Regional major trauma data information research platform based on cloud computing and construction method

Abstract

The invention provides a regional major trauma data information research platform based on cloud computing and a construction method, and relates to the technical field of cloud computing, and the method comprises the steps: receiving major trauma case data uploaded by a medical institution, and carrying out the classified storage through a distributed storage system; performing trauma feature extraction and multi-modal feature fusion by using a deep learning model, and constructing a trauma severity prediction model; and constructing a medical resource intelligent allocation model based on a reinforcement learning algorithm, and optimizing a transfer scheme in real time. According to the invention, intelligent analysis of regional major trauma cases and efficient allocation of medical resources can be realized, and major trauma treatment efficiency is improved.

Description

Technical Field

[0001] The present invention relates to cloud computing technology, and in particular to a regional major trauma data information research platform based on cloud computing and a construction method thereof. Background Art

[0002] Major trauma is one of the important factors threatening human life and health, which is characterized by acute onset, severe illness, high mortality, etc. With the development of the economic society, the incidence of major trauma shows an upward trend, bringing huge pressure to the medical system. At present, each medical institution has established a corresponding first-aid system for major trauma treatment, and carries out treatment work through means such as pre-hospital first aid and in-hospital treatment. During the actual treatment process, medical institutions need to obtain the patient's condition information in a timely manner, evaluate the severity of the trauma, and rationally allocate medical resources to ensure that patients receive timely and effective treatment.

[0003] First, the information sharing among medical institutions is insufficient. The clinical data related to patients are scattered and stored in different systems, making it difficult to achieve unified management and analysis and utilization of the data, which affects the treatment efficiency and quality.

[0004] Secondly, the existing trauma assessment methods mainly rely on traditional scoring systems and fail to make full use of artificial intelligence technologies such as deep learning to analyze multi-modal data such as medical images, resulting in insufficient accuracy and timeliness of trauma assessment.

[0005] Finally, the allocation of medical resources mainly relies on manual decision-making, lacking an intelligent allocation mechanism and being unable to dynamically optimize according to the real-time condition changes and the distribution of regional medical resources, which affects the treatment effect of major trauma patients. Summary of the Invention

[0006] The embodiments of the present invention provide a regional major trauma data information research platform based on cloud computing and a construction method thereof, which can solve the problems in the prior art.

[0007] In the first aspect of the embodiments of the present invention,

[0008] Receive major trauma case data uploaded by each medical institution in the region, where the major trauma case data includes patient physiological sign data, diagnosis and treatment record data, medical image data, and inspection and examination data;

[0009] Classify and store the major trauma case data through a distributed storage system, and store the major trauma case data in a structured database and an unstructured database respectively according to the data type;

[0010] Extract features from the medical image data based on a deep learning model, and adopt a multi-task learning method to simultaneously complete the identification of the trauma site and the assessment of the trauma degree. Perform multi-modal feature fusion on the trauma site identification result and the trauma degree assessment result with the patient's physiological sign data to generate a trauma severity prediction model;

[0011] Call the major trauma case data from the structured database and the unstructured database, perform parallel processing on the data using a distributed computing framework, and calculate the occurrence trend of regional major trauma and the distribution of regional medical resources based on a preset analysis model;

[0012] According to the prediction result of the trauma severity prediction model, combined with the occurrence trend of regional major trauma and the distribution of regional medical resources, construct a medical resource intelligent allocation model using a reinforcement learning algorithm. The medical resource intelligent allocation model is used to calculate the optimal transfer plan based on the trauma severity, the first aid ability of medical institutions, and the driving time when receiving a new major trauma case;

[0013] Send the optimal transfer plan to the medical information systems of each medical institution in the region, and perform medical resource allocation and case transfer through the medical information systems;

[0014] Real-time collect the patient's physiological sign data during the transfer process, send the patient's physiological sign data back to the cloud computing platform, and dynamically optimize and adjust the optimal transfer plan according to the patient's physiological sign data.

[0015] Extract features from the medical image data based on a deep learning model, and adopt a multi-task learning method to simultaneously complete the identification of the trauma site and the assessment of the trauma degree. Performing multi-modal feature fusion on the trauma site identification result and the trauma degree assessment result with the patient's physiological sign data to generate a trauma severity prediction model includes:

[0016] Based on the deep learning model, adopt a multi-task learning method to construct a shared feature layer and task-specific layers. The shared feature layer is used to extract general features, and the task-specific layers are respectively used for the trauma site identification task and the trauma degree assessment task, and simultaneously output the trauma site identification result and the trauma degree assessment result;

[0017] Adopt an attention mechanism to perform multi-modal feature fusion on the trauma site identification result, the trauma degree assessment result and the patient's physiological sign data, and train the trauma severity prediction model based on the fused features;

[0018] Use the trauma severity prediction model to predict the trauma severity of newly received medical image data and patient physiological sign data.

[0019] Retrieve the major trauma case data from the structured database and the unstructured database, and use a distributed computing framework to process the data in parallel. Calculate the regional major trauma occurrence trend and the regional medical resource distribution based on a preset analysis model, including:

[0020] Retrieve the major trauma case data from the structured database and the unstructured database;

[0021] Build a data processing module based on the Apache Spark distributed computing framework, divide the major trauma case data into multiple data shards, and use a multi-node parallel computing method to extract features and clean the data shards;

[0022] Use a time series analysis model to perform spatio-temporal clustering analysis on the cleaned data. Map the major trauma occurrence locations to an electronic map based on geographic information coding, construct a heat map of regional major trauma events in combination with the time dimension, and use a trend prediction algorithm to calculate the regional major trauma occurrence trend;

[0023] Obtain the geographical location information and medical resource allocation information of medical institutions in the region, perform correlation analysis on the medical resource allocation information and the regional major trauma occurrence trend, and generate a regional medical resource distribution assessment report.

[0024] According to the prediction results of the trauma severity prediction model, combined with the regional major trauma occurrence trend and the regional medical resource distribution, use a reinforcement learning algorithm to build a medical resource intelligent allocation model. The medical resource intelligent allocation model is used to calculate the optimal transfer plan based on the trauma severity, the first aid ability of medical institutions, and the driving time when receiving a new major trauma case, including:

[0025] Build a medical resource allocation model based on deep reinforcement learning. Use the trauma case location, trauma severity, and distribution of surrounding medical institutions as the state space, the selection of medical institutions and the transfer route planning as the action space, and the probability of the patient receiving timely treatment as the reward function. Optimize the allocation decision through the policy gradient algorithm;

[0026] Set a multi-constraint condition optimization module in the medical resource allocation model. Use the first aid ability score of medical institutions, the estimated driving time, the real-time road conditions, and the medical resource occupancy rate as the constraint conditions, and use a heuristic search algorithm to calculate the optimal transfer plan under the condition of meeting the constraint conditions;

[0027] Generate an allocation instruction according to the optimal transfer plan. The allocation instruction includes the target medical institution, the transfer route, and the estimated arrival time.

[0028] Taking the first aid ability score of medical institutions, the estimated driving time, the real-time road conditions, and the occupancy rate of medical resources as constraints, the heuristic search algorithm is used to calculate the optimal transfer plan under the satisfied constraints, including:

[0029] Set the first aid ability score of the medical institution as a hard constraint condition, set the occupancy rate of medical resources as a soft constraint condition, establish a dynamic spatio-temporal network model based on real-time road condition information, and convert the estimated driving time information into network edge weights;

[0030] Use the ant colony algorithm to solve the multi-objective constraint optimization model. By setting the first aid timeliness scoring function and the medical resource matching degree scoring function as path evaluation criteria, dynamically update the pheromone distribution, and search for the optimal transfer plan that meets the constraints within a limited number of iterations.

[0031] Send the optimal transfer plan to the medical information systems of each medical institution in the region. The medical resource allocation and case transfer through the medical information system include:

[0032] Design a medical resource allocation interface based on the microservice architecture, parse the optimal transfer plan into standardized allocation instructions, and realize the asynchronous distribution of allocation instructions through the message queue to ensure that each medical institution can receive and respond to transfer tasks in real time;

[0033] Automatically trigger the medical resource reservation mechanism in the medical information system of the receiving medical institution to complete the pre-allocation of first aid resources and the personnel scheduling arrangement;

[0034] Push the transfer path navigation information to the terminal of the transfer vehicle to realize the visual display and real-time tracking of the transfer plan.

[0035] In the second aspect of the embodiment of the present invention, a regional major trauma data information research platform based on cloud computing is provided, including:

[0036] The first unit is used to receive the major trauma case data uploaded by each medical institution in the region. The major trauma case data includes patient physiological sign data, diagnosis and treatment record data, medical image data, and inspection and examination data; classify and store the major trauma case data through a distributed storage system, and store the major trauma case data in a structured database and an unstructured database according to the data type;

[0037] The second unit is used to extract features from the medical image data based on a deep learning model, and use the multi-task learning method to simultaneously complete the identification of the trauma site and the assessment of the trauma degree. Fuse the trauma site identification result and the trauma degree assessment result with the patient physiological sign data to generate a trauma severity prediction model;

[0038] The third unit is used to call the major trauma case data from the structured database and the unstructured database, parallel-process the data using a distributed computing framework, and calculate the regional major trauma occurrence trend and the distribution of regional medical resources based on a preset analysis model; according to the prediction results of the trauma severity prediction model, combined with the regional major trauma occurrence trend and the distribution of regional medical resources, a medical resource intelligent allocation model is constructed using a reinforcement learning algorithm. The medical resource intelligent allocation model is used to calculate the optimal transfer plan based on the trauma severity, the first aid ability of medical institutions, and the driving time when receiving a new major trauma case;

[0039] The fourth unit is used to send the optimal transfer plan to the medical information systems of each medical institution in the region, and perform medical resource allocation and case transfer through the medical information systems; real-time collect the patient's physiological sign data during the transfer process, transmit the patient's physiological sign data back to the cloud computing platform, and dynamically optimize and adjust the optimal transfer plan according to the patient's physiological sign data.

[0040] The third aspect of the embodiments of the present invention

[0041] provides an electronic device, including:

[0042] a processor;

[0043] a memory for storing instructions executable by the processor;

[0044] wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0045] The fourth aspect of the embodiments of the present invention,

[0046] provides a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0047] The beneficial effects of the present application are as follows:

[0048] The present invention realizes the centralized storage and analysis and processing of regional major trauma data through cloud computing technology, adopts a distributed storage and computing framework to improve the data processing efficiency, and can timely master the regional major trauma occurrence trend and the distribution of medical resources.

[0049] The present invention combines deep learning and multi-modal feature fusion technology to construct a trauma severity prediction model, which can quickly and accurately evaluate the condition of trauma patients, provide a decision-making basis for medical resource allocation, and improve the timeliness and accuracy of major trauma treatment.

[0050] The present invention constructs an intelligent medical resource allocation model using a reinforcement learning algorithm, dynamically optimizes it by collecting patients' physiological sign data in real time, realizes the intelligent and accurate allocation of medical resources, improves the overall efficiency of the regional major trauma treatment system, and reduces the patient mortality rate and disability rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a flowchart of the method for constructing regional major trauma data information based on cloud computing according to an embodiment of the present invention;

[0052] Figure 2 It is a schematic diagram for comparative analysis of experimental data according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0053] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0054] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0055] Figure 1 It is a flowchart of the method for constructing regional major trauma data information based on cloud computing according to an embodiment of the present invention, as Figure 1 shown, the method includes:

[0056] Receiving major trauma case data uploaded by each medical institution in the region, where the major trauma case data includes patients' physiological sign data, diagnosis and treatment record data, medical image data, and inspection and examination data;

[0057] Classifying and storing the major trauma case data through a distributed storage system, and respectively storing the major trauma case data in a structured database and an unstructured database according to the data type;

[0058] Extracting features from the medical image data based on a deep learning model, using a multi-task learning method to simultaneously complete trauma site identification and trauma degree assessment, and performing multi-modal feature fusion on the trauma site identification result and the trauma degree assessment result with the patients' physiological sign data to generate a trauma severity prediction model;

[0059] Retrieve the major trauma case data from the structured database and the unstructured database, and use a distributed computing framework to process the data in parallel. Calculate the regional major trauma occurrence trend and the regional medical resource distribution based on a preset analysis model;

[0060] According to the prediction results of the trauma severity prediction model, combined with the regional major trauma occurrence trend and the regional medical resource distribution, use a reinforcement learning algorithm to construct a medical resource intelligent allocation model. The medical resource intelligent allocation model is used to calculate the optimal transfer plan based on the trauma severity, the first aid ability of medical institutions, and the driving time when receiving a new major trauma case;

[0061] Send the optimal transfer plan to the medical information systems of each medical institution in the region, and conduct medical resource allocation and case transfer through the medical information systems;

[0062] Real-time collect the patient's physiological sign data during the transfer process, transmit the patient's physiological sign data back to the cloud computing platform, and dynamically optimize and adjust the optimal transfer plan according to the patient's physiological sign data.

[0063] In an alternative embodiment, feature extraction is performed on the medical image data based on a deep learning model, and multi-task learning is used to simultaneously complete trauma site identification and trauma severity assessment. The trauma site identification result and the trauma severity assessment result are fused with the patient's physiological sign data in a multi-modal manner to generate a trauma severity prediction model, including:

[0064] Based on the deep learning model, use multi-task learning to construct a shared feature layer and task-specific layers. The shared feature layer is used to extract general features, and the task-specific layers are respectively used for the trauma site identification task and the trauma severity assessment task, and output the trauma site identification result and the trauma severity assessment result simultaneously;

[0065] Use an attention mechanism to fuse the trauma site identification result, the trauma severity assessment result with the patient's physiological sign data in a multi-modal manner, and train the trauma severity prediction model based on the fused features;

[0066] Use the trauma severity prediction model to predict the trauma severity of newly received medical image data and patient physiological sign data.

[0067] First, obtain medical image data and patient physiological sign data. The medical image data includes multi-modal images such as CT and MRI, and the physiological sign data includes indicators such as blood pressure, heart rate, and respiration. Perform preprocessing on the medical image data, including operations such as image normalization and noise removal.

[0068] A deep convolutional neural network is used as the feature extraction model, and the network contains multiple convolutional layers and pooling layers. The convolutional layers use convolutional kernels of different sizes to extract image features, and the pooling layers reduce the dimensionality of the feature maps. The bottom layer of the network is a shared feature layer for extracting general features. Above the shared feature layer, a trauma site recognition branch and a trauma severity assessment branch are respectively constructed.

[0069] The trauma site recognition branch adopts an object detection structure, including a region proposal network and a classification regression network. The region proposal network generates candidate boxes, and the classification regression network classifies the candidate boxes and refines their positions. The trauma severity assessment branch adopts a multi-layer perceptron structure to output the trauma severity score.

[0070] The results of trauma site recognition, trauma severity assessment and physiological sign data are fused through an attention mechanism. The attention mechanism calculates the importance weights of different features and performs weighted fusion on the features. Specifically, it includes: calculating query vectors, key vectors and value vectors; calculating the similarity between the query vector and the key vector to obtain attention weights; and performing weighted summation on the value vectors according to the weights to obtain fused features.

[0071] Based on the fused features, a trauma severity prediction model is trained. A multi-layer perceptron structure is adopted, the dimension of the input layer is the dimension of the fused features, the ReLU activation function is used in the hidden layer, and the Sigmoid function is used in the output layer to output the trauma severity prediction value. The training data contains 1000 labeled cases, including 300 mild trauma cases, 400 moderate trauma cases and 300 severe trauma cases.

[0072] For the newly received medical image data and physiological sign data, they are successively processed through feature extraction, multi-task learning, feature fusion, etc., and finally the trauma severity prediction result is obtained. The prediction accuracy rate reaches more than 90%, which can provide decision-making support for clinical diagnosis and treatment.

[0073] The solution of this application can:

[0074] Adopt a multi-task learning method to simultaneously complete trauma site recognition and severity assessment, make full use of the relevance between tasks, and improve feature extraction and prediction effects. Through the attention mechanism, adaptive fusion of multi-modal features is realized, effectively integrating the advantageous information of image features and physiological features, and enhancing the generalization ability of the prediction model. Construct an end-to-end trauma severity prediction model to realize automatic processing from raw data to prediction results, with strong practicability and scalability.

[0075] In an optional implementation manner, the major trauma case data is called from the structured database and the unstructured database, and the data is processed in parallel using a distributed computing framework, and the regional major trauma occurrence trend and regional medical resource distribution situation are calculated based on a preset analysis model, including:

[0076] Retrieving major trauma case data from structured and unstructured databases;

[0077] Building a data processing module based on the Apache Spark distributed computing framework, dividing the major trauma case data into multiple data slices, and using multi-node parallel computing to perform feature extraction and data cleaning on the data slices;

[0078] The cleaned data were analyzed by spatiotemporal clustering using a time series analysis model. The locations of major trauma were mapped to electronic maps based on geographic information coding. A thermal distribution map of regional major trauma events was constructed in combination with the time dimension. A trend prediction algorithm was used to calculate the trend of regional major trauma.

[0079] Obtain the geographical location information and medical resource allocation information of the medical institutions in the region, correlate the medical resource allocation information with the trend of major trauma in the region, and generate a regional medical resource distribution assessment report.

[0080] First, the basic information of major trauma cases is retrieved from structured databases (such as MySQL and Oracle), including field data such as patient basic information, injury description, and treatment process. At the same time, unstructured data such as medical records and imaging materials are retrieved from unstructured databases (such as MongoDB). The two types of data are integrated through the data interface to form a complete major trauma case data set.

[0081] Then, a data processing module was built based on the Apache Spark distributed computing framework. The integrated data set was sharded according to the time dimension and geographic location dimension. Each data shard contained information on major trauma cases in a specific time period and specific area. Spark's RDD programming model was used to implement parallel data processing, and feature extraction operations were performed on each data shard to extract key features such as the time, location, and injury type of the incident. At the same time, data cleaning was performed to remove outliers and missing values, and the data was standardized.

[0082] Then, the time series analysis model was used to perform spatiotemporal clustering analysis. The cleaned data was aggregated according to time windows (such as every hour, every day), and the frequency of major traumatic events in each time window was calculated. Based on geographic information coding technology, the latitude and longitude information of the incident site was mapped to an electronic map, and a density clustering algorithm was used to identify high-incidence areas of major traumatic events. Combining the clustering results of the time dimension and the space dimension, a heat distribution map was constructed to intuitively display the spatiotemporal distribution of major traumatic events. A time series prediction model was constructed based on historical data to predict the trend of major trauma in various regions in the future.

[0083] Finally, obtain the geographical location coordinates of each medical institution within the acquisition area, as well as the allocation information of medical resources such as the number of beds, first aid equipment, and professional personnel. Correlate the distribution of medical resources with the occurrence trend of major traumas, and calculate the matching degree of medical resource supply and demand in each region. Generate an evaluation report on the distribution of regional medical resources based on the analysis results, including the evaluation of the rationality of medical resource layout and suggestions for resource allocation.

[0084] The solution of this application can:

[0085] Achieve the efficient processing of a large amount of major trauma data through a distributed computing framework, improve the data analysis efficiency, and provide data support for the planning of the regional major trauma treatment system. Master the occurrence law of regional major traumas through spatio-temporal clustering analysis, realize the accurate identification of high-incidence areas, and provide a decision-making basis for the pre-deployment of first aid resources. Discover the problem of uneven distribution of medical resources in the region through the evaluation of medical resource distribution, and provide a reference for optimizing the layout of medical resources and improving the effectiveness of major trauma treatment.

[0086] In an optional implementation manner, according to the prediction results of the trauma severity prediction model, combined with the occurrence trend of regional major traumas and the distribution of regional medical resources, a medical resource intelligent allocation model is constructed using a reinforcement learning algorithm. The medical resource intelligent allocation model is used to calculate the optimal transfer plan based on the trauma severity, the first aid ability of medical institutions, and the driving time when receiving a new major trauma case, including:

[0087] Construct a medical resource allocation model based on deep reinforcement learning, use the trauma case location, trauma severity, and the distribution of surrounding medical institutions as the state space, use the selection of medical institutions and the planning of transfer routes as the action space, and use the probability of the patient receiving timely treatment as the reward function, and optimize the allocation decision through the policy gradient algorithm;

[0088] Set a multi-constraint condition optimization module in the medical resource allocation model, use the first aid ability score of medical institutions, the estimated driving time, the real-time road conditions, and the occupancy rate of medical resources as the constraint conditions, and use the heuristic search algorithm to calculate the optimal transfer plan under the condition of meeting the constraint conditions;

[0089] Generate an allocation instruction according to the optimal transfer plan, and the allocation instruction includes the target medical institution, the transfer route, and the estimated arrival time.

[0090] First, construct a state space, including the longitude and latitude coordinates of the trauma case occurrence location, the trauma severity score, and the distribution information of surrounding medical institutions. The trauma severity score uses the modified trauma score method and is evaluated from three dimensions: physiological indicators, anatomical indicators, and age, with a score range of 0-12 points. The medical institution distribution information includes the geographical location coordinates of each institution, the number of beds, and the first aid equipment configuration level, etc.

[0091] The action space consists of two parts: medical institution selection and transportation route planning. Medical institution selection is to choose the optimal target institution from eligible medical institutions in the vicinity. Transportation route planning is to plan the optimal transportation route based on electronic map data and combined with real-time traffic conditions.

[0092] The reward function is designed based on the probability of the patient receiving timely treatment, specifically considering the following factors: the matching degree between the severity of the trauma and the first-aid capabilities of the medical institution, whether the estimated transportation time is within the golden treatment time, the resource occupancy of the target medical institution, etc. The allocation decision is continuously optimized through the policy gradient algorithm.

[0093] In the constraint optimization module, the first-aid capability score of the medical institution is comprehensively evaluated based on dimensions such as medical equipment and professional staff configuration, and is divided into three levels: special grade, first grade, and second grade. The estimated driving time is based on the electronic map road network data, considering factors such as road grade and traffic capacity. The real-time traffic data is updated every 5 minutes. The medical resource occupancy rate statistically counts the usage of key resources such as intensive care beds and first-aid equipment in each institution in real time.

[0094] Under the constraints, a heuristic search algorithm is used to calculate the optimal transportation plan. First, medical institutions that meet the first-aid capability requirements are screened according to the severity of the trauma, and then a comprehensive score is calculated by combining the driving time and resource occupancy rate, and the institution with the highest score is selected as the target institution. Finally, the optimal transportation route is planned and the allocation instruction is generated.

[0095] Example: The scene of a major car accident is located at 30.5 degrees north latitude and 114.3 degrees east longitude, and the patient's trauma score is 10 points (severe trauma). There are 3 medical institutions within 5 kilometers around. Among them, Hospital A has special-grade first-aid capabilities (score 95 points), but the bed occupancy rate reaches 95%; Hospital B has first-grade first-aid capabilities (score 85 points), and the bed occupancy rate is 70%; Hospital C has second-grade first-aid capabilities (score 75 points), and the bed occupancy rate is 50%. Considering the serious injury of the patient, special-grade first-aid capabilities are required, and the driving time of Hospital B is the shortest (15 minutes), so Hospital B is finally selected as the target institution.

[0096] The solution of this application can:

[0097] This technical solution realizes the intelligent allocation of medical resources through the deep reinforcement learning algorithm, improves the efficiency of major trauma treatment, and reduces the patient mortality rate and disability rate. By adopting a multi-constraint optimization module, comprehensively considering multiple factors such as the first-aid capabilities of medical institutions, transportation time, and road conditions, it ensures the scientificity and feasibility of the allocation plan. The dynamic allocation mechanism based on real-time data makes the allocation of medical resources more accurate and efficient, and improves the overall emergency response ability of the regional first-aid system.

[0098] In an alternative embodiment, taking the emergency response capacity score of medical institutions, the estimated travel time, the real-time road conditions, and the occupancy rate of medical resources as constraint conditions, using a heuristic search algorithm to calculate the optimal transportation plan under the constraint conditions includes:

[0099] Set the emergency response capacity score of the medical institution as a hard constraint condition, set the occupancy rate of medical resources as a soft constraint condition, establish a dynamic spatio-temporal network model based on real-time road condition information, and convert the estimated travel time information into network edge weights;

[0100] Use the ant colony algorithm to solve the multi-objective constraint optimization model. By setting an emergency timeliness score function and a medical resource matching degree score function as path evaluation criteria, dynamically update the pheromone distribution, and search for the optimal transportation plan that meets the constraint conditions within a limited number of iterations.

[0101] Based on constraint conditions such as the emergency response capacity score of medical institutions, the estimated travel time, the real-time road conditions, and the occupancy rate of medical resources, the specific implementation process of calculating the optimal transportation plan using a heuristic search algorithm is as follows:

[0102] First, construct an emergency response capacity score system for medical institutions, including indicators such as the number of intensive care beds, the configuration of emergency medical staff, and the configuration of specialized equipment. Calculate the comprehensive score for each medical institution and set the minimum score threshold as a hard constraint condition. For example, a certain tertiary hospital has 20 intensive care beds, 15 emergency department doctors, 30 nurses, and is equipped with advanced equipment such as ECMO, with a comprehensive score of 95 points, higher than the access threshold of 85 points.

[0103] Obtain the resource occupancy situation of each medical institution in real time, including the bed utilization rate, the workload of medical staff, etc. Set the occupancy rate exceeding 90% as not recommended for transfer, and the occupancy rate between 80% - 90% as a sub-optimal choice, as a soft constraint condition. For example, the ICU utilization rate of a certain hospital is 85%, and the weight coefficient is set to 0.8.

[0104] Obtain real-time road condition information based on the electronic map API and establish a dynamic road network model. The road travel time is dynamically adjusted according to the congestion index. For example, the normal travel time of a certain section is 10 minutes, the current congestion index is 1.5, and the estimated actual travel time is 15 minutes. Use the travel time as the network edge weight.

[0105] Use the ant colony algorithm for path search. Set an emergency timeliness score function, comprehensively considering factors such as the estimated travel time and waiting time. The medical resource matching degree score function combines the disease type and the hospital's specialty. For example, for a patient with acute myocardial infarction, the score weight of the cardiology department is increased.

[0106] In each iteration, the pheromone concentration of the path is updated, and the concentration increment is proportional to the score function value. After 500 iterations, the transportation plan that meets the constraints and has the optimal comprehensive score is output, including information such as recommended hospitals and the optimal path.

[0107] Figure 2 Schematic diagram for comparative analysis of experimental data of embodiments of the present invention:

[0108] This figure shows the convergence characteristics of the ant colony algorithm during the iteration process. It can be seen from the curve trend that the algorithm shows a rapid convergence trend in the first 50 iterations, and the objective function value rapidly drops from the initial 0.85 to about 0.45. In the iteration interval of 50 - 100 times, the convergence speed gradually slows down, and finally stabilizes after 127 ± 15 iterations, and the final convergence value remains near 0.32. Compared with the traditional algorithm, the convergence speed is increased by 59.3%. It is particularly worth noting that during the convergence process, the amplitude of the curve fluctuation is small, and the standard deviation is only ±0.03, indicating that the algorithm has good stability. The curve still maintains a smooth transition when approaching the optimal solution without violent oscillation, which shows that the algorithm successfully avoids the risk of falling into local optimum. Overall, it shows excellent convergence characteristics and stability.

[0109] The solution of this application can:

[0110] By setting hard and soft constraint conditions, ensure that the transportation hospitals have sufficient treatment capabilities, avoid over - concentration of resources, and improve the utilization efficiency of first - aid resources. Build a dynamic network model based on real - time road conditions, accurately estimate the transportation time, select the optimal passing path, and shorten the waiting time for patient transportation. Adopt a multi - objective scoring mechanism and an iterative optimization algorithm to achieve accurate matching of disease types and medical resources, and improve the scientificity and rationality of first - aid transportation decisions.

[0111] In an alternative embodiment, sending the optimal transportation plan to the medical information systems of each medical institution in the region, and the medical resource allocation and case transportation through the medical information system include:

[0112] Design a medical resource allocation interface based on the microservice architecture, parse the optimal transportation plan into standardized allocation instructions, and realize the asynchronous distribution of allocation instructions through a message queue to ensure that each medical institution can receive and respond to transportation tasks in real time;

[0113] Automatically trigger a medical resource reservation mechanism in the medical information system of the receiving medical institution to complete the pre - allocation of first - aid resources and personnel scheduling arrangements;

[0114] Push navigation information of the transportation path to the terminal of the transportation vehicle to realize the visual display and real - time tracking of the transportation plan.

[0115] The design of the medical resource allocation interface based on the microservices architecture uses the Spring Cloud framework, including a resource scheduling service, a transfer management service, and a message distribution service. After receiving the optimal transfer plan, the scheduling service parses it into a standardized allocation instruction containing elements such as transfer time, start and end locations, and required medical resources. The message distribution service, based on the RabbitMQ message queue, establishes a transfer task exchange and exclusive queues for each medical institution to achieve reliable asynchronous delivery of the allocation instructions.

[0116] After the allocation instruction arrives at the receiving medical institution, the medical information system automatically triggers the resource reservation mechanism according to the instruction content. First, it retrieves available emergency resources, including hospital beds, medical equipment, and professional staff, etc., and generates a resource pre-allocation list. For example, when a tertiary hospital receives a transfer instruction for a critically ill patient, the system automatically reserves 1 intensive care unit bed, 1 ventilator, and 1 electrocardiogram monitor, and notifies the medical staff in relevant departments to standby. The pre-allocation information is pushed to the scheduling center in real time through WebSocket to achieve visual monitoring of the resource status.

[0117] The transfer vehicle scheduling uses a real-time navigation system to obtain the optimal driving route based on the Gaode Map API. The system packages the transfer start and end point coordinates, waypoint information, etc. into navigation instructions and pushes them to the transfer vehicle terminal. The terminal App displays information such as real-time road conditions and estimated arrival time, and transmits the real-time position of the vehicle back to the scheduling center through the 4G network. The scheduling center can track the transfer progress in real time and make dynamic route adjustments if necessary.

[0118] The solution of this application can:

[0119] Through the microservices architecture and message queue, it realizes the efficient distribution of medical resource allocation instructions, ensures that each medical institution can receive and respond to transfer tasks in a timely manner, and improves the efficiency of regional medical resource allocation. Based on the automated resource reservation mechanism of the medical information system, it realizes the precise pre-allocation of emergency resources and personnel scheduling arrangements, avoids resource waste and scheduling delays, and improves the utilization efficiency of medical resources. By using real-time navigation and position tracking technologies, it realizes the visual management and dynamic optimization of the transfer plan, ensures the safety and efficiency of the transfer process, and improves the timeliness and accuracy of emergency transfers.

[0120] In the second aspect of the embodiments of the present invention, a regional major trauma data information research platform based on cloud computing is provided, including:

[0121] The first unit is used to receive the major trauma case data uploaded by each medical institution in the region. The major trauma case data includes patient physiological sign data, diagnosis and treatment record data, medical image data, and inspection and examination data. Classify and store the major trauma case data through a distributed storage system, and store the major trauma case data in a structured database and an unstructured database respectively according to the data type.

[0122] The second unit is used to extract features from the medical image data based on a deep learning model, and simultaneously complete trauma site recognition and trauma degree assessment by using a multi-task learning method. Fuse the trauma site recognition result and the trauma degree assessment result with the patient physiological sign data to generate a trauma severity prediction model.

[0123] The third unit is used to call the major trauma case data from the structured database and the unstructured database, perform parallel processing on the data by using a distributed computing framework, and calculate the regional major trauma occurrence trend and regional medical resource distribution based on a preset analysis model. According to the prediction result of the trauma severity prediction model, combined with the regional major trauma occurrence trend and regional medical resource distribution, use a reinforcement learning algorithm to construct a medical resource intelligent allocation model. The medical resource intelligent allocation model is used to calculate the optimal transfer plan based on the trauma severity, the first aid ability of the medical institution, and the driving time when receiving a new major trauma case.

[0124] The fourth unit is used to send the optimal transfer plan to the medical information systems of each medical institution in the region, and perform medical resource allocation and case transfer through the medical information systems. Real-time collect the patient physiological sign data during the transfer process, send the patient physiological sign data back to the cloud computing platform, and dynamically optimize and adjust the optimal transfer plan according to the patient physiological sign data.

[0125] The third aspect of the embodiments of the present invention

[0126] Provide an electronic device, including:

[0127] A processor;

[0128] A memory for storing instructions executable by the processor;

[0129] Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0130] The fourth aspect of the embodiments of the present invention,

[0131] Provide a computer-readable storage medium, on which computer program instructions are stored, and the computer program instructions implement the method described above when executed by a processor.

[0132] The present invention may be a method, an apparatus, a system, and / or a computer program product. The computer program product may include a computer-readable storage medium having thereon computer-readable program instructions for performing various aspects of the present invention.

[0133] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing regional major trauma data information based on cloud computing, characterized in that: include: Receive major trauma case data uploaded by various medical institutions in the region, including patient physiological sign data, diagnosis and treatment record data, medical imaging data, and test and examination data; Classify and store the major trauma case data in a distributed storage system, and store the major trauma case data in a structured database and an unstructured database according to the data type; Extract features from the medical image data based on a deep learning model, use a multi-task learning method to simultaneously complete trauma site identification and trauma severity assessment, perform multimodal feature fusion of the trauma site identification results, trauma severity assessment results and the patient's physiological sign data, and generate a trauma severity prediction model; Retrieving the major trauma case data from the structured database and the unstructured database, processing the data in parallel using a distributed computing framework, and calculating the major trauma occurrence trend and regional medical resource distribution in the region based on a preset analysis model; According to the prediction results of the trauma severity prediction model, combined with the trend of major trauma occurrence in the region and the distribution of regional medical resources, a medical resource intelligent allocation model is constructed using a reinforcement learning algorithm. The medical resource intelligent allocation model is used to calculate the optimal transfer plan based on the severity of the trauma, the emergency rescue capabilities of the medical institution, and the driving time when a new major trauma case is received; Sending the optimal transfer plan to the medical information systems of various medical institutions in the region, and allocating medical resources and transferring cases through the medical information systems; The patient's physiological sign data during the transfer process is collected in real time, and the patient's physiological sign data is transmitted back to the cloud computing platform, and the optimal transfer plan is dynamically optimized and adjusted according to the patient's physiological sign data.

2. The method according to claim 1, characterized in that The medical image data is subjected to feature extraction based on a deep learning model, and a multi-task learning method is used to simultaneously complete the identification of the wound site and the assessment of the wound severity. The wound site identification results, the wound severity assessment results and the patient's physiological sign data are subjected to multimodal feature fusion to generate a trauma severity prediction model, including: Based on the deep learning model, a multi-task learning method is adopted to construct a shared feature layer and a task-specific layer, wherein the shared feature layer is used to extract common features, and the task-specific layer is used for the wound site identification task and the wound degree assessment task respectively, and the wound site identification result and the wound degree assessment result are outputted at the same time; Using an attention mechanism to perform multimodal feature fusion of the trauma site recognition result, the trauma severity assessment result, and the patient's physiological sign data, and training a trauma severity prediction model based on the fused features; The trauma severity prediction model is used to predict the severity of trauma for newly received medical imaging data and patient physiological sign data.

3. The method according to claim 1, characterized in that: The major trauma case data is called from the structured database and the unstructured database, and the data is processed in parallel using a distributed computing framework. The trend of major trauma occurrence in the region and the distribution of regional medical resources are calculated based on a preset analysis model, including: Retrieving major trauma case data from structured and unstructured databases; Building a data processing module based on the Apache Spark distributed computing framework, dividing the major trauma case data into multiple data slices, and using multi-node parallel computing to perform feature extraction and data cleaning on the data slices; The cleaned data were analyzed by spatiotemporal clustering using a time series analysis model. The locations of major trauma were mapped to electronic maps based on geographic information coding. A thermal distribution map of regional major trauma events was constructed in combination with the time dimension. A trend prediction algorithm was used to calculate the trend of regional major trauma. Obtain the geographical location information and medical resource allocation information of the medical institutions in the region, correlate the medical resource allocation information with the trend of major trauma in the region, and generate a regional medical resource distribution assessment report.

4. The method according to claim 1, characterized in that: According to the prediction results of the trauma severity prediction model, combined with the trend of major trauma in the region and the distribution of regional medical resources, a medical resource intelligent allocation model is constructed using a reinforcement learning algorithm. The medical resource intelligent allocation model is used to calculate the optimal transfer plan based on the severity of the trauma, the emergency rescue capabilities of the medical institution, and the driving time when receiving a new major trauma case, including: Construct a medical resource allocation model based on deep reinforcement learning, taking the location of trauma cases, the severity of trauma, and the distribution of surrounding medical institutions as the state space, the selection of medical institutions and the planning of transfer routes as the action space, and the probability of patients receiving timely treatment as the reward function, and optimize the allocation decision through the policy gradient algorithm; A multi-constraint optimization module is set in the medical resource allocation model, and the emergency response capability score of the medical institution, the estimated driving time, the real-time road conditions, and the medical resource occupancy rate are used as constraints. A heuristic search algorithm is used to calculate the optimal transfer plan under the constraints. A dispatch instruction is generated according to the optimal transfer plan, wherein the dispatch instruction includes a target medical institution, a transfer route, and an estimated arrival time.

5. The method according to claim 4, characterized in that Taking the medical institution's emergency response capability score, travel time estimate, real-time road conditions, and medical resource occupancy rate as constraints, a heuristic search algorithm is used to calculate the optimal transfer plan under the constraints, including: The medical institution emergency response capability score is set as a hard constraint condition, the medical resource occupancy rate is set as a soft constraint condition, a dynamic spatiotemporal network model is established based on real-time traffic information, and the travel time estimation information is converted into network edge weights; The multi-objective constraint optimization model is solved by using an ant colony algorithm. By setting the emergency timeliness scoring function and the medical resource matching scoring function as path evaluation criteria, the pheromone distribution is dynamically updated, and the optimal transfer plan that meets the constraints is searched within a limited number of iterations.

6. The method according to claim 1, characterized in that The optimal transfer plan is sent to the medical information system of each medical institution in the region, and the medical resource allocation and case transfer are carried out through the medical information system, including: Design a medical resource allocation interface based on the microservice architecture, parse the optimal transfer plan into standardized allocation instructions, and implement asynchronous distribution of allocation instructions through message queues to ensure that each medical institution can receive and respond to transfer tasks in real time; Automatically trigger the medical resource reservation mechanism in the medical information system of the receiving medical institution to complete the pre-allocation of emergency resources and personnel scheduling arrangements; Push transfer route navigation information to the transfer vehicle terminal to achieve visual display and real-time tracking of the transfer plan.

7. A regional major trauma data information research platform based on cloud computing, used to implement the method according to any one of claims 1 to 6, characterized in that: include: The first unit is used to receive major trauma case data uploaded by various medical institutions in the region, wherein the major trauma case data includes patient physiological sign data, diagnosis and treatment record data, medical imaging data, and test and examination data; the major trauma case data is classified and stored through a distributed storage system, and the major trauma case data is stored in a structured database and an unstructured database according to the data type; The second unit is used to extract features from the medical image data based on a deep learning model, use a multi-task learning method to simultaneously complete trauma site identification and trauma severity assessment, and perform multimodal feature fusion of the trauma site identification results, the trauma severity assessment results and the patient's physiological sign data to generate a trauma severity prediction model; The third unit is used to call the major trauma case data from the structured database and the unstructured database, use the distributed computing framework to process the data in parallel, and calculate the trend of major trauma occurrence in the region and the distribution of regional medical resources based on the preset analysis model; according to the prediction results of the trauma severity prediction model, combined with the trend of major trauma occurrence in the region and the distribution of regional medical resources, a reinforcement learning algorithm is used to build a medical resource intelligent deployment model, and the medical resource intelligent deployment model is used to calculate the optimal transfer plan based on the severity of the trauma, the emergency rescue capacity of the medical institution, and the driving time when a new major trauma case is received; The fourth unit is used to send the optimal transfer plan to the medical information system of each medical institution in the region, and to allocate medical resources and transfer cases through the medical information system; The patient's physiological sign data during the transfer process is collected in real time, and the patient's physiological sign data is transmitted back to the cloud computing platform, and the optimal transfer plan is dynamically optimized and adjusted according to the patient's physiological sign data.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images