Emergency department patient information processing method and system based on big data

Abstract

The invention relates to the technical field of data processing, and discloses an emergency department patient information processing method and system based on big data, and the method comprises the steps: carrying out the standardization processing of an unstructured text and heterogeneous data through natural language processing and medical ontology mapping; the method comprises the following steps: dynamically monitoring vital sign data of a patient, extracting trend change and fluctuation amplitude, and mining a frequent pattern of illness state evolution from historical medical record data based on an FP-Growth algorithm; through a Transform-GNN-XGBoost model, the disease deterioration risk of the patient is predicted, and in combination with historical medical record data and the current state of the patient, the hospitalization duration and the treatment cost of the patient are predicted; dynamically generating an optimal triage strategy and a resource scheduling scheme through a genetic algorithm; the prediction result, the optimal triage strategy and the resource scheduling scheme are visually displayed to medical staff; the diagnosis and treatment efficiency and the resource utilization rate of the emergency department are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to an emergency department patient information processing method and system based on big data. BACKGROUND

[0002] In the daily operation of the emergency department, a large amount of patient information will be generated, which contains patient basic information, symptom description, diagnosis result, treatment process and other aspects of content; the traditional emergency department patient information processing method can only record and store data simply, and it is difficult to deeply mine and effectively use these complex and massive data. SUMMARY

[0003] The present application aims to solve the above problems, and designs an emergency department patient information processing method and system based on big data.

[0004] The first aspect of the present application provides an emergency department patient information processing method based on big data, which comprises the following steps:

[0005] Collecting patient information, real-time vital sign data, historical medical record data and emergency process data of the emergency department, and performing standardized processing on unstructured text and heterogeneous data through natural language processing and medical ontology mapping;

[0006] Using SparkStreaming for real-time stream processing, dynamically monitoring patient vital sign data, performing sliding window analysis on vital sign data, extracting trend changes and fluctuation amplitudes, and based on the FP-Growth algorithm, mining frequent patterns of disease evolution from historical medical record data;

[0007] Through the Transformer-GNN-XGBoost model, the risk of patient condition deterioration is predicted, and the length of hospital stay and treatment cost of the patient are predicted based on the historical medical record data and the current state of the patient;

[0008] Based on the prediction results output by the Transformer-GNN-XGBoost model, the optimal triage strategy and resource scheduling scheme are dynamically generated through the genetic algorithm;

[0009] The prediction results, the optimal triage strategy and the resource scheduling scheme are visualized and displayed to medical staff.

[0010] Optionally, in the first implementation manner of the first aspect of the present application, the collecting of patient information, real-time vital sign data, historical medical record data and emergency process data of the emergency department, and the standardized processing of unstructured text and heterogeneous data through natural language processing and medical ontology mapping comprises:

[0011] The unstructured text is split into sub-word units by using a BERT word segmentation model, and the boundaries of each sub-word are determined based on the context semantics to obtain a word segmentation result;

[0012] Based on the word segmentation result, a BioBERT medical entity recognition tool is used to extract key entities, and the recognized entities are matched with medical terms to confirm the standard name of the entity;

[0013] The text after recognizing the entity is standardized, and the heterogeneous data from different sources is mapped to the standard term. The KNN interpolation method is used to fill the missing data in the standardization process to obtain the standardized data.

[0014] Optionally, in the second implementation manner of the first aspect of the present application, the real-time stream processing using SparkStreaming is used to dynamically monitor the patient vital sign data, and the sliding window analysis is performed on the vital sign data to extract the trend change and fluctuation amplitude, including:

[0015] The real-time vital sign data is received through the Kafka message queue, the window size is set to 10 minutes, the sliding step is set to 5 minutes, the mean, standard deviation and trend slope of the vital sign in the window are calculated, and the statistical characteristics are obtained;

[0016] For each vital sign index, the statistical characteristics in the window are compared with the preset threshold, and for the index exceeding the threshold by 15%, an abnormal label is added, and the time window of the abnormality is recorded.

[0017] Optionally, in the third implementation manner of the first aspect of the present application, the frequent patterns of disease evolution are mined from the historical medical record data based on the FP-Growth algorithm, including:

[0018] Triplet sequences including symptoms, diagnoses and treatments are extracted from the historical medical record data, and are sorted according to patient ID and visit time to construct the disease evolution sequence of each patient;

[0019] The disease evolution sequence of each patient is split into individual items to form a transaction database, the transaction database is scanned, the support count of each item is counted, and items with a support less than 5% are filtered out;

[0020] An FP tree is constructed, the frequent items are arranged in descending order of support, the conditional pattern base is recursively generated from the leaf node of the FP tree, the conditional FP tree is constructed, and the frequent patterns meeting the minimum confidence of 70% are mined.

[0021] Optionally, in the fourth implementation form of the first aspect of the present application, the prediction of the patient's condition deterioration risk by the Transformer-GNN-XGBoost model, in combination with the historical medical record data and the current state of the patient, predicts the length of hospital stay and the treatment cost of the patient, comprising:

[0022] The real-time vital sign data of the patient is divided by time window to form a multi-dimensional time sequence, and is input into the Transformer-GNN-XGBoost model;

[0023] The Transformer encoder calculates the correlation weight between each time step in the multi-dimensional time sequence through the self-attention mechanism, and outputs a time sequence feature vector after multi-layer encoding, capturing the dynamic change pattern and potential correlation of the vital signs;

[0024] A heterogeneous graph containing four types of nodes of patients, diseases, symptoms and drugs is constructed, the GNN network is used for feature aggregation of the nodes, the weighted aggregated features are calculated by sampling neighbor nodes, and a graph embedding representation of the patient is generated;

[0025] The time sequence feature vector output by the Transformer encoder and the graph embedding representation generated by the GNN network are spliced to form a comprehensive feature vector;

[0026] The comprehensive feature vector is input into the XGBoost classifier to predict the patient's condition deterioration risk, and in combination with the historical medical record data and the current state of the patient, the length of hospital stay and the treatment cost of the patient are predicted.

[0027] Optionally, in the fifth implementation form of the first aspect of the present application, the optimal triage strategy and resource scheduling scheme are dynamically generated based on the prediction results output by the Transformer-GNN-XGBoost model through a genetic algorithm, comprising:

[0028] According to the number of patients to be triaged in the current emergency department and the available resources, 50 candidate triage schemes are randomly generated as an initial population;

[0029] Each candidate triage scheme is encoded as a chromosome, containing N genes, each gene representing the allocation information of a patient;

[0030] The fitness score of the chromosome is calculated by the fitness function, 5 schemes are randomly selected from the initial population, the fitness scores are compared, and the two schemes with the highest scores are selected as parents, and the selection operation is repeated to generate 25 pairs of parents;

[0031] Two crossover points on the chromosome are randomly selected, and the gene segments between the crossover points are exchanged between the parents to generate two offspring schemes after crossing;

[0032] The offspring scheme is mutated with a probability of 10%, and the five schemes with the highest fitness scores in the current population are directly entered into the next generation; the offspring schemes generated by crossing and mutation are combined with the elite individuals to form a new population, and the iterative evolution is performed for 100 generations;

[0033] After the iteration is completed, the scheme with the highest fitness score is selected as the optimal triage strategy and resource scheduling scheme.

[0034] Optionally, in the sixth implementation manner of the first aspect of the present application, the fitness function includes three dimensions of triage accuracy, resource utilization rate and patient waiting time, and the fitness score of each scheme is obtained by weighted summation of the scores of the three dimensions.

[0035] The second aspect of the present application provides an emergency department patient information processing system based on big data, which comprises:

[0036] A standardization processing module is configured to collect patient information, real-time vital sign data, historical medical record data and emergency process data of the emergency department, and perform standardization processing on unstructured text and heterogeneous data through natural language processing and medical ontology mapping.

[0037] An extraction module is configured to use SparkStreaming for real-time stream processing, dynamically monitor patient vital sign data, perform sliding window analysis on the vital sign data, extract trend changes and fluctuation amplitudes, and mine frequent patterns of disease evolution from historical medical record data based on an FP-Growth algorithm.

[0038] A prediction module is configured to predict patient condition deterioration risk through a Transformer-GNN-XGBoost model, and predict patient hospitalization length and treatment cost in combination with historical medical record data and current patient state.

[0039] A generation module is configured to dynamically generate an optimal triage strategy and resource scheduling scheme based on the prediction results output by the Transformer-GNN-XGBoost model through a genetic algorithm.

[0040] A visualization module is configured to visually display the prediction results, the optimal triage strategy and the resource scheduling scheme to medical staff.

[0041] The third aspect of the present application provides an emergency department patient information processing device based on big data, which comprises a memory and at least one processor, and the memory stores instructions; the at least one processor invokes the instructions in the memory to enable the emergency department patient information processing device based on big data to perform each step of the emergency department patient information processing method based on big data.

[0042] The fourth aspect of the present application provides a computer readable storage medium, which stores instructions, and the instructions are executed by a processor to implement each step of the emergency department patient information processing method based on big data according to any one of the above.

[0043] In the technical solution provided by the present application, the patient information, real-time vital sign data, historical medical record data and emergency process data of the emergency department are collected, and the unstructured text and heterogeneous data are standardized by natural language processing and medical ontology mapping; real-time stream processing is performed using SparkStreaming, the patient vital sign data is dynamically monitored, the sliding window analysis is performed on the vital sign data, the trend change and fluctuation amplitude are extracted, the frequent patterns of disease evolution are mined from the historical medical record data based on the FP-Growth algorithm; the patient condition deterioration risk is predicted by the Transformer-GNN-XGBoost model, and the patient hospitalization length and treatment cost are predicted in combination with the historical medical record data and the current state of the patient; the optimal triage strategy and resource scheduling scheme are dynamically generated by the genetic algorithm based on the prediction results output by the Transformer-GNN-XGBoost model; the prediction results, the optimal triage strategy and the resource scheduling scheme are visually displayed to medical staff; the present application realizes full-dimensional data standardization and integration, eliminates data silos, shortens the early warning time, reduces experience bias, improves resource utilization, speeds up emergency response, and significantly improves the efficiency of diagnosis and treatment and the resource utilization rate of the emergency department. BRIEF DESCRIPTION OF DRAWINGS

[0044] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included only to illustrate preferred embodiments and are not to be considered as limiting of the present application.

[0045] Figure 1 The flowchart of the emergency department patient information processing method based on big data provided by the embodiments of the present application;

[0046] Figure 2 The structural schematic diagram of the emergency department patient information processing system based on big data provided by the embodiments of the present application;

[0047] Figure 3 The structural schematic diagram of the emergency department patient information processing device based on big data provided by the embodiments of the present application. DETAILED DESCRIPTION

[0048] The terms "first", "second", "third", "fourth" and the like in the description and in the claims of the present application, and above-mentioned drawings, if any, are used to distinguish between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the use of the terms so used herein is interchangeable under appropriate circumstances such that the embodiments of the application described herein are, for example, capable of efficient implementation in other than the order illustrated and / or described herein. Moreover, the terms "comprising" or "having" and any variations thereof, are intended to cover a non-exclusive inclusion, for example, a process, method, article, or apparatus that comprises a list of steps or units not solely those specified, but can include other not specified steps or units inherent therein.

[0049] For the sake of understanding, the specific flow of the embodiments of the application is described below, please refer to Figure 1 The flowchart of the emergency department patient information processing method based on big data provided by the embodiments of the application, the method specifically comprises the following steps:

[0050] Step 101, collecting patient information, real-time vital sign data, historical medical record data and emergency process data of the emergency department, and standardizing unstructured text and heterogeneous data through natural language processing and medical ontology mapping;

[0051] In this embodiment, the patient basic information includes structured data such as name, gender, age, ID number, contact information, medical history, drug allergy history, medical insurance type, etc.

[0052] Real-time vital sign data: collected through multi-parameter monitors, electrocardiogram machines and other devices, including heart rate, blood pressure, blood oxygen saturation, body temperature, respiratory rate, etc., with a sampling frequency of 1 time / second.

[0053] Historical medical record data: inpatient records in the past 3 years, outpatient diagnosis results such as "acute myocardial infarction" and "stroke", examination reports such as blood routine and CT image conclusion, treatment plans such as drug name and dose, and operation type, etc.

[0054] Emergency process data: reception time, emergency measures such as cardiopulmonary resuscitation, tracheal intubation, medical staff operation records, transportation path, etc. unstructured text.

[0055] In this embodiment, when processing unstructured text using the BERT segmentation model, first convert the handwritten medical records, examination report descriptions and other content in the emergency process into character sequences that the model can recognize. Through the pre-trained word embedding layer, each character is mapped to a high-dimensional vector. The model will combine the context semantics to split the continuous character sequence, for example, when encountering acute lymphoblastic leukemia, it will be split into sub-word units such as acute / lymphoblastic / leukemia according to the common combination of medical terms, ensuring that each sub-word retains independent semantics and reflects the association with the previous and subsequent sub-words. Finally, generate a segmentation result containing sub-word position information and semantic association;

[0056] Based on the segmentation results, use the BioBERT medical entity recognition tool. The tool will perform sequence labeling on each sub-word, analyze the syntactic role and semantic features of the sub-word in the context through the fine-tuned Transformer architecture, and identify key entities such as "myocardial infarction" (disease), "aspirin" (drug), and "fever" (symptom). Then, compare these identified entities with the pre-constructed medical terminology library, for example, match heart infarction to the standard term myocardial infarction, and match fever-reducing drugs to specific drug names such as acetaminophen, to confirm the standard name of each entity and ensure the consistency of entity representation.

[0057] When processing the text after identifying the entity, the text format is unified through regular expressions, such as converting blood pressure 120 / 80 mmHg to systolic pressure 120 mmHg and diastolic pressure 80 mmHg, and adjusting the visit time 2023-7-5 to 2023-07-05. For heterogeneous data from different sources, based on the SNOMEDCT medical ontology library and ICD-10 disease coding, different terms such as "HR" and "heart rate" are mapped to heart rate (beats / min), and "no obvious abnormalities" in the image report are mapped to normal = 1. In this process, if there is missing data, use the KNN interpolation method to select the 5 most similar samples from the historical data of the same type of patients, calculate the weighted average value based on the feature similarity between the samples and the target data, and fill in the missing value with the average value. Finally, get the standardized data with unified format and complete content.

[0058] Step 102, use SparkStreaming for real-time stream processing, dynamically monitor patient vital sign data, perform sliding window analysis on vital sign data, extract trend changes and fluctuation amplitudes, and based on the FP-Growth algorithm, mine frequent patterns of disease evolution from historical medical record data.

[0059] In this example, a topic named "emergency_vital_signs" is created through Kafka with 8 partitions based on the number of emergency rescue units and 3 replicas for each partition to ensure high availability. Each medical device such as a monitor or ventilator is assigned an independent producer client that sends the collected real-time vital sign data such as timestamps, patient IDs, heart rates, blood pressures, etc., serialized in JSON format to the corresponding partition with a 7-day message retention policy to support historical data analysis;

[0060] The Spark Streaming context is initialized with a batch interval of 5 seconds to ensure that the system can process all incoming data in a batch within that time. The checkpoint directory is configured to HDFS, and Exactly-Once semantics are enabled to ensure data processing accuracy. The Kafka Direct API is used to subscribe to the "emergency_vital_signs" topic, and a DStream object is created to deserialize the received JSON data into structured patient vital sign records;

[0061] A sliding window operation is applied to the DStream with a window size of 10 minutes and a sliding step of 5 minutes. The data within the window is grouped by patient ID to ensure that each patient's vital sign data forms an independent data stream. For each group, all vital sign indicators such as heart rate and blood oxygen saturation are extracted, sorted by timestamp, and a time series data structure is constructed;

[0062] For each patient's windowed time series data, key statistical features are calculated:

[0063] Basic statistics: The mean, median, and standard deviation of indicators such as heart rate and blood pressure are calculated to reflect the central tendency and dispersion of the data;

[0064] Trend slope calculation: Linear regression is used to fit the time series data to calculate the slope of the indicator's change per unit time, such as the number of heart rate changes per minute, to identify upward or downward trends;

[0065] Extreme value detection: The maximum and minimum values within the window are marked, along with the time of occurrence and the degree of deviation from the normal value range;

[0066] For each vital sign indicator, the windowed statistical features are compared to pre-set medical thresholds:

[0067] Normal range judgment: Whether the heart rate is within the range of 60-100 beats per minute, and whether the blood oxygen saturation is ≥95%;

[0068] Wave amplitude calculation: compare the current window with the previous window of the index rate of change such as systolic blood pressure change more than 15%;

[0069] Abnormality marking: add an abnormal label to the index that exceeds the normal range or the wave amplitude exceeds the threshold, and record the time window when the abnormality occurs;

[0070] Read the emergency medical history data of the past 5 years from HDFS, extract the triple sequence of symptoms, diagnosis and treatment. For each case, convert the symptoms, diagnosis and treatment measures into standardized codes. Sort the cases by patient ID and visit time, and construct the patient's disease evolution sequence to provide input for frequent pattern mining.

[0071] Split each patient's disease evolution sequence into individual items to form a transaction database, scan the transaction database, and count the support count of each item. Filter out items with a support count below 5%;

[0072] Build an FP tree, arrange the frequent items in descending order of support, start from the leaf node of the FP tree, recursively generate a conditional pattern base, build a conditional FP tree, and mine frequent patterns that meet the minimum confidence of 70%.

[0073] Step 103, predict the patient's disease deterioration risk through the Transformer-GNN-XGBoost model, and combine the historical medical record data and the patient's current state to predict the patient's hospitalization length and treatment cost;

[0074] In this embodiment, the patient's real-time vital sign data is divided into time windows to form a multi-dimensional time sequence, and input into the Transformer-GNN-XGBoost model; the Transformer encoder calculates the correlation weight between each time step in the multi-dimensional time sequence through self-attention mechanism, and outputs a time sequence feature vector after multiple layers of encoding, capturing the dynamic change pattern and potential correlation of vital signs; build a heterogeneous graph containing four types of nodes: patients, diseases, symptoms and drugs, aggregate features of nodes through GNN network, calculate weighted aggregated features by sampling neighbor nodes, and generate graph embedding representation of patients; concatenate the time sequence feature vector output by the Transformer encoder and the graph embedding representation generated by the GNN network to form a comprehensive feature vector;

[0075] Input the fused feature vector into the XGBoost classifier, and the training target is to predict whether the patient needs to be transferred to the ICU within the next 24 hours; XGBoost trains multiple decision trees through iteration, each tree fits the residual of the previous round, and uses gradient boosting mechanism to optimize the model; During training, use 5-fold cross-validation to evaluate model performance, and optimize F1 score by adjusting learning rate, tree depth and other hyperparameters; Finally output the probability value of patient disease deterioration;

[0076] The XGBoost regression model is constructed to predict the length of hospital stay; the actual length of hospital stay of historical cases is taken as the target value, and the mean square error is taken as the loss function; during the training process, the long-tail distribution of the length of hospital stay data is logarithmically transformed to alleviate the influence of outliers; the optimal hyperparameter combination such as the number of trees and the minimum sample number of leaf nodes is determined through grid search; the model outputs the predicted length of hospital stay, and estimates the uncertainty of the prediction through the confidence interval;

[0077] For treatment cost prediction, a quantile regression forest model is constructed to predict the median, 25% quantile and 75% quantile of the cost, forming a cost prediction interval; the total treatment cost of historical cases is taken as the target value, and medical resource usage such as ICU days and surgery type is taken as the feature; by minimizing the quantile loss function, three independent models are trained to estimate different quantile points; finally, the cost prediction interval with 90% confidence is output;

[0078] The outputs of the three XGBoost models (deterioration risk, length of hospital stay, and treatment cost) are integrated to generate a unified prediction report; the deterioration risk probability is mapped to a risk level, the length of hospital stay prediction value is converted to the format of "expected days ± error range", and the treatment cost interval is converted to a visual column chart form; at the same time, the confidence index of each prediction result is calculated to evaluate the prediction reliability; finally, the report is stored in the database;

[0079] Actual outcomes of new inpatients are collected daily, such as whether to deteriorate or the actual length of hospital stay, forming an incremental training set; online learning strategy is used to fine-tune the parameters of the XGBoost model, keeping the model adaptive to the latest data distribution; the entire Transformer-GNN-XGBoost model is retrained regularly to include the latest cases and medical knowledge graph updates; by comparing the prediction results with the actual outcomes, the model is calibrated, and the prediction threshold is adjusted to balance the false positive and false negative rates.

[0080] Step 104, based on the prediction results output by the Transformer-GNN-XGBoost model, the optimal triage strategy and resource scheduling scheme are dynamically generated through genetic algorithm;

[0081] In this embodiment, 50 candidate triage schemes are randomly generated as the initial population according to the number of patients waiting for triage and available resources in the current emergency department; each candidate triage scheme is encoded as a chromosome, containing N genes, each gene representing the allocation information of a patient; the fitness score of the chromosome is calculated by the fitness function, 5 schemes are randomly selected from the initial population, the fitness scores are compared, the two schemes with the highest scores are selected as parents, the selection operation is repeated to generate 25 pairs of parents; two crossover points on the chromosome are randomly selected, the gene segments between the crossover points are exchanged between the parents, and two offspring schemes are generated after crossing; the offspring schemes are mutated with a probability of 10%, a gene on the chromosome is randomly selected and its value is replaced with another legal value. For example, change the doctor assigned to a patient from A to B, or adjust the equipment usage time, reevaluate the feasibility of the offspring after mutation; select the top 5 schemes with the highest fitness scores from the current population to directly enter the next generation, combine the offspring schemes generated by crossing and mutation with the elite individuals to form a new population, and iterate evolution for 100 generations; after iteration, the scheme with the highest fitness score is selected as the optimal triage strategy and resource scheduling scheme.

[0082] In this embodiment, the fitness function includes three dimensions of triage accuracy, resource utilization rate and patient waiting time, and the fitness score of each scheme is obtained by weighted sum of the scores of the three dimensions;

[0083] Triage accuracy: based on the risk level of patient's condition (I-IV level), calculate the matching degree of the allocated resource level and risk level, such as I level patient must be allocated the highest priority resource;

[0084] Resource utilization rate: calculate the idle time proportion of various resources such as beds, equipment and medical staff, the goal is to control the idle rate within 10%;

[0085] Patient waiting time: calculate the average waiting time of each patient from admission to treatment, the goal is to control the average waiting time within 30 minutes.

[0086] Step 105, visualize the prediction results, optimal triage strategy and resource scheduling scheme to medical staff.

[0087] In this embodiment, by constructing a multi-role adaptive visual interaction platform, the prediction results such as patient condition deterioration risk, hospitalization length prediction, treatment cost interval, etc. are presented, for example, the patient condition deterioration risk is presented in a red-yellow-green three-color heat map, red represents high risk and real-time flashing warning, the hospitalization length prediction is displayed in a blue column chart with error range, and the treatment cost interval is presented in an orange line chart with upper and lower limits. The triage strategy is stratified according to the condition level, and the patients of grade I-IV are respectively corresponding to identification cards of different colors, with priority allocation instructions. The resource scheduling scheme includes a 24-hour occupation Gantt chart of devices such as beds and ventilators, and a dynamic update of the time axis form of the nurse scheduling table. The doctor interface highlights the real-time vital curve and risk warning details of the patients under his charge, supports clicking to review the treatment plan of similar cases, and the head nurse interface focuses on the resource utilization efficiency board, which can manually fine-tune the scheduling scheme through drag operation and real-time view the fitness score change after adjustment. All data are automatically refreshed every 5 seconds to ensure that medical staff can intuitively and efficiently obtain key information and quickly respond.

[0088] Please refer to Figure 2 The structure diagram of the emergency department patient information processing system based on big data provided by the embodiment of the application, the system comprises:

[0089] A standardized processing module is configured to collect patient information, real-time vital sign data, historical medical record data and emergency process data of the emergency department, and perform standardized processing on unstructured text and heterogeneous data through natural language processing and medical ontology mapping.

[0090] An extraction module is configured to use SparkStreaming for real-time stream processing, dynamically monitor patient vital sign data, perform sliding window analysis on vital sign data, extract trend changes and fluctuation amplitudes, and mine frequent patterns of disease evolution from historical medical record data based on an FP-Growth algorithm.

[0091] A prediction module is configured to predict patient condition deterioration risk through a Transformer-GNN-XGBoost model, and predict patient hospitalization length and treatment cost in combination with historical medical record data and current state of the patient.

[0092] A generation module is configured to dynamically generate an optimal triage strategy and resource scheduling scheme based on the prediction results output by the Transformer-GNN-XGBoost model through a genetic algorithm.

[0093] A visualization module is configured to visually display the prediction results, the optimal triage strategy and the resource scheduling scheme to medical staff.

[0094] Figure 3is a structural schematic view of an emergency department patient information processing device based on big data provided by an embodiment of the present application. The emergency department patient information processing device based on big data 300 can have a large difference due to different configurations or performances, and can include one or more processors (central processing units, CPUs) 310 (for example, one or more processors) and a memory 320, one or more storage media 330 (for example, one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and the storage media 330 can be temporary storage or persistent storage. The programs stored in the storage media 330 can include one or more modules (not shown in the figure), and each module can include a series of instruction operations in the emergency department patient information processing device based on big data 300. Further, the processor 310 can be configured to communicate with the storage media 330 and execute a series of instruction operations in the storage media 330 on the emergency department patient information processing device based on big data 300 to implement the method provided by the above embodiment.

[0095] The emergency department patient information processing device based on big data 300 can also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Server, MacOS X, Unix, Linux, FreeBSD, etc. Those skilled in the art can understand that, Figure 3 The structure of the emergency department patient information processing device based on big data shown does not constitute a limitation on the computer device provided by the present application, and can include more or fewer components than shown, or combine certain components, or different component arrangements.

[0096] The present application also provides a computer readable storage medium, which can be a non-volatile computer readable storage medium or a volatile computer readable storage medium. The computer readable storage medium has instructions stored therein, and when the instructions are run on a computer, the computer executes the steps of the emergency department patient information processing method based on big data provided by the above embodiments.

[0097] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described device or apparatus, unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0098] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or say the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0099] The basic principles, main features and advantages of the present application are shown and described above. Those skilled in the art should understand that the present application is not limited by the above embodiments, and the above embodiments and descriptions in the specification are only preferred examples of the present application and are not intended to limit the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for processing emergency department patient information based on big data, characterized in that: The method comprises the following steps: Collect patient information, real-time vital signs data, historical medical records data, and emergency process data from the emergency department, and standardize unstructured text and heterogeneous data through natural language processing and medical ontology mapping; Use Spark Streaming for real-time stream processing to dynamically monitor patient vital signs data. Perform sliding window analysis on vital sign data to extract trend changes and fluctuations. Use the FP-Growth algorithm to mine frequent patterns of disease evolution from historical medical records. The Transformer-GNN-XGBoost model is used to predict the risk of a patient's condition worsening. It also combines historical medical records with the patient's current condition to predict the patient's hospitalization duration and treatment costs. Based on the prediction results output by the Transformer-GNN-XGBoost model, the optimal triage strategy and resource scheduling plan are dynamically generated through genetic algorithms; The prediction results, optimal triage strategy and resource scheduling plan are visually displayed to medical staff.

2. The method for processing emergency department patient information based on big data according to claim 1, characterized in that: The method collects patient information, real-time vital signs data, historical medical records data, and emergency process data from the emergency department, and standardizes unstructured text and heterogeneous data through natural language processing and medical ontology mapping, including: The BERT word segmentation model is used to split unstructured text into subword units, and the boundaries of each subword are determined based on the contextual semantics to obtain the word segmentation results. Based on the word segmentation results, the BioBERT medical entity recognition tool is used to extract key entities, and the identified entities are matched with medical terms to confirm the standard names of the entities; The text after entity recognition is standardized, and terminology standardization mapping is performed on heterogeneous data from different sources. The KNN interpolation method is used to fill in the missing data that occurs during the standardization process to obtain standardized data.

3. The method for processing emergency department patient information based on big data according to claim 1, characterized in that: The method uses Spark Streaming for real-time stream processing, dynamically monitors patient vital signs data, performs sliding window analysis on vital signs data, and extracts trend changes and fluctuation amplitudes, including: Receive real-time vital sign data through the Kafka message queue, set the window size to 10 minutes and the sliding step size to 5 minutes, calculate the mean, standard deviation, and trend slope of the vital signs within the window, and obtain statistical features; For each vital sign indicator, the statistical features within the window are compared with the preset threshold. For indicators that exceed the threshold by 15%, an abnormal label is added, and the time window when the abnormality occurs is recorded.

4. The method for processing emergency department patient information based on big data according to claim 1, characterized in that: The FP-Growth algorithm is used to mine frequent patterns of disease evolution from historical medical records, including: Extract triplet sequences including symptoms, diagnosis, and treatment from historical medical records, sort them by patient ID and consultation time, and construct a sequence of each patient's condition evolution; Split each patient's condition evolution sequence into separate items to form a transaction database. Scan the transaction database, calculate the support count of each item, and filter out items with support less than 5%; Construct an FP tree, sort the frequent items in descending order of support, start from the leaf node of the FP tree, recursively generate the conditional pattern base, construct a conditional FP tree, and mine frequent patterns that meet the minimum confidence of 70%.

5. The method for processing emergency department patient information based on big data according to claim 1, characterized in that: The Transformer-GNN-XGBoost model is used to predict the risk of a patient's condition worsening. It also combines historical medical records with the patient's current condition to predict the patient's hospitalization duration and treatment costs, including: The patient's real-time vital sign data is split into time windows to form a multidimensional time series sequence, which is then input into the Transformer-GNN-XGBoost model; The Transformer encoder calculates the association weights between each time step in a multi-dimensional time series through a self-attention mechanism. After multi-layer encoding, it outputs a time series feature vector that captures the dynamic change patterns and potential correlations of vital signs. Construct a heterogeneous graph containing four types of nodes: patients, diseases, symptoms, and drugs. Use the GNN network to aggregate node features. By sampling neighbor nodes and calculating weighted aggregate features, a graph embedding representation of the patient is generated. The temporal feature vector output by the Transformer encoder and the graph embedding representation generated by the GNN network are concatenated to form a comprehensive feature vector; The comprehensive feature vector is input into the XGBoost classifier to predict the risk of the patient's condition worsening, and the patient's hospitalization time and treatment costs are predicted by combining historical medical record data and the patient's current condition.

6. The method for processing emergency department patient information based on big data according to claim 1, characterized in that: The prediction results output by the Transformer-GNN-XGBoost model are used to dynamically generate the optimal triage strategy and resource scheduling plan through a genetic algorithm, including: According to the current number of patients to be triaged in the emergency department and the available resources, 50 candidate triage plans are randomly generated as the initial population; Each candidate triage scheme is encoded as a chromosome, containing N genes, each gene representing the allocation information of a patient; The fitness score of the chromosome is calculated using the fitness function. Five schemes are randomly selected from the initial population, and the fitness scores are compared. The two schemes with the highest scores are selected as parents. The selection operation is repeated to generate 25 pairs of parents. Randomly select two crossover points on the chromosome, exchange the gene segments between the crossover points between the parents, and generate two offspring schemes after crossover; The offspring solutions are mutated with a probability of 10%, and the five solutions with the highest fitness scores are selected from the current population to directly enter the next generation. The offspring solutions generated by crossover and mutation are merged with the elite individuals to form a new population, and the evolution is iterated for 100 generations. After the iteration, the solution with the highest fitness score is selected as the optimal triage strategy and resource scheduling solution.

7. The method for processing emergency department patient information based on big data according to claim 6, characterized in that: The fitness function includes three dimensions: triage accuracy, resource utilization, and patient waiting time. The scores of the three dimensions are weighted and summed to obtain the fitness score of each solution.

8. An emergency department patient information processing system based on big data, characterized in that: The system includes: The standardization processing module is used to collect patient information, real-time vital signs data, historical medical records data, and emergency process data from the emergency department. It standardizes unstructured text and heterogeneous data through natural language processing and medical ontology mapping. The extraction module uses Spark Streaming for real-time stream processing to dynamically monitor patient vital signs data, perform sliding window analysis on vital signs data, extract trend changes and fluctuation amplitudes, and mine frequent patterns of disease evolution from historical medical records based on the FP-Growth algorithm. The prediction module uses the Transformer-GNN-XGBoost model to predict the risk of a patient's condition worsening and, by combining historical medical records with the patient's current condition, to predict the patient's length of hospital stay and treatment costs. The generation module is used to dynamically generate the optimal triage strategy and resource scheduling plan based on the prediction results output by the Transformer-GNN-XGBoost model through a genetic algorithm; The visualization module is used to visualize the prediction results, optimal triage strategy and resource scheduling plan to medical staff.

9. An emergency department patient information processing device based on big data, characterized in that: The big data-based emergency department patient information processing device includes a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the big data-based emergency department patient information processing device performs each step of the big data-based emergency department patient information processing method as described in any one of claims 1-7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the various steps of the method for processing emergency department patient information based on big data as described in any one of claims 1 to 7 are implemented.

Images