The model is stored in the backend server. The system preloads the model from the file during startup, and the system and the model can be directly connected to call the prediction function of the model. The prediction classification interface of the model can be called to output the patient's risk level; (4) The system collects and obtains data related to emergency patients, inputs it into the model as an influencing factor, and calls the model for prediction.

10. A system for assessing the risk level of emergency patients based on an attention mechanism, characterized in that: include: The data acquisition module is configured to obtain basic information of emergency patients and historical data of physiological related information, and pre-process and divide the data into a training set, a validation set, and a test set; A model building and training module is configured to build an evaluation model, wherein the evaluation model is a multi-head self-attention mechanism model, train the evaluation model using a training set, evaluate the trained model using a validation set and perform parameter tuning, and test the performance of the evaluation model using a test set; The evaluation module is configured to obtain basic information and physiological related information data of emergency patients, and use the evaluation model that has passed the performance test to process the acquired data to obtain the risk level evaluation result of the emergency patients.

Images