Out-of-hospital patient medical record generation method and system

Abstract

The invention relates to the field of medical informatization, and discloses an out-of-hospital patient medical record generation method and system.The method comprises the steps that physical sign data, symptom description data and geographic space information of a patient are obtained and preprocessed; data from different sources are constructed into graph nodes through a spatio-temporal topology network model, and a graph convolutional network and a long-short term memory network are utilized to process spatial relevance and time dependency of the data respectively; fusing the multi-modal data, and complementing missing data through Bayesian inference; and finally, generating a patient electronic medical record based on the fused data, and outputting the patient electronic medical record to a doctor for follow-up visit and health management. According to the method, the accuracy and integrity of medical record generation of the patient outside the hospital can be improved, and personalized health management and efficient doctor work support are provided.

Description

Technical Field

[0001] The present invention relates to the technical field of medical informatization, and specifically to a method and system for generating medical records of out-of-hospital patients. Background Art

[0002] With the development of medical informatization, the application of electronic medical records has become increasingly widespread. Especially in the health management of out-of-hospital patients, the generation and use of electronic medical records play a crucial role. Traditional methods for generating medical records usually rely on regular hospital visit records of patients and manually input and update patients' health data. However, in out-of-hospital follow-up, patients' physical sign data, symptom descriptions, and health conditions often rely on patients to provide actively. This mode often faces problems such as incomplete data, information lag, and inconsistent records.

[0003] In the prior art, although some physical sign data collection technologies based on Internet of Things devices have emerged, and some systems can collect patients' symptom description information, there are still great difficulties in integrating data from multiple sources. The effective combination of physical sign data, symptom description data, and geospatial information has not been fully solved. Especially, there is a lack of effective processing methods for data association in the spatial and temporal dimensions. Most existing medical record generation systems ignore the interaction relationship between patients' health status and the environment they are in, resulting in the generated electronic medical records being unable to fully reflect patients' health conditions.

[0004] In addition, although some methods attempt to use data filling and speculation of missing values, they often lack accurate algorithms to speculate missing data, resulting in incomplete or misinferred data, which affects the reliability and accuracy of the finally generated medical records. In this context, how to efficiently and accurately integrate multi-modal data from different sources and reasonably fill in missing data has become an important challenge faced by current technologies. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides a method and system for generating medical records of out-of-hospital patients, which solves the problems of insufficient multi-modal data fusion, inaccurate filling of missing data, and data inconsistency in the process of generating medical records of out-of-hospital patients in the prior art.

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: A method for generating medical records of out-of-hospital patients includes the following steps: Obtain the physical sign data of the patient, the symptom description data of the patient, and the geospatial information of the patient, and perform preprocessing; Based on the spatio-temporal topological network model, construct the physical sign data, symptom description data, and geospatial information of the patient as graph nodes, and process the spatial association relationship between the nodes through a graph convolutional network, and process the time dependence relationship of the nodes by using a long short-term memory network; Fuse multi-modal data, integrate spatio-temporal correlation data through a graph convolutional network, and use Bayesian inference to complete missing data; Generate an electronic medical record for the patient based on the fused data and output the generated medical record.

[0007] Preferably, the obtaining of the patient's vital sign data includes obtaining the patient's physiological vital sign data through Internet of Things devices; the obtaining of the patient's symptom description data includes text or voice data uploaded by the patient through a mini-program or voice input; the obtaining of the patient's geospatial information includes obtaining the patient's current location data through the patient's address information or positioning service.

[0008] Preferably, the step of constructing the patient's vital sign data, symptom description data, and geospatial information into graph nodes based on the spatio-temporal topology network model includes: Construct corresponding node representations for each type of data respectively, and use the node representations as basic elements in the graph; Establish edges between nodes according to the time sequence and spatial relationship to form the adjacency matrix of the graph, and the adjacency matrix is used to represent the relationship between data nodes; Use a graph convolutional network to process the node information in the graph, and transmit and update the spatial correlation between nodes through the adjacency matrix of the graph.

[0009] Preferably, the step of processing the spatial correlation relationship between nodes through a graph convolutional network includes: Use the node update rule of the graph convolutional network to perform weighted summation on the features of each node, combined with the weight information in the adjacency matrix; Hierarchically process the node features through a multi-layer graph convolutional network, gradually transmit and aggregate the information from neighbor nodes, so as to capture the spatial dependence of the data; Perform non-linear activation on the output result of each layer of the graph convolutional network.

[0010] Preferably, the step of processing the time dependence relationship of nodes by using a long short-term memory network includes: Use the node features output by the graph convolutional network as the input data of the long short-term memory network to capture the evolution law of the data in the time dimension; Use the long short-term memory network to process the state update of nodes at different time steps, and capture the time dependence of nodes through cyclic calculation; Update the time-dependent features of each node to generate node representations with time information.

[0011] Preferably, the step of fusing multi-modal data includes: Connect the node features processed by the graph convolutional network with the node features from different data sources to form multi-modal fusion features; Apply a further graph convolutional network layer to the multi-modal fusion features; In the multi-layer processing of the graph convolutional network, fuse different modal data according to the spatial correlation and temporal dependence of nodes.

[0012] Preferably, the step of integrating the spatio-temporal correlation data includes: Perform weighted summation on the node features from different time steps and spatial positions, and perform information propagation and synthesis between nodes through the adjacency matrix in the graph convolutional network; During the integration process, use a non-linear activation function to process the weighted features of nodes to further enhance the feature representation after data fusion; Utilize the hierarchical structure of the graph convolutional network to gradually extract comprehensive information from different time steps and spatial positions at each layer.

[0013] Preferably, the step of complementing missing data using Bayesian inference includes: Perform probability speculation on the missing data based on the existing data, and calculate the posterior probability distribution of the missing data through the Bayesian inference model; Utilize the patient's historical health data, similar patient data, and spatial and temporal correlation information to complement the missing data; Assign credibility weights to the complemented data, and use the estimation result of the missing data as the final output.

[0014] Preferably, the step of generating a patient's electronic medical record based on the fused data and outputting the generated medical record includes: Generate a comprehensive electronic medical record including the patient's physical sign data, symptom description data, health status information, and geospatial information based on the fused multi-modal data; Format the generated electronic medical record to make it suitable for doctors' review and subsequent follow-up, and the formatting process includes organizing the data into structured medical record information; Output the formatted electronic medical record to the doctor's terminal so that doctors can perform subsequent diagnosis, health management, and treatment tracking based on the generated medical record.

[0015] The present invention also provides an out-of-hospital patient medical record generation system, including: A data acquisition module for obtaining the patient's physical sign data through Internet of Things devices, obtaining the patient's symptom description data through mini-programs or voice input methods, and obtaining the patient's geospatial information; A data processing module for preprocessing the physical sign data, symptom description data, and geospatial information, including noise suppression of sensor data by Kalman filtering, and correction of symptom description data by OCR and natural language processing; A spatio-temporal correlation modeling module for constructing the physical sign data, symptom description data, and geospatial information as graph nodes based on a spatio-temporal topology network model, processing the spatial association relationships between nodes through a graph convolutional network, and processing the time dependence relationships of nodes using a long short-term memory network; A data fusion and completion module for fusing multi-modal data, integrating spatio-temporal correlation data through a graph convolutional network, and completing missing data using Bayesian inference; A medical record generation module for generating an electronic medical record of a patient based on the fused data and outputting the generated medical record to a doctor's terminal for the doctor to use for follow-up and health management.

[0016] The present invention provides a method and system for generating medical records of out-of-hospital patients. It has the following beneficial effects: Through multi-modal data fusion and a spatio-temporal topology network model, the present invention can effectively capture the relationships between a patient's physical sign data, symptom description data, and geospatial information, ensuring the accurate fusion and integrity of various types of data. In particular, the use of a graph convolutional network and a long short-term memory network (LSTM) to model the spatial and temporal dependencies of data improves the quality and reliability of the data.

[0017] By introducing Bayesian inference, the present invention can make reasonable inferences and completions based on known data in the case of missing patient data. The Bayesian inference model uses a patient's historical data and the data of similar patients to provide a high-confidence estimate of missing data, avoiding the situation where the generation of medical records is affected due to incomplete data.

[0018] Combining a patient's geospatial information and health status data, the present invention can provide a more personalized health assessment and management plan for doctors. The introduction of geospatial information takes into account the impact of environmental factors on a patient's health, enabling doctors to make more accurate follow-up and treatment decisions based on the specific situation of the patient.

[0019] After the electronic medical record generated by the present invention is formatted, it is convenient for doctors to quickly consult and use, reducing the cumbersome operations of traditional medical record management. Through automated data processing and medical record generation, doctors can more efficiently conduct patient follow-up, health assessment, and treatment plan formulation, improving work efficiency and the quality of diagnosis and treatment. Brief Description of the Drawings

[0020] Figure 1 It is a schematic diagram of the method flow of the present invention; Figure 2This is a schematic diagram of the system structure of the present invention.

[0021] Among them, 10 is the data acquisition module; 20 is the data processing module; 30 is the spatio-temporal correlation modeling module; 40 is the data fusion and completion module; 50 is the medical record generation module. Specific implementation manners

[0022] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0023] Please refer to the attached Figure 1 , the present invention provides a method for generating an out-of-hospital patient medical record, aiming to obtain the patient's physical sign data, symptom description data, and geospatial information, and combine advanced spatio-temporal topology network technology, graph convolutional network (GCN), long short-term memory network (LSTM), and Bayesian inference method to achieve multi-modal fusion of the patient's health data and completion of missing data, and finally generate a complete out-of-hospital patient electronic medical record.

[0024] As Figure 1 shown, the method for generating an out-of-hospital patient medical record may include the following steps: S1. Obtain the patient's physical sign data, the patient's symptom description data, and the patient's geospatial information, and perform preprocessing; S2. Based on the spatio-temporal topology network model, construct the patient's physical sign data, symptom description data, and geospatial information into graph nodes, process the spatial association relationship between the nodes through the graph convolutional network, and process the time-dependent relationship of the nodes by using the long short-term memory network; S3. Perform fusion on the multi-modal data, integrate the spatio-temporal correlation data through the graph convolutional network, and use Bayesian inference to complete the missing data; S4. Generate the patient's electronic medical record based on the fused data, and output the generated medical record.

[0025] Next, each step of the method of the present invention will be described in detail.

[0026] For step S1, in this embodiment, step S1 collects the multi-dimensional data of the patient, including physical sign data, symptom description data, and geospatial information, and performs preprocessing on the collected data. This step is the basis of the entire medical record generation method, ensuring the accuracy of subsequent data analysis and processing.

[0027] First, the patient's physical sign data is collected in real time through Internet of Things devices. The Internet of Things devices include but are not limited to smart bracelets, smart thermometers, sphygmomanometers, smart ear thermometers, etc. These devices can monitor and collect the patient's physiological sign data in real time. Specifically, the collected physical sign data includes the patient's heart rate, body temperature, blood pressure, respiratory rate, etc. These data can effectively reflect the patient's physiological state and provide basic information for subsequent health analysis and medical record generation.

[0028] During the process of collecting physical sign data, the device transmits the data to the central system through wireless communication technology. During the data transmission process, to ensure the privacy of the patient and the security of the data, the present invention uses encryption technology for data protection. Specifically, the AES encryption algorithm can be used to encrypt the collected physical sign data, and then the data is transmitted to the cloud platform through the TLS protocol.

[0029] Secondly, the patient's symptom description data is uploaded by the patient through a mini-program or voice input. The patient can fill in a form through the mini-program, input the self-reported symptom information, or describe the current physical discomfort through voice input. For example, the patient can describe symptoms such as headache, fatigue, nausea, etc., or report drug side effects, etc. The symptom description data usually includes the patient's self-feedback on their own physical health status and is an indispensable part of medical record generation.

[0030] For the processing of symptom description data, the present invention adopts natural language processing (NLP) technology. Through NLP technology, first, the symptom description input by voice is transcribed into text data, and the text data is processed. The processing process includes word segmentation, part-of-speech tagging, named entity recognition, etc., in order to extract the key information of the symptoms described by the patient and convert this information into a standardized data format that can be used for subsequent analysis.

[0031] Finally, the acquisition of geospatial information is another key part of this step. Geospatial information can be obtained in two ways: one is through the patient's address information, and the other is through a positioning service (such as GPS) to automatically obtain the patient's current location data. Geospatial information reflects the geographical location where the patient is located, and considering the impact of environmental factors (such as climate, air quality, etc.) on the patient's health status, in the process of medical record generation, geospatial information can provide a more comprehensive background for data analysis. Especially in some areas, environmental pollution and climate change may have a greater impact on the patient's health, so this information plays an important role in subsequent analysis.

[0032] The data preprocessing steps are also included in the operations of S1. Sign data and symptom description data are often interfered by noise. Especially, the sensor data may contain errors caused by the accuracy of the device itself, environmental factors, etc. To remove this noise and improve the accuracy of the data, the present invention uses the Kalman filter algorithm to suppress the noise of the collected sign data.

[0033] The Kalman filter algorithm is a recursive algorithm. By combining the current observation value and historical data, it estimates the current system state, thus effectively reducing the errors in the sensor data. Specifically, the Kalman filter algorithm updates each new observation value according to the state equation and observation equation of the system to generate the optimal estimate at the current moment. Through this process, the accuracy of the sensor data is significantly improved, thereby providing higher-quality input data for the subsequent spatio-temporal correlation modeling.

[0034] For the symptom description data, especially the text recognized by OCR, the present invention processes it through NLP. Specifically, first, the OCR technology is used to perform character recognition on the medical reports written or scanned by the patient, converting the text in the image into processable text. Then, NLP technology is used to further process the text recognized by OCR, including grammar correction, vocabulary matching, and semantic analysis, etc. The purpose of this process is to eliminate the misrecognition problems that may occur during the OCR process and ensure the accuracy of the text data.

[0035] For the obtained geospatial information, the present invention stores it in the form of longitude and latitude coordinates or area codes. During the processing of geospatial data, the system will convert the longitude and latitude coordinates into corresponding geographical regions, such as cities or counties, and at the same time, it can also combine external environmental data, such as climate and air quality index, etc., to provide a more comprehensive health assessment.

[0036] Finally, all the sign data, symptom description data, and geospatial information will be encrypted and transmitted to the cloud platform for storage and prepared for further processing in the subsequent steps.

[0037] For step S2, in this embodiment, step S2 models the multi-modal data through a spatio-temporal topology network model and uses a method combining the graph convolutional network (GCN) and the long short-term memory network (LSTM) to process the spatial correlation relationship and time dependence relationship of the data respectively. The specific operation process is as follows: First, the goal of this step is to convert the patient's sign data, symptom description data, and geospatial information into a graph structure for processing in the graph neural network. Each data type (such as sign data, symptom description data, geospatial information) is mapped to a node in the graph, and these nodes are connected by the edges in the graph.

[0038] During the construction of the graph, physical sign data such as heart rate, blood pressure, body temperature, etc. are collected in real time through Internet of Things devices, converted into node features, and mapped into the graph. The symptom description data is the text information extracted through natural language processing technology, which is converted into node features and forms another part of the graph nodes. Geospatial information (such as latitude and longitude data or regional codes) is modeled as the third feature of the graph nodes. In this way, the representation of each patient's health data in the graph includes three important dimensions: physical signs, symptoms, and geospatial information.

[0039] Next, based on the temporal and spatial correlations between the patient's physical sign data, symptom description data, and geospatial information, edges are established between the nodes. Specifically, for nodes that are closer in time, the weight of the edge is higher, while for nodes that are farther apart in time, the weight of the edge is lower; for nodes that are adjacent geographically, the weight of the edge is also higher, and vice versa. The construction of this adjacency matrix depends on the timestamp information and geographical location information of the patient's health data.

[0040] The calculation formula of the adjacency matrix is as follows: Where, represents the degree of association between node and node . This degree of association consists of two parts: one part comes from the time dependence (such as the time series of the patient's health data), and the other part comes from the spatial dependence (such as the patient's geographical location).

[0041] The graph convolutional network (GCN) uses this adjacency matrix to update the features of each node and capture the spatial dependence between the nodes in the graph. The update rule of graph convolution is as follows:

[0042] Where, represents the feature of node at the th layer, and are the weight matrix and bias term of the th layer respectively, is the normalization coefficient, is the activation function. This update rule shows that the features of a node are updated by weighted summing the features of its neighbor nodes.

[0043] The graph convolutional network (GCN) propagates information through layer-by-layer iteration, enabling each node to not only depend on its own features but also fuse the feature information of its neighbor nodes. In this way, the graph convolutional network can effectively capture the spatial correlation of the data, especially the spatial dependence relationship in the patient's health data.

[0044] On this basis, the Long Short-Term Memory network (LSTM) is used to further process the temporal dependencies in the data. LSTM is widely used for time series data modeling, and its advantage lies in being able to remember for a long time and capture the temporal dependencies within a long time range. In this step, LSTM receives the node features output by the graph convolutional network and updates the state of the nodes according to the changes in the patient's health data over time. Specifically, LSTM gradually updates the state of the nodes based on the input information at each time step, thereby capturing the features of each node changing over time.

[0045]

[0046]

[0047]

[0048]

[0049]

[0050] Among them, , , are the input gate, forget gate, and output gate respectively, is the memory cell state, is the output of LSTM, is the input at the current moment. LSTM generates new node representations based on the input health data and the previous state information.

[0051] By combining the graph convolutional network and the Long Short-Term Memory network, this embodiment can capture both the spatial relationships and temporal dependencies in the patient's health data simultaneously. This enables the system to understand the patient's health condition more accurately and provide more comprehensive information for subsequent data processing and medical record generation.

[0052] In this embodiment, the spatio-temporal topology network model combines the graph convolutional network (GCN) and the Long Short-Term Memory network (LSTM). Through the update and propagation of node features, it effectively processes the spatial and temporal correlations in the patient's health data. The implementation of this method not only enhances the correlation processing ability between different modality data but also improves the system's understanding and processing ability of the patient's health data, ensuring that the generated medical record information has higher accuracy and reliability.

[0053] For step S3, in this embodiment, step S3 fuses multi-modal data and uses a graph convolutional network to integratively process spatio-temporal correlation data. Through this step, effective information integration can be carried out between different data sources, and missing data can be complemented to ensure the integrity and high accuracy of the generated electronic medical record. The specific implementation steps are as follows: In this step, first, through the spatio-temporal correlation node features generated in the aforementioned step S2, data from multiple sources (such as vital sign data, symptom description data, geospatial information, etc.) are fused in a multi-modal manner. Each data type (vital sign data, symptom description data, geospatial information, etc.) has its corresponding node representation in the graph convolutional network, and these nodes are interconnected through a graph structure. At this stage, the data features from different modalities are weighted and fused through the graph convolutional network, and the graph convolutional network iteratively updates the node features, thereby effectively integrating the information of each modality.

[0054] Specifically, through the graph convolutional network, the features of a node not only include its own data features but also gradually enhance the expression ability of the fused features through information exchange with neighbor nodes. In the update rule of the graph convolutional network, the features of a node are updated by a weighted sum of the features of its neighbor nodes. As the number of layers increases, the graph convolutional network can gradually capture the deep spatio-temporal correlations between nodes.

[0055] After the graph convolutional network completes the fusion of multi-modal data, the system will update the final representation of the nodes according to the time series and spatial structure to ensure that all data types (such as the patient's vital sign data, symptom description data, and geospatial information) are fully fused in space and time.

[0056] However, in practical applications, data missing is inevitable. Especially, the patient's health data may be missing due to various reasons (such as sensor failures or the patient not uploading information). For this reason, Bayesian inference is introduced in this embodiment to complement the missing data.

[0057] Specifically, the Bayesian inference method estimates the missing data through conditional probability. Suppose there is a set of patient health data with some data missing. Bayesian inference can infer the most likely value of the missing data based on the existing observed data. The basic formula of Bayesian inference is as follows:

[0058] where, is the posterior probability of the missing data under the condition of the known data ; is the probability of the observed data assuming the missing data The likelihood function; is the missing data The prior probability, which represents the possible distribution of the missing data when no data has been observed. is the marginal probability, representing the data The total probability.

[0059] When applying Bayesian inference, first estimate the probability distribution of the missing data through the patient's historical data (such as previous health records, similar data of other patients, etc.). Then, based on this information, complete the missing data. The Bayesian inference method can provide the credibility of each missing data, that is, the completion result of the missing data will carry a reliability score, indicating the inference accuracy of the data.

[0060] Bayesian inference also allows dynamic adjustment of the speculation process according to the patient's health status, medical history, and other known information, so as to complete data in different situations. For each missing physical sign data, symptom description data, or geospatial information, Bayesian inference will generate a possible estimated value and give the speculation result of the missing value according to its probability distribution.

[0061] Through Bayesian inference, this embodiment can provide effective data completion in the case of missing data, making the generated electronic medical record complete and consistent.

[0062] In this embodiment, the multi-modal data is fused through a graph convolutional network, and Bayesian inference is combined to complete the missing data, effectively solving the problem of collaborative processing of different modal data and the impact of missing data on the result accuracy. While processing spatial relationships, the graph convolutional network realizes weighted fusion of multi-dimensional data through an adjacency matrix; Bayesian inference provides a scientific estimation method for missing data, ensuring that the finally generated electronic medical record has high-quality data support and can be provided to doctors for accurate health assessment and subsequent management.

[0063] For step S4, in this embodiment, step S4 generates a complete patient electronic medical record based on the multi-modal data fused and completed in the previous step S3, and outputs the generated medical record to the doctor terminal for follow-up and health management. This step mainly includes the following operations: generating an electronic medical record, formatting the medical record content, and outputting the medical record data.

[0064] First, when generating an electronic medical record, the fused data includes the patient's physical sign data, symptom description data, health status information, and geospatial information. These data will be combined and organized into a complete medical record file in this step. After being processed by GCN and LSTM, each data point already has a comprehensive feature representation of time series and spatial relationships. Based on these fused features, the system will generate an electronic medical record containing the following: the patient's basic personal information, historical health data, real-time monitoring data, symptom descriptions, disease diagnosis and treatment records, health management suggestions, etc.

[0065] Specifically, the physical sign data part includes the patient's physiological indicators (such as heart rate, blood pressure, body temperature, etc.) and their changing trends; the symptom description data part includes the self-reported information input by the patient through the mini-program or voice, including descriptions of physical discomfort, drug side effects, etc.; the health status information includes the patient's medical history, chronic disease status, past disease diagnosis and treatment records; the geospatial information includes the patient's geographical location and the possible impact of the area where they are located on health.

[0066] When generating the electronic medical record, these data will be transformed into standardized medical record content through a formatting process to ensure its adaptation to the format of the electronic medical record system used by doctors. The formatting process includes: Organizing numerical data (such as physical sign data, blood pressure, etc.) into a table form; Presenting text data (such as symptom descriptions, health status descriptions, etc.) in the form of text or reports; Organizing information such as historical health data and treatment records into chronological records to show the patient's health change trends.

[0067] The formatted electronic medical record not only retains the structured representation of the original data, but also makes the medical record content more intuitive and easy to understand through forms such as charts and text.

[0068] Secondly, the generated medical record needs to be sent to the doctor's device through the output module. The form of the electronic medical record output can be adjusted according to actual needs. Common formats include PDF, HTML format, or the format of a dedicated medical system. Through the doctor's terminal, doctors can access the patient's electronic medical record to perform operations such as health assessment, disease diagnosis, and treatment plan formulation. Important technical aspects in the output process include data security and privacy protection. To ensure the security of patient data, in this embodiment, data encryption technology is adopted, and the medical record data is encrypted through the AES encryption algorithm to prevent the medical record data from being illegally accessed or tampered with during transmission.

[0069] The output medical records will provide different views and functions according to the needs of doctors. Doctors can view the detailed information of the medical records and conduct follow-up visits, treatment decisions, or arrangements for further examinations based on the patient's health status. The output of electronic medical records can not only improve the work efficiency of doctors but also help patients with long-term health management and disease prevention.

[0070] In addition, in this embodiment, the output medical records can also be integrated with other information systems in the hospital (such as diagnosis systems, drug management systems, etc.) to achieve data interoperability and sharing. For example, doctors can issue prescriptions for patients based on the generated medical record information, or develop personalized treatment plans for patients through the health trend data in the medical records.

[0071] In this embodiment, through the process of generating electronic medical records based on fused data, the comprehensiveness and accuracy of the medical record content are ensured. At the same time, through formatting and secure output functions, an efficient and reliable electronic medical record viewing and management tool is provided for doctors. The generated medical records not only have high medical value but also provide strong support for subsequent health management, disease follow-up, and treatment, ultimately realizing the effective transmission and management of information and meeting the needs of out-of-hospital patient health management.

[0072] Generally speaking, the present invention generates complete electronic medical records through the fusion and processing of multi-modal data. The method includes obtaining the patient's vital sign data, symptom description data, and geospatial information, and preprocessing the data through technologies such as Kalman filtering and natural language processing. On the basis of data processing, a spatio-temporal topological network model is used to process the spatial correlation and time dependence of the data by combining GCN and LSTM. Further, Bayesian inference is used to complete the missing data to ensure the integrity of the data. Finally, the generated medical records are output after formatting for doctors to use for follow-up visits and health management. The present invention can effectively improve the accuracy and integrity of out-of-hospital patient medical record generation and provide more accurate health assessments and treatment support for doctors.

[0073] To better understand the present invention, the above method will be described in detail below in conjunction with specific embodiments. Embodiment

[0074] This embodiment provides a specific implementation method for generating out-of-hospital patient medical records, mainly including vital sign data collection, symptom description data input, data preprocessing, spatio-temporal topological modeling, data fusion, missing data completion, and finally the generation and output of electronic medical records.

[0075] Step 1: Data collection and preprocessing In this embodiment, the patient collects data through an intelligent health management mini-program and Internet of Things devices. This mini-program can be connected to a variety of intelligent devices (such as smart bracelets, sphygmomanometers, thermometers, etc.) to collect the patient's vital sign data in real time, such as blood pressure, heart rate, body temperature, respiratory rate, etc. Taking the smart bracelet as an example, after the patient wears the bracelet, the device automatically monitors and records data such as heart rate, steps, and body temperature, and the data is encrypted and transmitted to the cloud platform in real time.

[0076] In addition to vital sign data, the patient also inputs their own symptom descriptions through the mini-program. For example, the patient can input "feeling headache and fatigue" through the keyboard or describe discomfort through voice input. The voice input is converted into text by the mini-program and uploaded to the cloud platform. Natural language processing technologies (such as word segmentation, sentiment analysis, etc.) analyze the symptom description data to extract relevant symptom features, such as "headache" and "fatigue".

[0077] In addition, the patient's geospatial information is automatically collected through the positioning function of the mini-program or manually input by the patient (such as the current address, the area where they are located). Through the geospatial information, the system can understand the geographical environment where the patient is located and further analyze the potential impact of environmental factors on health.

[0078] Step 2: Spatiotemporal Topological Network Modeling After data collection and preprocessing, Step 2 performs spatiotemporal topological network modeling on the data. All collected vital sign data, symptom description data, and geospatial information are converted into graph nodes. Specifically, vital sign data (such as blood pressure, heart rate, etc.) is converted into node features, symptom description data extracts keywords (such as "headache", "fatigue", etc.) through natural language processing technology and is converted into text features, and geospatial information (such as longitude and latitude) is also added as node features.

[0079] Next, association edges are established between the nodes according to the time sequence and spatial relationship. Data points that are temporally close will have higher weights, and nodes that are spatially close (such as patients in the same area) will also have higher association weights. The features of each node and the information of adjacent nodes are aggregated and updated through a graph convolutional network (GCN) to capture the spatial and temporal dependencies between the data. The GCN passes and fuses the information of adjacent nodes through its node update rules, thereby enhancing the feature expression ability of the nodes.

[0080] Subsequently, a long short-term memory network (LSTM) further processes the temporal dependencies of the nodes. The LSTM updates the state of each node according to historical data and the feature information at the current moment, captures the dynamic changes of health data over time, and further optimizes the representation of the nodes.

[0081] Step 3: Data Fusion and Missing Data Completion After the spatio-temporal topological modeling is completed, step 3 fuses the multi-modal data. During the fusion process, nodes from different data sources (vital sign data, symptom description data, geospatial information) are fused into comprehensive features through a graph convolutional network, and these fused features can better represent the patient's health status.

[0082] For missing data, Bayesian inference is used in this embodiment for completion. Suppose the vital sign data (such as blood pressure, body temperature) of some patients fails to be collected. The Bayesian inference method will make inferences based on the patient's historical data, data of similar patients, and spatio-temporal correlation, calculate the posterior probability of the missing data, and complete the missing values. Each missing value is accompanied by a credibility score to ensure the rationality and accuracy of the completion.

[0083] Step 4: Generate and output an electronic medical record After the data is fused and the missing data is completed, step 4 generates the patient's electronic medical record. The generated medical record includes the patient's basic information (such as name, age, gender), vital sign data (such as heart rate, blood pressure, etc.), symptom description, health status information (such as chronic diseases, past medical history), and geospatial information (such as the potential impact of the region and environmental factors on health). All information is formatted, including tabular display of numerical data and structured presentation of text descriptions.

[0084] After the electronic medical record is generated, the system encrypts and protects the medical record data through encryption technology, and uses the AES encryption algorithm to ensure the security of the medical record data during transmission. The generated electronic medical record is output to the doctor's terminal device. Doctors can view the medical record information through the terminal, conduct health assessments, diagnoses, and develop personalized treatment plans. At the same time, the output format of the medical record can be PDF, HTML, or the hospital's dedicated electronic medical record format to adapt to the needs of different terminal devices.

[0085] Through the above embodiments, the present invention can effectively integrate the patient's health data from different sources, including vital sign data, symptom description data, and geospatial information, and use the spatio-temporal topological network model for deep fusion. Combining the graph convolutional network and the long short-term memory network, the present invention realizes the spatial and temporal correlation modeling of health data, effectively improving the accuracy and integrity of the data. At the same time, the introduction of Bayesian inference also ensures the reasonable completion of missing data, and the finally generated electronic medical record can provide a comprehensive and accurate health assessment tool for doctors, helping to achieve precise follow-up and health management.

[0086] The out-of-hospital patient medical record generation system described below can be mutually referred to the out-of-hospital patient medical record generation method described above.

[0087] Please refer to the appendix Figure 2 , the present invention also provides an out-of-hospital patient medical record generation system, including: A data acquisition module 10, configured to obtain the physical sign data of a patient through an Internet of Things device, obtain the symptom description data of the patient through a mini-program or voice input method, and obtain the geospatial information of the patient; A data processing module 20, configured to preprocess the physical sign data, symptom description data, and geospatial information, including using Kalman filtering to suppress noise in the sensor data, and using OCR and natural language processing to correct the symptom description data; A spatio-temporal correlation modeling module 30, configured to construct the physical sign data, symptom description data, and geospatial information into graph nodes based on a spatio-temporal topology network model, process the spatial association relationship between the nodes through a graph convolutional network, and process the time dependence relationship of the nodes using a long short-term memory network; A data fusion and completion module 40, configured to fuse multi-modal data, integrate the spatio-temporal correlation data through a graph convolutional network, and use Bayesian inference to complete the missing data; A medical record generation module 50, configured to generate an electronic medical record of the patient based on the fused data, and output the generated medical record to a doctor's terminal for the doctor to use for follow-up and health management.

[0088] The system of this embodiment can be used to execute the above method embodiment, and its principle and technical effect are similar, which will not be elaborated here.

[0089] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for generating medical records of out-of-hospital patients, characterized in that: The following steps are involved: Obtaining the patient's vital sign data, the patient's symptom description data, and the patient's geographic spatial information, and performing pre-processing; Based on the spatiotemporal topological network model, the patient's physical sign data, symptom description data and geographic space information are constructed as graph nodes, and the spatial association relationship between nodes is processed by the graph convolutional network, and the time dependency relationship of nodes is processed by the long short-term memory network; Fuse multimodal data, integrate spatiotemporal correlation data through graph convolutional networks, and use Bayesian inference to complete missing data; Generate the patient's electronic medical record based on the fused data and output the generated medical record.

2. The method for generating medical records of outpatient patients according to claim 1, characterized in that: The obtaining of the patient's vital sign data includes obtaining the patient's physiological sign data through an Internet of Things device; the obtaining of the patient's symptom description data includes text or voice data uploaded by the patient through a mini-program or voice input; the obtaining of the patient's geographic spatial information includes obtaining the patient's current location data through the patient's address information or positioning service.

3. The method for generating medical records of outpatient patients according to claim 1, characterized in that: The step of constructing the patient's physical sign data, symptom description data and geographic space information into graph nodes based on the spatiotemporal topological network model includes: Construct corresponding node representations for each type of data, and use the node representations as basic elements in the graph; Establish edges between nodes according to the time sequence and spatial relationship to form an adjacency matrix of the graph, wherein the adjacency matrix is ​​used to represent the relationship between data nodes; A graph convolutional network is used to process the node information in the graph, and the spatial associations between the nodes are transmitted and updated through the adjacency matrix of the graph.

4. The method for generating medical records of outpatient patients according to claim 3, characterized in that: The step of processing the spatial association relationship between nodes by using a graph convolutional network comprises: Use the node update rule of the graph convolutional network to perform weighted summation of the features of each node, combined with the weight information in the adjacency matrix; The node features are processed hierarchically through a multi-layer graph convolutional network, which gradually transfers and aggregates information from neighboring nodes to capture the spatial dependency of the data. Nonlinear activation is performed on the output results of each layer of graph convolutional network.

5. The method for generating medical records of outpatient patients according to claim 4, characterized in that: The step of processing the time dependency of nodes using the long short-term memory network comprises: The node features output by the graph convolutional network are used as the input data of the long short-term memory network to capture the evolution law of the data in the time dimension; Use long short-term memory networks to process node state updates at different time steps, and capture the temporal dependencies of nodes through cyclic calculations; The time-dependent features of each node are updated to generate a node representation with time information.

6. The method for generating medical records of outpatient patients according to claim 1, characterized in that: The step of fusing the multimodal data comprises: Connect the node features processed by the graph convolutional network with the node features from different data sources to form multimodal fusion features; Applying a further graph convolutional network layer to the multimodal fusion features; In the multi-layer processing of graph convolutional networks, different modal data are fused according to the spatial correlation and temporal dependency of nodes.

7. A method for generating medical records of outpatient patients according to claim 6, characterized in that: The step of integrating the spatiotemporal correlation data comprises: Perform weighted summation of node features from different time steps and spatial positions, and propagate and synthesize information between nodes through the adjacency matrix in the graph convolutional network; During the integration process, nonlinear activation functions are used to process the weighted features of the nodes to further enhance the feature representation after data fusion; By utilizing the hierarchical structure of graph convolutional networks, comprehensive information of different time steps and spatial positions is gradually extracted at each layer.

8. The method for generating medical records of outpatient patients according to claim 7, characterized in that: The step of using Bayesian inference to complete missing data includes: Make probability inferences about missing data based on existing data, and calculate the posterior probability distribution of missing data through the Bayesian inference model; Use patients’ historical health data, similar patients’ data, and spatial and temporal association information to fill in missing data; Assign credibility weights to the completed data, and use the estimated results of the missing data as the final output.

9. The method for generating medical records of outpatient patients according to claim 1, characterized in that: The step of generating the patient's electronic medical record based on the fused data and outputting the generated medical record comprises: Generate a comprehensive electronic medical record including patient vital signs data, symptom description data, health status information and geospatial information based on the fused multimodal data; Formatting the generated electronic medical record to make it suitable for doctors' review and subsequent follow-up, wherein the formatting process includes organizing the data into structured medical record information; The formatted electronic medical records are output to the doctor's terminal so that the doctor can perform subsequent diagnosis, health management and treatment tracking based on the generated medical records.

10. A system for generating medical records of out-of-hospital patients, used to execute the method for generating medical records of out-of-hospital patients according to any one of claims 1 to 9, characterized in that: include: The data collection module is used to obtain the patient's vital sign data through the IoT device, obtain the patient's symptom description data through the mini-program or voice input, and obtain the patient's geographic spatial information; A data processing module, used for preprocessing the physical sign data, symptom description data and geographic space information, including Kalman filtering to suppress noise on sensor data, and OCR and natural language processing to correct symptom description data; A spatiotemporal correlation modeling module, which is used to construct the physical sign data, symptom description data and geographic space information into graph nodes based on a spatiotemporal topological network model, and process the spatial correlation relationship between nodes through a graph convolutional network, and process the temporal dependency relationship of nodes using a long short-term memory network; Data fusion and completion module, which is used to fuse multimodal data, integrate spatiotemporal correlation data through graph convolutional networks, and use Bayesian inference to complete missing data; The medical record generation module is used to generate the patient's electronic medical record based on the fused data, and output the generated medical record to the doctor's terminal for the doctor to use for follow-up and health management.

Images