Intelligent clinical support decision-making system and method for emergency triage

Abstract

The invention relates to the field of intelligent medical treatment, in particular to an intelligent clinical support decision-making system and method for emergency triage. The intelligent clinical support decision-making system for emergency triage comprises a medical resource adjustment module, a patient risk assessment module and a patient disease triage module. According to the method, the historical case data is analyzed, the relationship between various symptoms and potential diseases is identified, the occurrence frequency of the symptoms in different diseases is deeply analyzed, the corresponding weight value is allocated to each symptom, and the weight values are applied to the disease triage model, so that the triage accuracy is improved, and the disease triage efficiency is improved. Medical staff can be helped to quickly identify possible disease types of the patient, and the risk of misdiagnosis and missed diagnosis is reduced; by means of the method, doctors in the emergency department can make triage decisions more confidently, and therefore the treatment experience and the treatment effect of patients are improved.

Description

Technical Field

[0001] The present invention relates to the field of smart medical care, and in particular to an intelligent clinical support decision-making system and method for emergency triage. Background Art

[0002] In the field of emergency medicine, accurate and rapid triage is crucial for timely treatment of patients; emergency doctors are often faced with the need to quickly identify patients' symptoms and determine priorities; a common challenge is that many symptoms are common in different diseases, making it difficult to accurately determine which specific disease a patient has based on symptoms alone; for example, fever may be caused by a variety of infectious diseases, while chest pain may be related to heart disease or respiratory disease; the non-specificity of this symptom requires that the emergency triage system must be able to comprehensively consider multiple factors to improve the accuracy and efficiency of diagnosis.

[0003] In the traditional emergency triage process, medical staff mainly rely on personal experience and intuition to judge the relationship between patients' symptoms and potential diseases; although this method is flexible, it has certain limitations, including strong subjectivity, low efficiency and easy errors; therefore, the present invention proposes an intelligent clinical support decision system and method for emergency triage, providing a more accurate and efficient emergency triage method; the system will provide patients with accurate disease triage and department recommendation by deeply analyzing historical case data and combining information such as patients' symptoms, signs, age and medical history; this method is expected to significantly improve the accuracy of emergency triage, reduce misdiagnosis and missed diagnosis, and thus improve the overall quality of emergency medical services. Summary of the invention

[0004] The present invention analyzes historical case data, identifies the relationship between various symptoms and potential diseases, conducts in-depth analysis of the frequency of occurrence of symptoms in different diseases, assigns a corresponding weight value to each symptom, and applies the weight value to the disease triage model, thereby improving the accuracy of triage. It can also help medical staff quickly identify the possible disease type of the patient and reduce the risk of misdiagnosis and missed diagnosis. Through this method, emergency doctors can make triage decisions with greater confidence, thereby improving patients' medical experience and treatment effects.

[0005] Intelligent clinical decision support methods for emergency triage, including: In recent For any type of disease, based on the disease type in the most recent The number of patients per day in a day is used to determine whether the disease type has entered a trending stage; for any disease type in a trending stage, the number of patients of the disease type in the next day is predicted, and the predicted value of the number of patients of the disease type in the next day is obtained. Based on the predicted value of the number of patients, the medical resources of the corresponding department of the disease type are adjusted; When the hospital emergency department receives a new patient, the patient's vital signs data is obtained, including the patient's heart rate, blood pressure, respiratory rate, blood oxygen saturation and consciousness level; the vital signs data is used as the input of the risk assessment model, and the probability of each risk level of the patient is output. The risk levels include level 1 risk, level 2 risk, level 3 risk and level 4 risk, and the patient's risk level with the highest probability is the patient's risk level; the risk level of the hospital emergency department patients is used to determine the order in which each patient is treated; Obtain a historical case data set, and calculate triage weight values ​​of all symptoms existing in the historical case data set; If the patient's risk level is level one, emergency measures are taken immediately; if the patient's risk level is level two, three or four, the patient's symptom data, age data and medical history data are obtained, and the symptom data includes several symptoms that the patient has; multi-hot encoding is applied to the symptom data and medical history data to generate symptom input vectors and medical history input vectors respectively, and the age data is normalized at the same time; the symptom input vector, the triage weight value corresponding to the symptom input vector, the medical history input vector and the normalized age data are used as inputs to the disease triage model, and the disease triage result is output. The disease triage result is the patient's disease type, and the corresponding treatment department is recommended to the patient based on the patient's disease type.

[0006] Preferably, for any type of disease, based on the disease type in the recent The number of patients per day determines whether the disease type has entered the trend stage. The specific operations are as follows: Record the number of patients admitted to the hospital emergency department and the type of illness each patient has every day. days as the current sliding time window; for any type of disease, calculate the current disease type before the sliding time window -The average number of patients per day within 3 days, and use the calculated result as the benchmark value b of the current sliding time window; for any day within 3 days after the sliding time window, use the formula Calculate the first Daily variation ;in, Indicates that the current disease type is the first in the sliding time window. Number of patients per day, =1, 2, …, ; If the calculated change amplitudes of the last three days are all greater than the preset increase threshold, the current symptom type enters the trend stage; For any symptom type in the trend stage, if the change amplitude of the symptom type in the last 3 days within the sliding time window is lower than the preset decrease threshold, the trend stage of the symptom type ends.

[0007] Preferably, for any disease type in the trending stage, the number of patients with the disease type in the next day is predicted to obtain the predicted value of the number of patients with the disease type in the next day. The specific operation is as follows: Using the formula Calculate the predicted value of the number of patients one day after acquisition ;in, Indicates the first The data weight of the day, Select and obtain through the group optimization algorithm, .

[0008] Preferably, a historical case data set is obtained, and the triage weight values ​​of all symptoms existing in the historical case data set are calculated and obtained. The specific operations are as follows: Obtain a historical case data set, and for any symptom that has appeared in the historical case data set, extract the department where the current symptom exists. Based on the total number of times the current symptom has appeared and the number of times the current symptom has appeared in each department, calculate the proportion of the current symptom in each department. ; =1, 2, …, ; Indicates the number of departments with current symptoms, ; Using the formula Calculate and obtain the concentration C of the current symptoms. The higher the concentration C, the more the current symptoms are concentrated in a few departments. Based on the concentration of all symptoms obtained, select the maximum concentration and minimum concentration ; Then use the formula Calculate and obtain the triage weight value W of the current symptom.

[0009] Preferably, the disease triage model is established based on the MLP model, including a first input layer, a symptom weighting layer, a feature extraction layer, a feature fusion layer, a hidden layer and a first output layer; the first input layer is used to input a symptom input vector, a triage weight value corresponding to the symptom input vector, a medical history input vector and normalized age data; the symptom weighting layer is used to apply the triage weight value to weight the input symptom input vector; the feature extraction layer includes a symptom branch, an age branch and a medical history branch, which are respectively used to extract features of the symptom input vector, the age data and the medical history input vector, and obtain a symptom feature vector, an age feature vector and a medical history feature vector; the feature fusion layer is used to splice and fuse the symptom feature vector, the age feature vector and the medical history feature vector to obtain a fusion feature; the hidden layer is used to further process the fusion feature and output the optimized final fusion feature; the first output layer is used to apply the softmax activation function to output the probability prediction of each disease type, and the disease category with the highest probability is the disease triage result; The specific operations for training the disease triage model are as follows: Obtain several disease triage training samples with labeled disease triage results, each disease triage training sample comprising a patient's symptom input vector, a triage weight value corresponding to the symptom input vector, a medical history input vector and normalized age data; divide all disease triage training samples into a disease triage training set and a disease triage verification set, use the disease triage training set to train the disease triage model, and verify the disease triage model through the disease triage verification set to obtain a first verification result; set a first training condition, determine whether the first verification result obtained meets the first training condition, and if so, output the trained disease triage model; if not, continue to use the disease triage training set to train the disease triage model.

[0010] Preferably, the risk assessment model is established based on the CNN model, including a second input layer, a convolution layer, a pooling layer, a fully connected layer and a second output layer; the second input layer is used to receive the patient's vital signs data; the convolution layer is used to extract local features of the patient's vital signs data; the pooling layer is used to perform maximum pooling on the output of the convolution layer to reduce the feature dimension; the fully connected layer is used to flatten the output of the pooling layer into a one-dimensional vector, perform linear transformation and nonlinear activation, integrate features and generate high-level representations; the second output layer is used to map the output of the fully connected layer to the probability of 4 risk levels, use the Softmax function to normalize the output to a probability distribution, obtain the probability of each risk level, and output the risk assessment result, which is the probability of each risk level; The specific operations for training the risk assessment model are as follows: Obtain risk assessment training samples with labeled risk assessment results, each risk assessment training sample containing vital signs data of a patient; divide all acquired risk assessment training samples into a risk assessment training set and a risk assessment verification set; train the risk assessment model using the risk assessment training set, and verify the risk assessment model using the risk assessment verification set to obtain a second verification result; set a second training condition to determine whether the acquired second verification result satisfies the second training condition, and if so, output the trained risk assessment model; if not, continue to train the risk assessment model using the risk assessment training set.

[0011] Preferably, a swarm optimization algorithm is used to select and obtain the Data weight of the day , the specific operations are as follows: Step 1: Randomly generate a particle swarm, the position of each particle represents a candidate weight combination, and set the maximum number of iterations; Step 2: For any particle, initialize the velocity of the particle and use the current position as the individual optimal position of the particle; calculate the inverse of the prediction error when applying the corresponding weight combination of the particle, and use the inverse of the prediction error obtained by calculation as the fitness value of the particle; traverse all particles and use the position corresponding to the particle with the highest fitness value as the global optimal position; Step 3: For any particle, update the speed and position of the particle, and recalculate the fitness value; if the updated fitness value is higher than the fitness value of the individual best position of the particle, update the individual best position of the particle to the current position; Step 4: Traverse all particles. If the fitness value of the individual best position of any particle is higher than the fitness value of the global best position, update the global best position to the individual best position of the particle. Step 5: Repeat steps 3-4 until the maximum number of iterations is reached. The global best position obtained is the best weight combination. .

[0012] Preferably, the intelligent clinical decision support system for emergency triage includes: The medical resource adjustment module includes a trend disease identification unit, a patient quantity prediction unit and a medical resource adjustment unit; the trend disease identification unit is used to determine the trend of any disease type based on the recent trend of the disease type. The number of patients per day in a day is used to determine whether the disease type has entered a trending stage; the patient quantity prediction unit is used to predict the number of patients of any disease type in a trending stage in the next day, and obtain the predicted value of the number of patients of the disease type in the next day; the medical resource adjustment unit is used to adjust the medical resources of the department corresponding to a certain disease type based on the predicted value of the number of patients of the disease type; The patient risk assessment module is used to obtain the patient's vital signs data and use the vital signs data as the input of the risk assessment model to output the probability of each risk level of the patient. The risk levels include level 1 risk, level 2 risk, level 3 risk and level 4 risk. The risk level with the highest probability is the patient's risk level. The patient symptom triage module is used to calculate the triage weight values ​​of all symptoms existing in the historical case data set; when the patient's risk level is level one, emergency measures are taken immediately; when the patient's risk level is level two, three or four, the patient's symptom data, age data and medical history data are obtained; multi-hot encoding is applied to the symptom data and medical history data to generate symptom input vectors and medical history input vectors respectively, and the age data is normalized at the same time; the symptom input vector, the triage weight value corresponding to the symptom input vector, the medical history input vector and the normalized age data are used as the input of the symptom triage model, the symptom triage results are output, and the corresponding treatment department is recommended for the patient based on the obtained symptom triage results.

[0013] The present invention has the following advantages: The present invention analyzes historical case data, identifies the relationship between various symptoms and potential diseases, conducts in-depth analysis of the frequency of occurrence of symptoms in different diseases, assigns a corresponding weight value to each symptom, and applies the weight value to the disease triage model, thereby improving the accuracy of triage. It can also help medical staff quickly identify the possible disease type of the patient and reduce the risk of misdiagnosis and missed diagnosis. Through this method, emergency doctors can make triage decisions with greater confidence, thereby improving patients' medical experience and treatment effects.

[0014] The present invention can dynamically adjust the allocation of medical resources by real-time monitoring and predicting changes in the number of patients in the emergency department; based on the prediction of the number of patients and the trend analysis of the types of symptoms, it can promptly identify departments that need additional resources, thereby effectively avoiding waste and shortage of medical resources; this optimized resource allocation not only improves the work efficiency of the emergency department, but also ensures that patients can obtain the required medical services in the shortest time; through the reasonable allocation of medical resources, it helps to improve the overall quality of emergency medical services and meet the needs of patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1This is a schematic diagram of the structure of an intelligent clinical decision support system for emergency triage adopted in an embodiment of the present invention. DETAILED DESCRIPTION

[0016] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0017] Embodiment 1, an intelligent clinical decision support method for emergency triage, comprising: In recent For any type of disease, based on the disease type in the most recent The number of patients per day in a day determines whether the disease type has entered a trend stage; by counting the patient data received every day, the system can identify which disease types have a significant increase or decrease in the number of patients; this trend analysis is based on the dynamic monitoring of time series data, aiming to identify possible epidemiological changes or seasonal fluctuations. For example, in autumn and winter, especially in winter, the spread of influenza virus will lead to a significant increase in the number of patients with respiratory diseases; in snowy weather, the number of patients with trauma caused by traffic accidents and falls will increase due to ice or snow on the road; extremely cold weather may increase the severity of patients with heart diseases, such as myocardial infarction and heart attack; for any disease type in the trend stage, the number of patients of this disease type in the next day is predicted, and the predicted value of the number of patients of this disease type in the next day is obtained. Based on the predicted value of the number of patients, the medical resources of the corresponding department of this disease type are adjusted, such as adjusting the reception time, adding temporary beds, etc., to reduce patient waiting time and improve service quality. Reasonable allocation of resources helps control hospital operating costs and avoid financial pressure caused by over-allocation of resources; When the hospital emergency department receives a new patient, the patient's vital signs data is obtained. The vital signs data include the patient's heart rate, blood pressure, respiratory rate, blood oxygen saturation and consciousness level. These indicators can provide medical staff with direct information on the patient's current health status and are the key to assessing the patient's urgency and stability. The vital signs data is used as the input of the risk assessment model to output the probability of each risk level of the patient. The risk levels include level 1 risk, level 2 risk, level 3 risk and level 4 risk. The patient's risk level is the one with the highest probability. The risk assessment is performed through the model to reduce the bias of human subjective judgment and improve the objectivity and accuracy of the assessment. The risk level of the patients in the hospital emergency department is used to determine the order of each patient's visit. Clear risk levels help optimize resource allocation and ensure that emergency patients can receive timely treatment. Obtain a historical case data set, and calculate the triage weight values ​​of all symptoms in the historical case data set; the reason for calculating the triage weight value of each symptom is that many symptoms are not specific to a single disease, but are common in multiple diseases; for example, symptoms such as fever, cough, or abdominal pain can be caused by a variety of different diseases, which makes it difficult to accurately determine which specific disease the patient has based on symptoms alone; by calculating the triage weight value of the symptom, it can help medical staff to more accurately divert patients to the appropriate department and improve the accuracy and efficiency of triage; this method helps to solve the problem of non-specific symptoms, optimize the allocation of resources in the emergency department, and ultimately improve the diagnosis and treatment effect of patients; If the patient's risk level is level one, emergency measures are taken immediately; if the patient's risk level is level two, three or four, the patient's symptom data, age data and medical history data are obtained, and the symptom data includes several symptoms that the patient has; multi-hot encoding is applied to the symptom data and medical history data to generate symptom input vectors and medical history input vectors respectively, and the age data is normalized at the same time; the symptom input vector, the triage weight value corresponding to the symptom input vector, the medical history input vector and the normalized age data are used as inputs to the disease triage model, and the disease triage result is output. The disease triage result is the patient's disease type, and the corresponding treatment department is recommended to the patient based on the patient's disease type.

[0018] For any disease type, based on the disease type in the recent The number of patients per day determines whether the disease type has entered the trend stage. The specific operations are as follows: Record the number of patients admitted to the hospital emergency department and the type of illness each patient has every day. days as the current sliding time window; for any type of disease, calculate the current disease type before the sliding time window -The average number of patients per day within 3 days, and use the calculated result as the benchmark value b of the current sliding time window; for any day within 3 days after the sliding time window, use the formula Calculate the first Daily variation ;in, Indicates that the current disease type is the first in the sliding time window. The number of patients per day, =1, 2, …, ; If the calculated change amplitudes of the last three days are all greater than the preset increase threshold, the current symptom type enters the trend stage; For any symptom type in the trend stage, if the change amplitude of the symptom type in the last 3 days within the sliding time window is lower than the preset decrease threshold, the trend stage of the symptom type ends.

[0019] For any disease type in the trending stage, predict the number of patients with this disease type in the next day, and obtain the predicted value of the number of patients with this disease type in the next day. The specific operations are as follows: Using the formula Calculate the predicted value of the number of patients 1 day after acquisition ;in, Indicates the first The data weight of the day, Select and obtain through the group optimization algorithm, .

[0020] Obtain a historical case data set, and calculate the triage weight values ​​of all symptoms in the historical case data set. The specific operations are as follows: Obtain a historical case data set, and for any symptom that has appeared in the historical case data set, extract the department where the current symptom exists. Based on the total number of times the current symptom has appeared and the number of times the current symptom has appeared in each department, calculate the proportion of the current symptom in each department. ; =1, 2, …, ; Indicates the number of departments with current symptoms, ; Using the formula Calculate and obtain the concentration C of the current symptoms. The higher the concentration C, the more the current symptoms are concentrated in a few departments. Based on the concentration of all symptoms obtained, select the maximum concentration and minimum concentration ; Then use the formula Calculate and obtain the triage weight value W of the current symptom.

[0021] The disease triage model is established based on the MLP model, including the first input layer, symptom weighting layer, feature extraction layer, feature fusion layer, hidden layer and the first output layer; the first input layer is used to input the symptom input vector, the triage weight value corresponding to the symptom input vector, the medical history input vector and the normalized age data; the symptom weighting layer is used to apply the triage weight value to weight the input symptom input vector; the feature extraction layer includes a symptom branch, an age branch and a medical history branch, which are respectively used to extract the features of the symptom input vector, the age data and the medical history input vector to obtain the symptom feature vector, the age feature vector and the medical history feature vector; the feature fusion layer is used to splice and fuse the symptom feature vector, the age feature vector and the medical history feature vector to obtain the fusion feature; the hidden layer is used to further process the fusion feature and output the optimized final fusion feature; the first output layer is used to apply the softmax activation function to output the probability prediction of each disease type, and the disease category with the highest probability is the disease triage result; The specific operations for training the disease triage model are as follows: Obtain several disease triage training samples with labeled disease triage results, each disease triage training sample comprising a patient's symptom input vector, a triage weight value corresponding to the symptom input vector, a medical history input vector and normalized age data; divide all disease triage training samples into a disease triage training set and a disease triage verification set, use the disease triage training set to train the disease triage model, and verify the disease triage model through the disease triage verification set to obtain a first verification result; set a first training condition, determine whether the first verification result obtained meets the first training condition, and if so, output the trained disease triage model; if not, continue to use the disease triage training set to train the disease triage model.

[0022] The risk assessment model is established based on the CNN model, including the second input layer, convolution layer, pooling layer, fully connected layer and the second output layer; the second input layer is used to receive the patient's vital signs data; the convolution layer is used to extract the local features of the patient's vital signs data; the pooling layer is used to perform maximum pooling on the output of the convolution layer to reduce the feature dimension; the fully connected layer is used to flatten the output of the pooling layer into a one-dimensional vector, perform linear transformation and nonlinear activation, integrate features and generate high-level representations; the second output layer is used to map the output of the fully connected layer to the probability of four risk levels, use the Softmax function to normalize the output to a probability distribution, obtain the probability of each risk level, and output the risk assessment result, which is the probability of each risk level; The specific operations for training the risk assessment model are as follows: Obtain risk assessment training samples with labeled risk assessment results, each risk assessment training sample containing vital signs data of a patient; divide all acquired risk assessment training samples into a risk assessment training set and a risk assessment verification set; train the risk assessment model using the risk assessment training set, and verify the risk assessment model using the risk assessment verification set to obtain a second verification result; set a second training condition to determine whether the acquired second verification result satisfies the second training condition, and if so, output the trained risk assessment model; if not, continue to train the risk assessment model using the risk assessment training set.

[0023] Use the swarm optimization algorithm to select and obtain the sliding time window Data weight of the day , the specific operations are as follows: Step 1: Randomly generate a particle swarm, the position of each particle represents a candidate weight combination, and set the maximum number of iterations; Step 2: For any particle, initialize the velocity of the particle and use the current position as the individual optimal position of the particle; calculate the inverse of the prediction error when applying the corresponding weight combination of the particle, and use the inverse of the prediction error obtained by calculation as the fitness value of the particle; traverse all particles and use the position corresponding to the particle with the highest fitness value as the global optimal position; Step 3: For any particle, update the speed and position of the particle, and recalculate the fitness value; if the updated fitness value is higher than the fitness value of the individual best position of the particle, update the individual best position of the particle to the current position; Step 4: Traverse all particles. If the fitness value of the individual best position of any particle is higher than the fitness value of the global best position, update the global best position to the individual best position of the particle. Step 5: Repeat steps 3-4 until the maximum number of iterations is reached. The global best position obtained is the best weight combination. .

[0024] Example 2, an intelligent clinical decision support system for emergency triage, such as Figure 1 As shown, including: The medical resource adjustment module includes a trend disease identification unit, a patient quantity prediction unit and a medical resource adjustment unit; the trend disease identification unit is used to determine the trend of any disease type based on the recent trend of the disease type. The number of patients per day in a day is used to determine whether the disease type has entered a trending stage; the patient quantity prediction unit is used to predict the number of patients of any disease type in a trending stage in the next day, and obtain the predicted value of the number of patients of the disease type in the next day; the medical resource adjustment unit is used to adjust the medical resources of the department corresponding to a certain disease type based on the predicted value of the number of patients of the disease type; The patient risk assessment module is used to obtain the patient's vital signs data and use the vital signs data as the input of the risk assessment model to output the probability of each risk level of the patient. The risk levels include level 1 risk, level 2 risk, level 3 risk and level 4 risk. The risk level with the highest probability is the patient's risk level. The patient symptom triage module is used to calculate the triage weight values ​​of all symptoms existing in the historical case data set; when the patient's risk level is level one, emergency measures are taken immediately; when the patient's risk level is level two, three or four, the patient's symptom data, age data and medical history data are obtained; multi-hot encoding is applied to the symptom data and medical history data to generate symptom input vectors and medical history input vectors respectively, and the age data is normalized at the same time; the symptom input vector, the triage weight value corresponding to the symptom input vector, the medical history input vector and the normalized age data are used as the input of the symptom triage model, the symptom triage results are output, and the corresponding treatment department is recommended for the patient based on the obtained symptom triage results.

[0025] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. An intelligent clinical decision support method for emergency triage, characterized in that: include: In recent For any type of disease, based on the disease type in the most recent The number of patients per day in a day is used to determine whether the disease type has entered a trending stage; for any disease type in a trending stage, the number of patients of the disease type in the next day is predicted, and the predicted value of the number of patients of the disease type in the next day is obtained. Based on the predicted value of the number of patients, the medical resources of the corresponding department of the disease type are adjusted; When a new patient is admitted to the hospital emergency department, the patient's vital signs data is obtained, including the patient's heart rate, blood pressure, respiratory rate, blood oxygen saturation and consciousness level; the vital signs data is used as the input of the risk assessment model, and the probability of each risk level of the patient is output. The risk levels include level 1 risk, level 2 risk, level 3 risk and level 4 risk. The risk level with the highest probability is the patient's risk level; The risk level of patients in the hospital emergency department is used to determine the order in which each patient is seen; Obtain a historical case data set, and calculate triage weight values ​​of all symptoms existing in the historical case data set; If the patient's risk level is level one, emergency measures are taken immediately; if the patient's risk level is level two, three or four, the patient's symptom data, age data and medical history data are obtained, and the symptom data includes several symptoms that the patient has; multi-hot encoding is applied to the symptom data and medical history data to generate symptom input vectors and medical history input vectors respectively, and the age data is normalized at the same time; the symptom input vector, the triage weight value corresponding to the symptom input vector, the medical history input vector and the normalized age data are used as inputs to the disease triage model, and the disease triage result is output. The disease triage result is the patient's disease type, and the corresponding treatment department is recommended to the patient based on the patient's disease type.

2. The intelligent clinical decision support method for emergency triage according to claim 1, characterized in that: For any disease type, based on the disease type in the recent The number of patients per day determines whether the disease type has entered the trend stage. The specific operations are as follows: Record the number of patients admitted to the hospital emergency department and the type of illness each patient has every day. days as the current sliding time window; for any disease type, calculate the current disease type within the sliding time window. -The average number of patients per day in the past 3 days, and use the calculated result as the benchmark value b of the current sliding time window; for any day in the last 3 days of the sliding time window, use the formula Calculate the first Daily variation ;in, Indicates that the current disease type is the first in the sliding time window. The number of patients per day, =1, 2, …, ; If the calculated change amplitudes of the last three days are all greater than the preset increase threshold, the current symptom type enters the trend stage; For any symptom type in the trend stage, if the change amplitude of the symptom type in the last 3 days within the sliding time window is lower than the preset decrease threshold, the trend stage of the symptom type ends.

3. The intelligent clinical decision support method for emergency triage according to claim 2, characterized in that: For any disease type in the trending stage, predict the number of patients with this disease type in the next day, and obtain the predicted value of the number of patients with this disease type in the next day. The specific operations are as follows: Using the formula Calculate the predicted value of the number of patients 1 day after acquisition ;in, Indicates the first The data weight of the day, Select and obtain through the group optimization algorithm, .

4. The intelligent clinical decision support method for emergency triage according to claim 3, characterized in that: Obtain a historical case data set, and calculate the triage weight values ​​of all symptoms in the historical case data set. The specific operations are as follows: Obtain a historical case data set, and for any symptom that has appeared in the historical case data set, extract the department where the current symptom exists. Based on the total number of times the current symptom has appeared and the number of times the current symptom has appeared in each department, calculate the proportion of the current symptom in each department. ; =1, 2, …, ; Indicates the number of departments with current symptoms, ; Using the formula Calculate and obtain the concentration C of the current symptoms. The higher the concentration C, the more the current symptoms are concentrated in a few departments. Based on the concentration of all symptoms obtained, select the maximum concentration and minimum concentration ; Then use the formula Calculate and obtain the triage weight value W of the current symptom.

5. The intelligent clinical decision support method for emergency triage according to claim 4, characterized in that: The disease triage model is established based on the MLP model, including the first input layer, symptom weighting layer, feature extraction layer, feature fusion layer, hidden layer and the first output layer; the first input layer is used to input the symptom input vector, the triage weight value corresponding to the symptom input vector, the medical history input vector and the normalized age data; The symptom weighting layer is used to apply the triage weight value to perform weighted processing on the input symptom input vector; the feature extraction layer includes a symptom branch, an age branch, and a medical history branch, which are respectively used to extract the features of the symptom input vector, the age data, and the medical history input vector, and obtain the symptom feature vector, the age feature vector, and the medical history feature vector; the feature fusion layer is used to splice and fuse the symptom feature vector, the age feature vector, and the medical history feature vector to obtain the fusion feature; the hidden layer is used to further process the fusion feature and output the optimized final fusion feature; the first output layer is used to apply the softmax activation function to output the probability prediction of each disease type, and the disease category with the highest probability is the disease triage result; The specific operations for training the disease triage model are as follows: Obtain a number of symptom triage training samples with labeled symptom triage results, each symptom triage training sample comprising a patient's symptom input vector, a triage weight value corresponding to the symptom input vector, a medical history input vector, and normalized age data; divide all symptom triage training samples into a symptom triage training set and a symptom triage verification set, use the symptom triage training set to train the symptom triage model, and use the symptom triage verification set to verify the symptom triage model, and obtain a first verification result; A first training condition is set to determine whether the first verification result obtained satisfies the first training condition. If so, the trained disease triage model is output; if not, the disease triage training set is continuously applied to train the disease triage model.

6. The intelligent clinical decision support method for emergency triage according to claim 5, characterized in that: The risk assessment model is established based on the CNN model, including the second input layer, convolution layer, pooling layer, fully connected layer and the second output layer; the second input layer is used to receive the patient's vital signs data; the convolution layer is used to extract the local features of the patient's vital signs data; the pooling layer is used to perform maximum pooling on the output of the convolution layer to reduce the feature dimension; The fully connected layer is used to flatten the output of the pooling layer into a one-dimensional vector, perform linear transformation and nonlinear activation, integrate features and generate high-level representations; the second output layer is used to map the output of the fully connected layer into the probability of four risk levels, use the Softmax function to normalize the output into a probability distribution, obtain the probability of each risk level, and output the risk assessment result, which is the probability of each risk level; The specific operations for training the risk assessment model are as follows: Obtain risk assessment training samples with labeled risk assessment results, each of which contains vital sign data of a patient; divide all acquired risk assessment training samples into a risk assessment training set and a risk assessment verification set; The risk assessment model is trained using the risk assessment training set, and the risk assessment model is verified using the risk assessment verification set to obtain a second verification result; A second training condition is set to determine whether the second verification result obtained satisfies the second training condition. If so, the trained risk assessment model is output; if not, the risk assessment training set is continued to be applied to train the risk assessment model.

7. The intelligent clinical decision support method for emergency triage according to claim 6, characterized in that: Use the swarm optimization algorithm to select and obtain the sliding time window Data weight of the day , the specific operations are as follows: Step 1: Randomly generate a particle swarm, the position of each particle represents a candidate weight combination, and set the maximum number of iterations; Step 2: For any particle, initialize the velocity of the particle and use the current position as the individual optimal position of the particle; calculate the inverse of the prediction error when applying the corresponding weight combination of the particle, and use the inverse of the prediction error obtained by calculation as the fitness value of the particle; traverse all particles and use the position corresponding to the particle with the highest fitness value as the global optimal position; Step 3: For any particle, update the speed and position of the particle, and recalculate the fitness value; if the updated fitness value is higher than the fitness value of the individual best position of the particle, update the individual best position of the particle to the current position; Step 4: Traverse all particles. If the fitness value of the individual best position of any particle is higher than the fitness value of the global best position, update the global best position to the individual best position of the particle. Step 5: Repeat steps 3-4 until the maximum number of iterations is reached. The global best position obtained is the best weight combination. .

8. An intelligent clinical decision support system for emergency triage, characterized in that: The system is applied to the intelligent clinical decision support method for emergency triage as described in any one of claims 1 to 7, comprising: The medical resource adjustment module includes a trend disease identification unit, a patient quantity prediction unit and a medical resource adjustment unit; the trend disease identification unit is used to determine the trend of any disease type based on the recent trend of the disease type. The number of patients per day in a day is used to determine whether the disease type has entered a trending stage; the patient quantity prediction unit is used to predict the number of patients of any disease type in a trending stage in the next day, and obtain the predicted value of the number of patients of the disease type in the next day; the medical resource adjustment unit is used to adjust the medical resources of the department corresponding to a certain disease type based on the predicted value of the number of patients of the disease type; The patient risk assessment module is used to obtain the patient's vital signs data and use the vital signs data as the input of the risk assessment model to output the probability of each risk level of the patient. The risk levels include level 1 risk, level 2 risk, level 3 risk and level 4 risk. The risk level with the highest probability is the patient's risk level. The patient symptom triage module is used to calculate the triage weight values ​​of all symptoms existing in the historical case data set; when the patient's risk level is level one, emergency measures are taken immediately; when the patient's risk level is level two, three or four, the patient's symptom data, age data and medical history data are obtained; multi-hot encoding is applied to the symptom data and medical history data to generate symptom input vectors and medical history input vectors respectively, and the age data is normalized at the same time; the symptom input vector, the triage weight value corresponding to the symptom input vector, the medical history input vector and the normalized age data are used as the input of the symptom triage model, the symptom triage results are output, and the corresponding treatment department is recommended for the patient based on the obtained symptom triage results.

Images