Medical data management method, system and storage medium based on artificial intelligence
By preprocessing and standardizing multi-source medical data, virtual medical twins are built, and the problems of data integration and personalized medical care in the existing system are solved, efficient resource allocation and decision-making support are achieved, and the quality and efficiency of medical services are improved. , and ensures data security.
Patent Information
- Application Number
- CN202411077418.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-07
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-08-07
AI Technical Summary
The existing medical data management system lacks effective data integration and sharing mechanisms, making it difficult to achieve personalized medical care, resource allocation is not dynamic, decision-making support tools lack intelligence, data security and privacy protection challenges, resulting in low quality of medical services and resource utilization efficiency. .
By preprocessing and standardizing multi-source medical data, building virtual medical twins, performing intelligent classification and annotation, data mining and feature extraction, generating personalized medical solutions and resource optimization strategies, and using multiple security measures to protect data privacy.
It has achieved effective integration of heterogeneous data, improved the accuracy of medical decision-making and resource utilization efficiency, met personalized medical needs, dynamically adjusted resource allocation, improved the quality of medical services and patient satisfaction, and reduced medical costs.
Smart Images

Figure CN118609743B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of data processing technology, and in particular to an artificial intelligence-based medical data management method, system and storage medium. Background Art
[0002] Existing medical data management systems mainly include electronic medical record systems, hospital information systems, laboratory information systems, and image archiving and communication systems. These systems can realize the electronic recording of patient information, basic medical data analysis, and simple hospital resource management. In recent years, with the development of artificial intelligence and Internet of Things technologies, some advanced medical institutions have begun to try to apply these new technologies to the medical field. For example, using deep learning algorithms for medical image-assisted diagnosis, using natural language processing technology to analyze electronic medical records, and using wearable devices and smart sensors to monitor patient physiological indicators in real time. In addition, some medical institutions have also begun to explore big data analysis technology to provide support for clinical decision-making by mining massive medical data.
[0003] However, there are still many shortcomings in existing technologies. Most systems are still relatively independent and lack effective data integration and sharing mechanisms, resulting in serious information island problems and difficulty in fully utilizing multi-source heterogeneous medical data. Secondly, existing systems generally lack dynamic adaptability and find it difficult to quickly adjust resource allocation and service processes according to real-time conditions. Thirdly, in terms of personalized medicine, existing systems can often only provide general recommendations based on statistics, making it difficult to achieve true precision medicine. Fourthly, hospital resource optimization management still relies mainly on manual experience and lacks intelligent and automated decision support tools. Finally, existing systems also face challenges in data security and privacy protection, especially in scenarios of data sharing and cross-institutional collaboration. These deficiencies have seriously restricted the improvement of medical service quality and the efficient use of medical resources. Summary of the invention
[0004] The present application provides an artificial intelligence-based medical data management method, system and storage medium for improving the efficiency and accuracy of artificial intelligence-based medical data management.
[0005] In the first aspect, the present application provides an artificial intelligence-based medical data management method, which includes: preprocessing multi-source medical data collected from electronic medical record systems, medical devices and wearable devices to obtain a standardized medical data set; intelligently classifying and labeling the standardized medical data set to obtain structured medical knowledge data; constructing a digital twin based on the structured medical knowledge data to obtain a virtual medical twin; performing data mining and feature extraction on the virtual medical twin to obtain medical analysis data; based on the medical analysis data, performing a comprehensive analysis of real-time collected patient individual data and hospital operation data to obtain personalized medical plans and resource optimization strategies; performing data optimization on the personalized medical plans and resource optimization strategies through pre-collected patient flow data to obtain a target management strategy, wherein the target management strategy includes: a resource allocation sub-strategy, a patient management sub-strategy, and a service optimization sub-strategy.
[0006] In combination with the first aspect, in a first implementation method of the first aspect of the present application, the multi-source medical data collected from the electronic medical record system, medical equipment and wearable devices are pre-processed to obtain a standardized medical data set, including: - performing data cleaning on the multi-source medical data to obtain initial cleaned data, and performing format consistency processing on the initial cleaned data to obtain unified format data; performing data desensitization processing on the unified format data to obtain privacy protection data, and performing time series alignment on the privacy protection data to obtain time consistent data; performing interpolation processing on missing values in the time consistent data to obtain a complete data set, and performing outlier detection and processing on the complete data set to obtain anomaly corrected data; performing data normalization processing on the anomaly corrected data to obtain normalized data, and performing dimensionality reduction processing on the normalized data through a principal component analysis algorithm to obtain reduced dimensionality data; performing data segmentation and encoding on the reduced dimensionality data to obtain encoded data; performing metadata annotation and index creation on the encoded data to obtain the standardized medical data set.
[0007] In combination with the first aspect, in a second implementation method of the first aspect of the present application, the intelligent classification and labeling of the standardized medical data set to obtain structured medical knowledge data includes: performing multi-dimensional feature extraction on the standardized medical data set to obtain a medical feature vector set, and performing unsupervised clustering analysis on the medical feature vector set to obtain an initial medical data category; semantically mapping the initial medical data category based on a preset medical ontology to obtain semantically enhanced medical data, and extracting multimodal relationships on the semantically enhanced medical data to obtain a medical entity association graph; performing medical terminology standardization on the medical entity association graph to obtain a standardized medical vocabulary network, and constructing a medical event time series chain based on the standardized medical vocabulary network to obtain medical time series association data; performing pattern recognition and mining on the medical time series association data to obtain a medical management model library; performing simulation evaluation on the medical decision-making process based on the medical management model library to obtain a decision-making influence indicator set; performing data fusion on the decision-making influence indicator set and the medical time series association data to obtain a fused medical knowledge graph; and performing structured processing and indexing on the fused medical knowledge graph to obtain the structured medical knowledge data.
[0008] In combination with the first aspect, in a third implementation method of the first aspect of the present application, the digital twin is constructed based on the structured medical knowledge data to obtain a virtual medical twin, including: performing semantic analysis on the structured medical knowledge data to obtain a medical entity relationship network, and constructing a three-dimensional space model based on the medical entity relationship network to obtain an initial virtual medical environment; dynamically mapping parameters of the initial virtual medical environment to obtain an interactive medical scene; accessing the interactive medical scene with real-time data streams to obtain a data-driven virtual medical model; establishing multi-dimensional evaluation indicators based on the data-driven virtual medical model to obtain a medical twin evaluation indicator set, and performing simulation verification on the medical twin evaluation indicator set to obtain a twin operation rule set; digitally mapping the medical process based on the twin operation rule set to obtain virtual medical operation data; and performing multi-scenario testing and optimization on the virtual medical operation data to obtain the virtual medical twin.
[0009] In combination with the first aspect, in a fourth implementation method of the first aspect of the present application, the data mining and feature extraction of the virtual medical twin to obtain medical analysis data includes: performing spatiotemporal data decomposition on the virtual medical twin to obtain a multi-dimensional medical data stream, and performing data cleaning and standardization processing on the multi-dimensional medical data stream to obtain a normalized medical data set; performing time series correlation analysis on the normalized medical data set to obtain a medical event association chain; constructing a medical decision tree based on the medical event association chain to obtain an initial medical decision rule set; cross-validating and pruning the initial medical decision rule set to obtain a streamlined medical decision rule; mining medical resource utilization patterns based on the streamlined medical decision rules to obtain a resource allocation feature graph; performing cluster analysis on the resource allocation feature graph to obtain a medical resource utilization efficiency evaluation index; constructing a multi-objective optimization function based on the medical resource utilization efficiency evaluation index to obtain a resource allocation strategy set; performing sensitivity analysis on the resource allocation strategy set to obtain a key influencing factor set; and performing parameter tuning on the virtual medical twin based on the key influencing factor set to obtain the medical analysis data.
[0010] In combination with the first aspect, in a fifth implementation of the first aspect of the present application, based on the medical analysis data, a comprehensive analysis is performed on the real-time collected patient individual data and hospital operation data to obtain a personalized medical plan and a resource optimization strategy, including: multi-dimensional decomposition of the medical analysis data to obtain a patient feature vector and a hospital resource status matrix; cluster analysis of the patient feature vector to obtain patient grouping data, and construct a personalized risk assessment index based on the patient grouping data to obtain a patient risk distribution map; dynamic load analysis of the hospital resource status matrix to obtain a resource utilization curve, and peak prediction based on the resource utilization curve to obtain a resource demand prediction table; cross-matching the patient risk distribution map and the resource demand prediction table to obtain an initial resource allocation plan, and constructing a multi-constraint optimization rule set based on the initial resource allocation plan to obtain a resource scheduling strategy; discrete event simulation of the resource scheduling strategy to obtain a strategy evaluation result, and iterative optimization of the resource scheduling strategy based on the strategy evaluation result to obtain an optimized resource allocation plan; correlation analysis of the optimized resource allocation plan with the patient individual data to obtain the personalized medical plan and resource optimization strategy.
[0011] In combination with the first aspect, in a sixth implementation method of the first aspect of the present application, the personalized medical plan and resource optimization strategy are optimized through pre-collected patient flow data to obtain a target management strategy, wherein the target management strategy includes: a resource allocation sub-strategy, a patient management sub-strategy, and a service optimization sub-strategy, including: performing a wavelet transform on the pre-collected patient flow data to obtain a multi-scale time series decomposition result; performing an autoregressive analysis on the multi-scale time series decomposition result to obtain a short-term prediction value of patient flow; calculating the resource utilization rate of each department based on the short-term prediction value of patient flow to obtain a dynamic resource load distribution map, and performing a hotspot analysis on the dynamic resource load distribution map to obtain a resource bottleneck identification report; generating a resource bottleneck identification report based on the resource bottleneck identification report The method comprises the following steps: forming a resource dynamic allocation instruction set to obtain a resource allocation sub-strategy; performing queue theory analysis on the pre-collected patient flow data to obtain a patient waiting time distribution function, and constructing a triage priority rule based on the patient waiting time distribution function to obtain an intelligent triage plan; combining the intelligent triage plan with the patient's individual characteristic data to obtain a personalized treatment path; constructing a patient guidance mechanism based on the personalized treatment path to obtain a patient management sub-strategy; collaboratively optimizing the resource allocation sub-strategy and the patient management sub-strategy to obtain a service process reconstruction plan, and constructing a service quality monitoring indicator based on the service process reconstruction plan to obtain a service optimization sub-strategy; integrating the resource allocation sub-strategy, the patient management sub-strategy and the service optimization sub-strategy to obtain the target management strategy.
[0012] In a second aspect, the present application provides an artificial intelligence-based medical data management system, the artificial intelligence-based medical data management system comprising:
[0013] A processing module, used to pre-process multi-source medical data collected from electronic medical record systems, medical devices and wearable devices to obtain a standardized medical data set;
[0014] The classification module is used to intelligently classify and label standardized medical data sets to obtain structured medical knowledge data;
[0015] A construction module, used to construct a digital twin based on the structured medical knowledge data to obtain a virtual medical twin;
[0016] An extraction module, used to perform data mining and feature extraction on the virtual medical twin to obtain medical analysis data;
[0017] An analysis module is used to conduct a comprehensive analysis of the real-time collected patient individual data and hospital operation data based on the medical analysis data to obtain personalized medical solutions and resource optimization strategies;
[0018] The optimization module is used to optimize the personalized medical plan and resource optimization strategy through the pre-collected patient flow data to obtain the target management strategy, wherein the target management strategy includes: resource allocation sub-strategy, patient management sub-strategy and service optimization sub-strategy.
[0019] The third aspect of the present application provides a computer-readable storage medium, which stores instructions. When the computer-readable storage medium is run on a computer, it enables the computer to execute the above-mentioned artificial intelligence-based medical data management method.
[0020] In the technical solution provided by this application, by preprocessing and standardizing multi-source medical data, the effective integration of heterogeneous data is achieved, the problem of information islands in traditional systems is solved, and a solid data foundation is laid for subsequent intelligent analysis. Secondly, by intelligently classifying and labeling standardized medical data sets, structured medical knowledge data is obtained, which not only improves the comprehensibility and availability of data, but also provides more comprehensive and accurate information support for medical decision-making. The virtual medical twin constructed based on structured medical knowledge data provides a comprehensive and dynamic digital mapping for medical management, enabling medical institutions to perform various simulations and predictions in a virtual environment, greatly reducing decision-making risks. Data mining and feature extraction of virtual medical twins are performed, and the medical analysis data obtained provides a scientific basis for the formulation of personalized medical plans and the generation of resource optimization strategies. This decision support based on deep data analysis significantly improves the accuracy of medical services and the efficiency of resource utilization. By comprehensively analyzing the real-time collected patient individual data and hospital operation data, this method can generate truly personalized medical plans, meet the needs of precision medicine, and can also dynamically adjust resource allocation according to real-time conditions, thereby improving the operational efficiency of hospitals. In particular, personalized medical plans and resource optimization strategies are further optimized through pre-collected patient flow data. The resulting target management strategy covers multiple aspects such as resource allocation, patient management, and service optimization, providing a comprehensive intelligent solution for hospital management. This dynamic and intelligent management method can effectively cope with the complexity and variability of the medical environment, and significantly improve the quality of medical services and patient satisfaction. In addition, the application of this method can also reduce medical costs, reduce the waste of medical resources, and create greater economic benefits for medical institutions. This method involves the processing of a large amount of sensitive medical data. During the construction and implementation process, data security and privacy protection issues are fully considered. Multiple security measures such as data desensitization and access control are adopted, which can effectively protect patient privacy and improve the efficiency and accuracy of medical data management based on artificial intelligence. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.
[0022] Figure 1 This is a schematic diagram of an embodiment of a medical data management method based on artificial intelligence in an embodiment of the present application;
[0023] Figure 2 This is a schematic diagram of an embodiment of an artificial intelligence-based medical data management system in an embodiment of the present application. DETAILED DESCRIPTION
[0024] Embodiments of the present application provide a medical data management method, system and storage medium based on artificial intelligence. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0025] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiments of the present application, an embodiment of the medical data management method based on artificial intelligence includes:
[0026] Step S101, preprocessing multi-source medical data collected from electronic medical record systems, medical devices and wearable devices to obtain a standardized medical data set;
[0027] It is understandable that the execution subject of the present application can be a medical data management system based on artificial intelligence, or a terminal or a server, which is not limited here. The present application embodiment is described by taking the server as the execution subject as an example.
[0028] Specifically, various types of medical data are obtained from different sources. The electronic medical record system provides text data such as basic patient information, diagnosis records, and treatment plans; medical equipment such as CT and MRI provide high-resolution medical imaging data; and wearable devices provide patients' real-time physiological indicators, such as heart rate, blood pressure, blood sugar and other continuous data. The diversity of these data types and formats brings challenges to subsequent processing.
[0029] The first step in data preprocessing is data cleaning, which aims to remove noise and outliers. For example, for heart rate data collected by wearable devices, abnormally high or low values may occur due to loose equipment, and these data need to be identified and processed. Data cleaning also includes processing missing values, which can be solved by interpolation or mean value filling. The next step is to unify the data format and convert data from different sources into a consistent format. For example, the data format output by medical devices from different manufacturers is unified into the standard DICOM format. Data desensitization is a key step in protecting patient privacy, which involves encrypting or replacing sensitive information such as name, ID number, etc. Time series alignment is an important step in processing multi-source data, ensuring that data from different devices are consistent in the time dimension. For example, align the heart rate data recorded every minute by the wearable device with the body temperature data recorded every day in the electronic medical record on the time axis. The aligned data may still have missing values and need to be interpolated. Common methods include linear interpolation, spline interpolation, etc. Outlier detection and processing are important steps to ensure data quality. Statistical methods or machine learning algorithms can be used to identify and process outliers. Data normalization is to convert data of different scales to the same scale, which is helpful for subsequent data analysis and model training. Principal component analysis (PCA) is a commonly used dimensionality reduction method that can reduce data dimensions while retaining the main information of the data. For high-dimensional medical data, such as gene expression data, PCA can effectively extract key features. Data segmentation and coding are to better organize and manage data. Data can be segmented according to factors such as time and disease type, and each data segment can be assigned a unique code.
[0030] Finally, metadata annotation and index creation are to facilitate subsequent data retrieval and analysis. Metadata includes information such as the source of the data, collection time, and data type, while indexing can speed up data query. The original multi-source medical data is converted into a standardized medical data set, laying the foundation for subsequent intelligent classification, annotation, and analysis. This standardized data set is not only consistent in format, but also aligned in time dimension, while ensuring data quality and privacy security.
[0031] For example, a hospital collected data from a patient for a week, including temperature data recorded daily in the electronic medical record, blood pressure data recorded hourly by medical equipment, and heart rate data recorded every minute by wearable devices. First, the data was cleaned to remove outliers caused by equipment failure, such as obvious erroneous values such as 0 or 300 in the heart rate data. Then, the timestamps of all data were unified to the hourly level, and the average was taken for more fine-grained data (such as heart rate). Next, the patient's personal information was desensitized, and the name and ID number were replaced with a unique ID. After time series alignment, it was found that there were missing values in the blood pressure data, and these missing values were filled by linear interpolation. Several abnormal temperature data points were detected by the box plot method, and these data points were adjusted to the normal range after confirmation by the doctor. Finally, the temperature, blood pressure, and heart rate data were normalized to the range of 0-1, and the original multidimensional data (including systolic blood pressure, diastolic blood pressure, etc.) were reduced to several main dimensions by the PCA algorithm. Finally, these processed data were given a unified encoding format and added with corresponding metadata annotations to form a standardized medical data set. This standardized dataset not only ensures the quality and consistency of the data, but also greatly improves the efficiency and accuracy of subsequent analysis.
[0032] Step S102: intelligently classify and label the standardized medical data set to obtain structured medical knowledge data;
[0033] Specifically, multi-dimensional feature extraction is performed on the standardized medical data set. This step uses various feature extraction algorithms to analyze the characteristics of the data from different angles. For example, for time series data, its statistical features (such as mean, variance), frequency domain features (such as Fourier transform coefficients) and time domain features (such as trends, periodicity), etc. can be extracted. For text data, natural language processing technology can be used to extract features such as keywords and syntactic structures. Unsupervised clustering analysis is performed on the medical feature vector set. Commonly used algorithms include K-means and hierarchical clustering. The purpose of this step is to automatically classify similar data to form initial medical data categories. For example, patients with similar symptom patterns can be clustered together, or similar diagnostic reports can be classified into one category. The advantage of unsupervised learning is that it can discover potential patterns in data without pre-defining categories.
[0034] In order to further improve the accuracy and semantic meaning of classification, semantic mapping of initial medical data categories is performed based on the preset medical ontology. Medical ontology is a formalized knowledge representation that contains concepts, relationships, and rules in the medical field. By mapping the initial categories to the concepts in the medical ontology, semantically enhanced medical data can be obtained, which not only have statistical similarities but also have clear medical semantics. Multimodal relationship extraction of semantically enhanced medical data is the next key step. Multimodality here refers to different types of medical data, such as text, images, and numerical values. The purpose of relationship extraction is to identify the relationship between different entities, such as the relationship between symptoms and diseases, the relationship between drugs and treatment effects, etc. The result of this step is the formation of a medical entity association graph, which is a complex network structure that shows various associations between medical concepts.
[0035] In order to ensure the standardization and consistency of data, it is necessary to standardize the medical terminology of the medical entity association map. This step unifies medical terms from different sources and with different expressions into a standardized medical vocabulary system, such as the International Classification of Diseases (ICD) or Systematic Medical Terminology (SNOMED CT). The result of the standardization process is a standardized medical vocabulary network, which not only contains standardized medical terms but also retains the relationship between them. Based on the standardized medical vocabulary network, constructing a medical event time series chain is the next important step. This step aims to capture the time sequence and causal relationship of medical events. For example, a complete time series chain of a patient from the first visit, diagnosis, treatment to recovery can be constructed, or a time series pattern of a disease from early symptoms to complications can be constructed. These time series chains form medical time series association data, which provide an important basis for subsequent pattern recognition and predictive analysis.
[0036] Pattern recognition and mining of medical time series correlation data is an in-depth analysis process, the purpose of which is to discover the laws and patterns hidden in the data. This can include common disease development patterns, treatment response patterns, drug interaction patterns, etc. Through this step, a medical management model library is obtained, which is a knowledge base containing various medical models. Simulating and evaluating the medical decision-making process based on the medical management model library is an important part of applying the extracted knowledge to actual decision-making. This step can simulate different medical decision-making scenarios and evaluate the potential impact and effect of the decision. Through this simulation and evaluation, a set of decision impact indicators is obtained, which quantifies the various impacts that different decisions may bring.
[0037] The purpose of data fusion of decision-influencing indicator sets and medical time series association data is to combine static decision evaluation results with dynamic time series data to form a more comprehensive knowledge system. This fusion process obtains a fused medical knowledge graph, which is a complex knowledge network that integrates time series relationships, decision influences, and medical knowledge. The fused medical knowledge graph is structured and indexed to obtain the final structured medical knowledge data. This step converts the knowledge graph into a structured data format that is easy to store, retrieve, and apply, and establishes corresponding indexes to improve data access efficiency.
[0038] For example, a hospital collected 5-year follow-up data of 1,000 diabetic patients, including regular blood sugar test results, medication records, complications, etc. First, multi-dimensional feature extraction was performed on these data, such as calculating the average blood sugar level, blood sugar fluctuation range, medication frequency, etc. of each patient, to form a feature vector for each patient. Then, the K-means algorithm was used to cluster these feature vectors, setting K=5 to divide the patients into 5 initial categories. Next, based on the preset diabetes medical ontology, these 5 categories were semantically mapped to obtain medically meaningful classifications such as "good blood sugar control type" and "high risk of complications type". Furthermore, by analyzing the patient's medication records and blood sugar changes, the drug-effect relationship was extracted and a medical entity association map was constructed. In the process of terminology standardization, different expressions used by different doctors (such as "diabetic nephropathy" and "diabetic nephropathy") were unified into standard terms. Based on the standardized data, a time series chain of disease development for each patient was constructed, including key time points such as first diagnosis, occurrence of complications, and treatment adjustment. By mining patterns in these time series chains, we discovered several typical disease progression patterns, such as "rapid complication development" and "stable control". Based on these patterns, we simulated and evaluated different treatment options and obtained the expected effect indicators of each option for different types of patients. Finally, we integrated these patterns, evaluation results, and the original time series data to form a comprehensive diabetes knowledge graph, which was converted into a structured data format for subsequent query and analysis. This structured knowledge data not only contains the original medical data, but also incorporates the patterns, relationships, and decision support information extracted from the data.
[0039] Step S103: constructing a digital twin based on the structured medical knowledge data to obtain a virtual medical twin;
[0040] Specifically, semantic analysis is performed on structured medical knowledge data, and natural language processing and knowledge graph technology are used to deeply understand the concepts, entities and relationships in the data. Through this analysis, a medical entity relationship network is obtained, which clearly shows the complex associations between various medical concepts. Based on the medical entity relationship network, building a three-dimensional space model is the next key step. Abstract medical knowledge is converted into a visual three-dimensional structure, including the physical layout of the hospital, equipment distribution, patient flow path, etc. This process involves computer graphics and spatial modeling technology, and ultimately obtains the initial virtual medical environment, which is a static digital representation of the medical system.
[0041] Dynamic parameter mapping of the initial virtual medical environment is the key to making the digital twin "alive". Real-time data, historical data, and predicted data are mapped into the virtual environment so that it can reflect the dynamic changes of the medical system. For example, information such as patient admission, transfer, and discharge is updated in real time into the virtual environment, or data such as equipment usage status and medical staff workload are dynamically displayed. This process results in an interactive medical scene that not only displays the current status, but also simulates system responses in different situations.
[0042] Real-time data stream access to interactive medical scenarios is the key to keeping the digital twin synchronized with the actual medical system. A data bridge is established between the actual medical system and the virtual environment to ensure that the virtual environment can reflect changes in the actual situation in real time. This includes access to data streams of various medical devices, real-time data from patient monitoring systems, updates to hospital information systems, etc. Through this real-time data access, a data-driven virtual medical model is obtained, which can accurately reflect the real-time status and dynamic changes of the medical system. The establishment of multi-dimensional evaluation indicators based on the data-driven virtual medical model is to comprehensively evaluate the performance and efficiency of the medical system. These indicators include but are not limited to patient waiting time, bed turnover rate, medical equipment utilization, and medical staff workload. Through these indicators, a set of medical twin evaluation indicators is obtained, which provides a quantitative basis for the management and optimization of the medical system.
[0043] Simulation verification of the medical twin evaluation index set is an important step to ensure that the digital twin can accurately reflect and predict the behavior of the actual medical system. This process continuously adjusts and optimizes the parameters and algorithms of the digital twin by comparing the prediction results in the virtual environment with the performance of the actual medical system. Through repeated verification and adjustment, the twin operation rule set is obtained, which defines how the virtual medical twin operates and responds to various situations. Digital mapping of medical processes based on the twin operation rule set is to transform abstract rules into specific virtual operation processes. Various processes in the medical system are simulated, such as patient consultation process, medical resource scheduling process, emergency response process, etc. Through this mapping, virtual medical operation data is obtained, which reflects the operating status and performance of the medical system under different circumstances. Finally, multi-scenario testing and optimization of virtual medical operation data is a key step to improve the virtual medical twin. This process simulates various possible medical scenarios, such as daily operations, peak management, emergency situation handling, etc., and continuously optimizes the parameters and algorithms of the twin according to the test results. Through this repeated testing and optimization, a highly accurate and reliable virtual medical twin is finally obtained.
[0044] For example, a tertiary hospital decided to build a digital twin to optimize the operation of its emergency department. First, semantic analysis was performed on the structured medical knowledge data of the emergency department to identify key entities such as "patient", "doctor", "nurse", "examination room", "examination equipment", etc., as well as the relationships between them such as "patient-waiting-examination room" and "doctor-use-examination equipment", forming a medical entity relationship network. Based on this network, a three-dimensional spatial model of the emergency department was constructed, including the physical layout of the waiting area, examination room, emergency room, etc., as well as the distribution of medical staff and equipment. Then, dynamic parameters such as patient waiting time and doctor workload were mapped to this model to create an interactive virtual emergency scene. By accessing real-time data streams, such as the patient arrival rate updated every 15 minutes and the average frequency of examination equipment use of 10 times per hour, the virtual scene can reflect the actual situation in real time. Based on this model, a set of evaluation indicators including average waiting time, doctor utilization rate, bed turnover rate, etc. was established. By comparing the prediction results in the virtual environment (such as the predicted average waiting time during peak hours of 2 hours) with the actual data (the actual average waiting time during peak hours is 1.8 hours), the twin's prediction algorithm is continuously optimized. Based on the optimized rules, the entire emergency treatment process is digitally mapped, including the entire process from patient arrival, triage, waiting, treatment to departure. Finally, by simulating the operating conditions under different situations (such as ordinary working days, holiday peaks, public health emergencies, etc.), the virtual twin is continuously adjusted and optimized.
[0045] Step S104: performing data mining and feature extraction on the virtual medical twin to obtain medical analysis data;
[0046] Specifically, the virtual medical twin is subjected to spatiotemporal data decomposition, and the complex twin data is split according to the time and space dimensions to obtain a multi-dimensional medical data stream. The time dimension can include different granularities such as hours, days, weeks, and months, while the space dimension can include different departments, wards, equipment, etc. This decomposition enables subsequent analysis to be performed at different spatiotemporal scales, improving the flexibility and accuracy of the analysis. Data cleaning and standardization of multi-dimensional medical data streams are aimed at improving data quality and ensuring the accuracy of subsequent analysis. Data cleaning includes removing outliers, processing missing data, etc., while standardization unifies data of different dimensions to the same scale for easy comparison and analysis. Through this process, a standardized medical data set is obtained, which is the basis for subsequent analysis.
[0047] Time series correlation analysis of normalized medical data sets is a key step in mining time patterns. Time series analysis techniques, such as autocorrelation analysis and cross-correlation analysis, are used to explore the intrinsic connections of data in the time dimension. For example, the relationship between patient flow and medical resource utilization at different time points is analyzed, or the time delay effect between therapeutic intervention and patient recovery speed is explored. Through this analysis, a medical event association chain is obtained, which shows the mutual influence and causal relationship of medical events in time. Constructing a medical decision tree based on the medical event association chain is an important step in converting the results of time series analysis into actionable decisions. The decision tree is an intuitive decision support tool that guides decision makers to choose the best action path based on different conditions and situations. In this process, machine learning algorithms, such as ID3 or C4.5 algorithms, are used to extract decision rules from the medical event association chain to form a preliminary medical decision rule set. This rule set contains a series of "if-then" decision logic, which provides a preliminary guiding framework for medical decision-making.
[0048] Cross-validation and pruning optimization of the preliminary medical decision rule set are the key to improving the accuracy and generalization ability of the decision tree. Cross-validation evaluates the stability and accuracy of the model by dividing the data set into multiple subsets, repeatedly training and testing the model. Pruning optimization reduces the risk of overfitting and improves the performance of the model on new data by removing unnecessary branches in the decision tree. Through this process, streamlined medical decision rules are obtained, which are more concise, efficient, and have stronger generalization ability. Mining medical resource utilization patterns based on streamlined medical decision rules is an important part of optimizing resource allocation. Using data mining techniques, such as association rule mining and sequence pattern mining, potential patterns of resource utilization are discovered from a large amount of medical data. For example, resource waste or bottlenecks in certain treatment processes are discovered, or efficient resource allocation schemes are identified. Through this mining, a resource allocation feature map is obtained, which intuitively shows the various characteristics and patterns of medical resource utilization.
[0049] Cluster analysis of resource allocation feature graphs is a step to further summarize and generalize resource utilization patterns. Cluster analysis uses algorithms such as K-means or hierarchical clustering to group similar resource allocation patterns into one category, thereby obtaining medical resource utilization efficiency evaluation indicators. These indicators can include resource utilization rate, turnover efficiency, cost-effectiveness ratio, etc., providing a quantitative basis for resource optimization. Constructing a multi-objective optimization function based on medical resource utilization efficiency evaluation indicators is to find a balance between multiple objectives. This function takes into account multiple objectives such as resource utilization efficiency, patient satisfaction, and medical quality, and obtains a resource allocation strategy set through mathematical modeling and optimization algorithms, such as genetic algorithms or particle swarm optimization algorithms. This strategy set contains the optimal resource allocation solutions under different circumstances.
[0050] Sensitivity analysis of the resource allocation strategy set is an important step in evaluating the stability and adaptability of the strategy. Sensitivity analysis changes the values of different parameters and observes the impact of these changes on the strategy effect, thereby obtaining a set of key influencing factors. These factors are the variables that most significantly affect the effect of resource allocation and need to be focused on and controlled in actual operations. Based on the set of key influencing factors, the parameters of the virtual medical twin are tuned, and the results of all previous analyses are applied to the twin. By adjusting the various parameters of the twin, it can more accurately reflect and predict the behavior of the actual medical system. Through this process, medical analysis data is obtained, which not only contains in-depth analysis results of the current medical system, but also includes predictions and optimization suggestions for future situations.
[0051] For example, the virtual medical twin of the emergency department of a general hospital contains detailed operational data for the past year. First, the data is decomposed in time and space, and the data is split into time scales such as hours, days, and weeks, as well as spatial dimensions such as different consulting rooms and examination rooms. After data cleaning and standardization, a normalized data set containing 10 key indicators (such as patient waiting time, doctor workload, equipment utilization rate, etc.) is obtained. Through time series correlation analysis, it is found that there is a 2-hour lag effect between patient arrival rate and average waiting time, that is, the arrival rate at the current moment will affect the waiting time after 2 hours. Based on this association, a decision tree is constructed to predict resource demand during peak hours. After cross-validation and pruning, a streamlined decision rule containing 15 key decision nodes is obtained. Through the analysis of these rules, it is found that adding a triage nurse between 9 and 11 on Monday morning can reduce the average waiting time from 40 minutes to 25 minutes. The resource allocation pattern is clustered using the K-means algorithm, and three typical resource utilization patterns are obtained: efficient, balanced, and overloaded. Based on these patterns, a multi-objective optimization function considering resource utilization, patient waiting time, and medical quality was constructed. A strategy set containing 50 optimal resource allocation solutions under different circumstances was obtained through genetic algorithm. Sensitivity analysis showed that the number of triage nurses and the working hours of doctors were the two most critical factors affecting system performance. Finally, based on these analysis results, the parameters of the virtual medical twin were tuned so that it could accurately predict system performance in different situations and provide real-time suggestions for resource allocation.
[0052] Step S105: Based on the medical analysis data, a comprehensive analysis is performed on the real-time collected individual patient data and hospital operation data to obtain a personalized medical plan and resource optimization strategy;
[0053] Specifically, the medical analysis data is decomposed in multiple dimensions. This step splits the complex medical data according to different dimensions to obtain the patient feature vector and the hospital resource status matrix. The patient feature vector contains information such as the patient's physiological indicators, medical history, and treatment response, while the hospital resource status matrix reflects the real-time usage and availability of various hospital resources. Cluster analysis is performed on the patient feature vector, and machine learning algorithms such as K-means or hierarchical clustering are used to group patients with similar characteristics to obtain patient grouping data. This grouping helps to identify different types of patient groups and provides a basis for subsequent personalized treatment. Based on the patient grouping data, a personalized risk assessment indicator is constructed. This process involves multi-factor analysis and risk modeling, and finally a patient risk distribution map is obtained. This distribution map intuitively shows the health risk level of different patient groups and provides an important reference for the formulation of personalized medical plans. Dynamic load analysis is performed on the hospital resource status matrix. This step uses time series analysis techniques, such as moving average, exponential smoothing, etc., to analyze the dynamic trend of resource usage and obtain a resource utilization curve. This curve reflects the usage of various hospital resources in different time periods. Based on the resource utilization curve, peak forecasting is performed, and time series forecasting models such as ARIMA or Prophet are used to forecast resource demand in the future to obtain a resource demand forecast table. This forecast table provides forward-looking guidance for the rational allocation of resources.
[0054] Cross-matching the patient risk distribution map and the resource demand forecast table is a key step. This process uses heuristic algorithms or optimization algorithms to optimally match the patient's needs with the hospital's resource supply to obtain an initial resource allocation plan. This plan is a preliminary allocation strategy based on patient risk and resource availability. Based on the initial resource allocation plan, a multi-constraint optimization rule set is constructed. This step considers multiple constraints, such as medical quality, cost-effectiveness, patient satisfaction, etc., and uses mathematical programming methods, such as linear programming or integer programming, to obtain a resource scheduling strategy. This strategy not only meets the basic needs of resource allocation, but also seeks a balance between multiple goals.
[0055] Discrete event simulation of resource scheduling strategy is an important step to verify the effectiveness of the strategy. This process uses discrete event simulation software to simulate the execution effect of the strategy under different circumstances and obtain the strategy evaluation results. Based on the strategy evaluation results, the resource scheduling strategy is iteratively optimized. This step uses heuristic algorithms, such as genetic algorithms or simulated annealing algorithms, to continuously adjust and improve the strategy, and finally obtain the optimized resource allocation plan. This plan has been optimized and verified for many rounds and can achieve good results in practical applications. The optimized resource allocation plan is associated with the individual patient data. This process uses data mining techniques, such as association rule mining or decision tree analysis, to combine the resource allocation plan with the needs of specific patients to obtain personalized medical plans and resource optimization strategies. This final output not only takes into account the optimal allocation of overall resources, but also provides personalized treatment recommendations for each patient's specific situation.
[0056] For example, the cardiovascular department of a tertiary hospital is using this system to optimize its medical services. First, the system performs multi-dimensional decomposition of the medical analysis data of the past year, and obtains feature vectors containing 10,000 patients (each vector contains 50 features, such as age, blood pressure, blood lipids, etc.) and a state matrix reflecting the use of 30 key resources in the cardiovascular department. Through the K-means clustering algorithm, the patients are divided into five main groups, such as hypertension group, coronary heart disease group, etc. Based on these groupings, the system constructs a risk assessment model to obtain the average risk score of each group of patients (1-10 points). At the same time, the resource status matrix is analyzed in time series, and the resource demand for the next week is predicted using the ARIMA model. It is found that there will be a peak in the use of inspection equipment from 9 to 11 am on Monday.
[0057] The system matches the patient risk distribution with the resource demand forecast and initially formulates a resource allocation plan, such as recommending that high-risk patients (risk score>8) be given priority for treatment during periods with low resource utilization (such as 2-4 pm on Tuesday). Based on this initial plan, the system constructs a multi-objective optimization model that takes into account medical quality, waiting time, and cost, and obtains a better resource scheduling strategy through integer programming. This strategy is verified by discrete event simulation. The simulation results show that the new strategy can reduce the average patient waiting time from 45 minutes to 30 minutes, while reducing the average workload of doctors from 85% to 75%.
[0058] After multiple rounds of iterative optimization, the system finally generated an optimized resource allocation plan. This plan not only optimizes the allocation of overall resources, but also generates personalized medical advice for each patient. For example, for a 65-year-old hypertensive patient, the system recommends a routine check-up at 10 a.m. on Wednesday and arranges a 15-minute consultation with a specialist at 2 p.m. on the same day, while recommending a personalized drug treatment plan. This comprehensive plan not only improves the utilization efficiency of medical resources, but also ensures that every patient can receive timely and appropriate treatment, reflecting the concept of precision medicine.
[0059] Step S106: Optimize the personalized medical plan and resource optimization strategy through the pre-collected patient flow data to obtain the target management strategy, wherein the target management strategy includes: resource allocation sub-strategy, patient management sub-strategy and service optimization sub-strategy.
[0060] Specifically, the pre-collected patient flow data is subjected to wavelet transformation. This step uses wavelet analysis technology to decompose the time series data into components of different frequencies to obtain multi-scale time series decomposition results. Wavelet transformation can effectively capture various periodic patterns and emergencies in patient flow data, providing a rich information basis for subsequent analysis. Autoregressive analysis is performed on the multi-scale time series decomposition results. This process uses time series prediction models, such as the autoregressive integrated moving average model (ARIMA), to analyze the data change trends at different time scales and obtain short-term patient flow prediction values. This prediction value reflects the changing trend of patient flow in the future (such as the next 24 hours), providing forward-looking guidance for resource allocation.
[0061] The resource utilization rate of each department is calculated based on the short-term forecast of patient flow. This step compares the predicted patient flow with the resource capacity of each department to obtain a dynamic resource load distribution map. This distribution map intuitively shows the resource pressure of different departments in different time periods. Hot spot analysis is performed on the dynamic resource load distribution map. Spatial statistical methods such as hot spot analysis (Getis-Ord Gi* statistic) are used to identify areas and time periods with abnormally high or low resource utilization, and a resource bottleneck identification report is obtained. This report lists potential resource bottleneck points in detail and provides clear goals for subsequent resource allocation. Based on the resource bottleneck identification report, a set of resource dynamic allocation instructions is generated. This process uses heuristic algorithms or optimization algorithms, such as genetic algorithms, to develop a series of resource allocation plans and obtain resource allocation sub-strategies. This sub-strategy contains specific resource allocation instructions, such as increasing the number of medical staff in a department at a specific time or adjusting the equipment usage plan.
[0062] At the same time, the pre-collected patient flow data is analyzed by queue theory. This step uses queue theory models, such as the M / M / c model, to analyze the distribution characteristics of patient waiting time and obtain the patient waiting time distribution function. This function describes the probability distribution of patient waiting time under different circumstances and provides a theoretical basis for optimizing patient management. Based on the patient waiting time distribution function, the triage priority rules are constructed. This process takes into account factors such as the patient's urgency, waiting time, and resource availability to obtain an intelligent triage plan. This plan can dynamically adjust the patient's treatment order and balance the efficiency of medical resource utilization and the patient's waiting experience. Combine the intelligent triage plan with the patient's individual characteristic data. This step uses machine learning algorithms, such as decision trees or random forests, to combine the triage rules with the patient's specific situation to obtain a personalized treatment path. This path tailors the optimal medical process for each patient, including appointment time, examination order, etc. A patient guidance mechanism is constructed based on the personalized treatment path. This process constructs a set of intelligent navigation and reminder systems to obtain a patient management sub-strategy. This sub-strategy can guide patients in real time and optimize their medical experience. The resource allocation sub-strategy and the patient management sub-strategy are collaboratively optimized. This step uses multi-objective optimization algorithms, such as Pareto optimization, to seek a balance between resource efficiency and patient experience, and obtain a service process reconstruction plan. This plan comprehensively considers resource utilization and patient needs and comprehensively optimizes the medical service process. Service quality monitoring indicators are constructed based on the service process reconstruction plan. This process defines a series of key performance indicators (KPIs), such as patient satisfaction and average waiting time, to obtain a service optimization sub-strategy. This sub-strategy provides a complete set of service quality evaluation and improvement mechanisms.
[0063] Integrate the resource allocation sub-strategy, patient management sub-strategy and service optimization sub-strategy. This step uses system integration technology to organically combine the three sub-strategies to form a unified management framework to obtain the target management strategy. This target management strategy is a comprehensive, dynamic and intelligent medical management solution that can respond to changes in patient flow in real time, optimize resource allocation and improve service quality.
[0064] For example, the emergency department of a general hospital is using this system to optimize its operations. First, the system performs a wavelet transform on the patient flow data for the past year, decomposing the data into high-frequency (hourly), medium-frequency (daily), and low-frequency (weekly) components. Through autoregressive analysis, the system predicts the patient flow for the next 24 hours and finds that there will be a peak in visits from 8 to 10 a.m. on Monday, and the number of patients is expected to increase by 30% compared to usual. Based on this prediction, the system calculates the resource utilization rate of each area of the emergency department and finds that during the predicted peak period, the resource utilization rate of the triage area and the internal medicine clinic will exceed 90%, forming a potential bottleneck.
[0065] In response to this situation, the resource allocation sub-strategy recommends adding two triage nurses and one internal medicine physician on Monday morning, and opening two additional clinics. At the same time, through queue theory analysis, the system found that the average waiting time for patients during peak hours may exceed 60 minutes. Based on this, the patient management sub-strategy built a dynamic triage plan, which divides patients into three levels according to their urgency and generates a personalized treatment path for each patient. For example, for a 65-year-old patient with chest pain, the system recommends that he or she undergo an electrocardiogram first and then arrange for an internal medicine physician to diagnose immediately. The service optimization sub-strategy built a set of monitoring indicators, including the average waiting time of patients, the utilization rate of doctors' treatment time, etc., and set corresponding target values. Through the coordinated implementation of these strategies, the system is expected to reduce the average waiting time during peak hours from 60 minutes to 30 minutes, and reduce the average workload of doctors from the original 95% to 85%, greatly improving the operational efficiency and service quality of the emergency department. This comprehensive target management strategy not only optimizes resource allocation, but also significantly improves patients' medical experience, reflecting the refined and personalized modern hospital management concept.
[0066] In the embodiment of the present application, by preprocessing and standardizing multi-source medical data, the effective integration of heterogeneous data is achieved, the problem of information islands in traditional systems is solved, and a solid data foundation is laid for subsequent intelligent analysis. Secondly, by intelligently classifying and labeling standardized medical data sets, structured medical knowledge data is obtained, which not only improves the comprehensibility and availability of data, but also provides more comprehensive and accurate information support for medical decision-making. The virtual medical twin constructed based on structured medical knowledge data provides a comprehensive and dynamic digital mapping for medical management, enabling medical institutions to perform various simulations and predictions in a virtual environment, greatly reducing decision-making risks. Data mining and feature extraction of virtual medical twins are performed, and the obtained medical analysis data provides a scientific basis for the formulation of personalized medical plans and the generation of resource optimization strategies. This decision support based on deep data analysis significantly improves the accuracy of medical services and the efficiency of resource utilization. By comprehensively analyzing the real-time collected patient individual data and hospital operation data, this method can generate truly personalized medical plans, meet the needs of precision medicine, and can also dynamically adjust resource allocation according to real-time conditions, thereby improving the operational efficiency of hospitals. In particular, personalized medical plans and resource optimization strategies are further optimized through pre-collected patient flow data. The resulting target management strategy covers multiple aspects such as resource allocation, patient management, and service optimization, providing a comprehensive intelligent solution for hospital management. This dynamic and intelligent management method can effectively cope with the complexity and variability of the medical environment, and significantly improve the quality of medical services and patient satisfaction. In addition, the application of this method can also reduce medical costs, reduce the waste of medical resources, and create greater economic benefits for medical institutions. This method involves the processing of a large amount of sensitive medical data. During the construction and implementation process, data security and privacy protection issues are fully considered. Multiple security measures such as data desensitization and access control are adopted, which can effectively protect patient privacy and improve the efficiency and accuracy of medical data management based on artificial intelligence.
[0067] In a specific embodiment, the process of executing step S101 may specifically include the following steps:
[0068] (1) Perform data cleaning on multi-source medical data to obtain initial cleaned data, and then perform format consistency processing on the initial cleaned data to obtain data in a unified format;
[0069] (2) Perform data desensitization on the unified format data to obtain privacy-preserving data, and perform time series alignment on the privacy-preserving data to obtain time-consistent data;
[0070] (3) Interpolate the missing values in the time-consistent data to obtain a complete data set, and detect and process outliers on the complete data set to obtain outlier-corrected data;
[0071] (4) Performing data normalization on the abnormal correction data to obtain normalized data, and performing dimension reduction processing on the normalized data through the principal component analysis algorithm to obtain reduced dimension data;
[0072] (5) Segment and encode the dimension-reduced data to obtain encoded data;
[0073] (6) Metadata annotation and index creation are performed on the coded data to obtain a standardized medical data set.
[0074] Specifically, data cleaning is performed on multi-source medical data. This step includes removing duplicate data, correcting obvious erroneous values, and unifying data formats. For example, the date format of different sources is unified to "YYYY-MM-DD", or abnormal body temperature records (such as 50°C) are adjusted to the normal range. Through this process, the initial cleaned data is obtained, which has removed obvious errors and inconsistencies. The initial cleaned data is formatted in a consistent manner. This step converts data from different sources into a unified format and unit. For example, all blood pressure data is unified into mmHg units, or electrocardiogram data output by different devices is converted into a standard ECG format. After this processing, data in a unified format is obtained, laying the foundation for subsequent analysis.
[0075] Data desensitization of uniformly formatted data is a key step in protecting patient privacy. This process includes encrypting or replacing sensitive information such as name, ID number, etc. For example, use a hash algorithm to encrypt the patient ID, or replace the real name with a random code. Through this process, privacy-preserving data is obtained, which not only protects patient privacy but also does not affect subsequent data analysis. Time series alignment of privacy-preserving data ensures that data from different sources remain consistent in the time dimension. For example, align the temperature data recorded once an hour with the blood pressure data recorded once a day on the time axis. This process obtains time-consistent data, which provides a basis for subsequent time series analysis.
[0076] Interpolation of missing values in time-consistent data is an important step to improve data integrity. Common methods include linear interpolation, spline interpolation, or using machine learning algorithms to predict missing values. For example, for missing blood glucose data, linear interpolation can be performed based on the values of the previous and next time points. Through this process, a complete data set is obtained, which improves the continuity and analyzability of the data. Outlier detection and processing is performed on the complete data set. This step uses statistical methods or machine learning algorithms to identify and process outliers. For example, a box plot method is used to detect abnormal blood pressure values, or a clustering algorithm is used to identify abnormal heart rate patterns. Through this process, abnormality-corrected data is obtained, which improves the reliability of the data.
[0077] The purpose of data normalization for abnormal correction data is to eliminate the dimensional differences between different features. Common methods include minimum-maximum normalization or Z-score normalization. For example, all physiological indicators are normalized to the range of 0-1. After this processing, normalized data is obtained, so that different features can be directly compared. Dimensionality reduction of normalized data through principal component analysis algorithm is an important step to reduce data redundancy and extract key features. The PCA algorithm converts the original features into a set of new, unrelated features through linear transformation. By selecting the first few principal components, the data dimension can be significantly reduced while retaining most of the information. This process obtains reduced dimensionality data, which not only reduces the computational complexity but also retains the main features of the data.
[0078] The purpose of data segmentation and coding of dimensionality-reduced data is to better organize and manage data. This step may include segmenting continuous time series data into fixed-length segments, or segmenting according to specific events (such as hospitalization, surgery, etc.). Each data segment will be assigned a unique code to facilitate subsequent retrieval and analysis. Through this process, coded data is obtained, which improves the structured degree of the data.
[0079] Finally, metadata is annotated and indexes are created for the coded data. Metadata includes information such as the source of the data, the time of collection, and the data type, while indexes are used to speed up data query and retrieval. This step adds rich descriptive information to each data segment and establishes an efficient index structure. Through this process, a standardized medical data set is finally obtained, which is not only highly structured but also easy to manage and analyze.
[0080] For example, a hospital collected 5-year follow-up data from 1,000 diabetic patients, including daily blood glucose levels, monthly glycated hemoglobin (HbA1c) test results, quarterly complication checks, etc. First, during the data cleaning process, 5% of the blood glucose data were found to be significantly abnormal (such as 0 mg / dL or 1000 mg / dL), and these data were marked and deleted. Format consistency processing unified the blood glucose units recorded by different devices into mg / dL. Data desensitization replaced the patient's name with a randomly generated ID. Time series alignment aligned daily blood glucose data, monthly HbA1c, and quarterly examination data on the time axis to create a unified time index. For the 20% missing blood glucose data, linear interpolation was used to fill in. Outlier detection found that 3% of HbA1c values were abnormally high (>15%), and these values were adjusted to the normal range (4-14%). Data normalization scaled both blood glucose and HbA1c values to the range of 0-1. PCA dimensionality reduction reduced the original 20 features (including various complication indicators) to 8 principal components, retaining 95% of the information. Data segmentation divided the 5-year data of each patient into 20 quarterly segments, each with a unique code. Finally, metadata was added to each data segment, including data source, collection time, and basic patient information (age, gender, etc.).
[0081] In a specific embodiment, the process of executing step S102 may specifically include the following steps:
[0082] (1) Perform multi-dimensional feature extraction on the standardized medical data set to obtain a medical feature vector set, and perform unsupervised cluster analysis on the medical feature vector set to obtain the initial medical data category;
[0083] (2) Based on the preset medical ontology, the initial medical data categories are semantically mapped to obtain semantically enhanced medical data, and multimodal relationships are extracted from the semantically enhanced medical data to obtain a medical entity association graph;
[0084] (3) Standardize the medical terminology of the medical entity association graph to obtain a standardized medical vocabulary network, and build a medical event time series chain based on the standardized medical vocabulary network to obtain medical time series association data;
[0085] (4) Perform pattern recognition and mining on medical time series correlation data to obtain a medical management model library;
[0086] (5) Conduct simulation evaluation on the medical decision-making process based on the medical management model library to obtain a set of decision-making impact indicators;
[0087] (6) Fusing the decision-making influencing indicator set with the medical time series correlation data to obtain a fused medical knowledge graph;
[0088] (7) Structural processing and indexing of the integrated medical knowledge graph are performed to obtain structured medical knowledge data.
[0089] Specifically, multi-dimensional feature extraction of standardized medical data sets involves a variety of feature extraction techniques, including statistical feature extraction (such as mean, variance, kurtosis, etc.), frequency domain feature extraction (such as Fourier transform coefficients), and time domain feature extraction (such as trend, periodicity, etc.). Through these techniques, the original medical data is converted into a series of numerical features to form a medical feature vector set. This vector set contains the main feature information of the original data and lays the foundation for subsequent analysis. Unsupervised cluster analysis is performed on the medical feature vector set. This step uses clustering algorithms (such as K-means or hierarchical clustering) to classify similar feature vectors into one category. The purpose of cluster analysis is to discover the inherent structure in the data and group medical data with similar features. Through this process, initial medical data categories are obtained, which reflect the natural grouping of data but lack clear medical semantics. In order to give these data categories clear medical meanings, semantic mapping is performed on the initial medical data categories based on the preset medical ontology. Medical ontology is a formalized knowledge representation that contains concepts, relationships, and rules in the medical field. By mapping data categories to concepts in the medical ontology, semantically enhanced medical data is obtained. This process not only gives clear medical meaning to data categories, but also establishes the connection between data and professional medical knowledge. Multimodal relationship extraction of semantically enhanced medical data is a key step in building a medical entity association graph. This process uses natural language processing and graph analysis techniques to identify and extract relationships between entities from different types of medical data (such as text, images, numerical values, etc.). For example, the relationship between symptoms and diseases is extracted from medical record text, or the relationship between physiological indicators and diagnoses is extracted from examination results. Through this step, a medical entity association graph is obtained, which shows a complex network of associations between medical concepts. In order to ensure the consistency and standardization of medical terminology, the medical entity association graph is subjected to medical terminology standardization. This process unifies medical terms with different expressions into standardized medical vocabulary systems, such as the International Classification of Diseases (ICD) or Systematic Medical Terminology (SNOMED CT). Through this process, a standardized medical vocabulary network is obtained, which not only contains standardized medical terms but also retains the relationships between them.
[0090] Constructing a medical event time chain based on a standardized medical vocabulary network is an important step in capturing the temporal order and causal relationship of medical events. This process uses time series analysis techniques, such as time series pattern mining algorithms, to extract the temporal order and mutual relationship of events from medical data. Medical time series association data are obtained, which reflect the association and evolution of medical events in the time dimension. Pattern recognition and mining of medical time series association data is the process of discovering the laws and patterns hidden in the data. This step uses various data mining techniques, such as frequent pattern mining, sequence pattern mining, etc., to discover recurring patterns or laws from a large amount of medical time series data. Through this process, a medical management model library is obtained, which contains various medical management-related models, such as common disease development models, treatment response models, etc.
[0091] Simulating and evaluating the medical decision-making process based on the medical management model library is an important step in applying the extracted knowledge to actual decision-making. This process uses techniques such as decision tree analysis or Monte Carlo simulation to evaluate the possible results and impacts of different decision paths. Through this simulation evaluation, a set of decision impact indicators is obtained, which quantifies the various impacts that different decisions may bring. The purpose of data fusion of the decision impact indicator set with medical time series association data is to combine static decision evaluation results with dynamic time series data. This fusion process uses data integration techniques such as feature fusion or decision-level fusion to organically combine the information of the two data sources. Through this step, a fused medical knowledge graph is obtained, which is a complex knowledge network that integrates time series relationships, decision impacts, and medical knowledge.
[0092] The fused medical knowledge graph is structured and indexed. This step converts the complex knowledge graph into a structured data format that is easy to store, retrieve, and apply. By establishing a multidimensional index structure, the retrieval efficiency of the data is improved. This process ultimately results in structured medical knowledge data, which not only contains rich medical knowledge and relationships, but also has good structure and operability.
[0093] For example, a hospital collected 5-year follow-up data of 1,000 patients with type 2 diabetes, including daily blood sugar levels, monthly HbA1c test results, quarterly complication checks, etc. First, multi-dimensional feature extraction was performed on these data to extract 20 key features, such as blood sugar fluctuation amplitude and HbA1c change trend, to form a feature vector for each patient. The feature vectors were clustered using the K-means algorithm, with K=5 set to divide the patients into 5 initial categories. Then, based on the preset diabetes medical ontology, the 5 categories were semantically mapped to obtain medically meaningful classifications such as "good blood sugar control type" and "high risk of complications type". Next, by analyzing the patient's medication records and blood sugar changes, the drug-effect relationship was extracted and a medical entity association map was constructed. In the process of terminology standardization, different expressions used by different doctors (such as "diabetic nephropathy" and "diabetic nephropathy") were unified into the standard terms in ICD-10. Based on the standardized data, a time series chain of disease development for each patient was constructed, including key time points such as first diagnosis, occurrence of complications, and treatment adjustment. By mining patterns in these time series chains, we discovered several typical disease progression patterns, such as "rapid complication development type" and "stable control type". Based on these patterns, we simulated and evaluated different treatment options, such as evaluating the impact of early insulin intervention on complication development. The evaluation results showed that for patients with "rapid complication development type", early insulin intervention may reduce the complication rate from 60% to 40% within 5 years. These evaluation results were fused with the original time series data to form a comprehensive diabetes knowledge graph. Finally, this knowledge graph was converted into a structured data format to establish a multidimensional index based on disease progression stage, treatment options, and complication risk.
[0094] In a specific embodiment, the process of executing step S103 may specifically include the following steps:
[0095] (1) Perform semantic analysis on structured medical knowledge data to obtain a medical entity relationship network, and build a three-dimensional space model based on the medical entity relationship network to obtain the initial virtual medical environment;
[0096] (2) Dynamically map the parameters of the initial virtual medical environment to obtain an interactive medical scene;
[0097] (3) Access real-time data streams of interactive medical scenarios to obtain data-driven virtual medical models;
[0098] (4) Establish multi-dimensional evaluation indicators based on the data-driven virtual medical model to obtain the medical twin evaluation indicator set, and simulate and verify the medical twin evaluation indicator set to obtain the twin operation rule set;
[0099] (5) Digitally map the medical process based on the twin operation rule set to obtain virtual medical operation data;
[0100] (6) Conduct multi-scenario testing and optimization on virtual medical operation data to obtain a virtual medical twin.
[0101] Specifically, natural language processing and knowledge graph technologies are used to deeply understand the concepts, entities, and relationships in the data. Medical terms are semantically encoded by applying word vector models (such as Word2Vec or BERT), and then the relationship extraction algorithm (such as remote supervised learning or attention mechanism) is used to identify the relationship between entities, and finally a complex medical entity relationship network is formed. This network not only includes medical entities (such as diseases, symptoms, drugs, etc.), but also includes various relationships between them (such as "cause", "treatment", "concurrency", etc.).
[0102] Constructing a three-dimensional spatial model based on the medical entity relationship network is to transform abstract medical knowledge into a visualized three-dimensional structure. This process involves graphics and spatial modeling technology, such as using force-directed graph drawing to transform network relationships into spatial positions, and then visualizing them through three-dimensional rendering technology (such as OpenGL or WebGL). The constructed three-dimensional spatial model includes the physical layout of the hospital, equipment distribution, patient flow path, etc., forming an initial virtual medical environment. Dynamic parameter mapping of the initial virtual medical environment is the key to making the digital twin "alive". This step maps real-time data, historical data, and predicted data into the virtual environment, and uses data binding technology to associate data with virtual objects. For example, information such as patient admission, transfer, and discharge is updated in real time to the virtual environment, or data such as equipment usage status and medical staff workload are dynamically displayed. Through this process, an interactive medical scene is obtained, which can not only show the current state, but also simulate system responses in different situations. Real-time data stream access to the interactive medical scene is the key to keeping the digital twin synchronized with the actual medical system. This step uses stream processing technology (such as Apache Kafka or Apache Flink) to build a data bridge between the actual medical system and the virtual environment to ensure that the virtual environment can reflect changes in the actual situation in real time. This includes access to data streams from various medical devices, real-time data from patient monitoring systems, updates to hospital information systems, etc. Through this real-time data access, a data-driven virtual medical model is obtained, which can accurately reflect the real-time status and dynamic changes of the medical system.
[0103] The purpose of establishing multi-dimensional evaluation indicators based on data-driven virtual medical models is to comprehensively evaluate the performance and efficiency of medical systems. This process involves indicator construction and data analysis, using statistical analysis and machine learning techniques to evaluate system performance from multiple perspectives. Evaluation indicators include but are not limited to patient waiting time, bed turnover rate, medical equipment utilization rate, and medical staff workload. Through these indicators, a medical twin evaluation indicator set is obtained, which provides a quantitative basis for the management and optimization of the medical system. Simulation verification of the medical twin evaluation indicator set is an important step to ensure that the digital twin can accurately reflect and predict the behavior of the actual medical system. This process uses discrete event simulation technology (such as SimPy or AnyLogic) to simulate various medical scenarios, and continuously adjusts and optimizes the parameters and algorithms of the digital twin by comparing the predicted results in the virtual environment with the performance of the actual medical system. Through repeated verification and adjustment, a twin operation rule set is obtained, which defines how the virtual medical twin operates and responds to various situations.
[0104] Digital mapping of medical processes based on twin operation rule sets is to transform abstract rules into specific virtual operation processes. This process uses workflow modeling technology (such as BPMN or UML activity diagrams) to simulate various processes in the medical system, such as patient consultation processes, medical resource scheduling processes, emergency response processes, etc. Through this mapping, virtual medical operation data is obtained, which reflects the operating status and performance of the medical system under different circumstances. Multi-scenario testing and optimization of virtual medical operation data is a key step in improving virtual medical twins. This process uses scenario analysis and optimization algorithms (such as genetic algorithms or reinforcement learning) to simulate various possible medical scenarios, such as daily operations, peak management, emergency handling, etc., and continuously optimizes the parameters and algorithms of the twins based on the test results. Through this repeated testing and optimization, a highly accurate and reliable virtual medical twin is finally obtained.
[0105] For example, a tertiary hospital decided to build a digital twin to optimize the operation of its emergency department. First, semantic analysis was performed on the structured medical knowledge data of the emergency department to identify key entities such as "patient", "doctor", "nurse", "examination room", "examination equipment", etc., as well as the relationships between them such as "patient-waiting-examination room" and "doctor-use-examination equipment", forming a medical entity relationship network. Based on this network, a three-dimensional spatial model of the emergency department was constructed, including the physical layout of the waiting area, examination room, emergency room, etc., as well as the distribution of medical staff and equipment. Then, dynamic parameters such as patient waiting time and doctor workload were mapped to this model to create an interactive virtual emergency scene. By accessing real-time data streams, such as the patient arrival rate updated every 15 minutes and the average frequency of examination equipment use of 10 times per hour, the virtual scene can reflect the actual situation in real time. Based on this model, a set of evaluation indicators including average waiting time, doctor utilization rate, bed turnover rate, etc. was established. Through simulation verification, it was found that the average waiting time during the peak period predicted in the virtual environment was 120 minutes, while the actual data showed 110 minutes, and the prediction model was fine-tuned accordingly. Based on the optimized rules, the entire emergency treatment process was digitally mapped, including the entire process from patient arrival, triage, waiting, treatment to departure. Finally, by simulating the operating conditions under different situations (such as ordinary working days, holiday peaks, public health emergencies, etc.), the virtual twin was continuously adjusted and optimized. For example, when simulating the holiday peak, it was found that the average waiting time could be reduced from the original 90 minutes to 60 minutes by adding a triage nurse. This highly accurate virtual medical twin can not only reflect the operating status of the emergency department in real time, but also predict the impact of different decisions, providing a powerful decision support tool for hospital management.
[0106] In a specific embodiment, the process of executing step S104 may specifically include the following steps:
[0107] (1) Perform spatiotemporal data decomposition on the virtual medical twin to obtain a multi-dimensional medical data stream, and perform data cleaning and standardization on the multi-dimensional medical data stream to obtain a standardized medical data set;
[0108] (2) Perform time series correlation analysis on the standardized medical data set to obtain the medical event association chain;
[0109] (3) Construct a medical decision tree based on the medical event association chain to obtain the initial medical decision rule set;
[0110] (4) Cross-validate and prune the initial medical decision-making rule set to obtain streamlined medical decision-making rules;
[0111] (5) Mining medical resource utilization patterns based on streamlined medical decision-making rules to obtain resource allocation feature maps;
[0112] (6) Perform cluster analysis on the resource allocation characteristic diagram to obtain the evaluation index of medical resource utilization efficiency;
[0113] (7) Construct a multi-objective optimization function based on the medical resource utilization efficiency evaluation index to obtain a resource allocation strategy set;
[0114] (8) Conduct sensitivity analysis on the resource allocation strategy set to obtain the key influencing factor set;
[0115] (9) Based on the set of key influencing factors, the parameters of the virtual medical twin are tuned to obtain medical analysis data.
[0116] Specifically, this process uses multidimensional data analysis techniques, such as principal component analysis (PCA) or tensor decomposition, to split complex twin data by time and space dimensions to obtain multidimensional medical data streams. These data streams include time series (such as hourly patient flow), spatial distribution (such as resource utilization in different departments), and attribute information (such as physiological indicators of patients). For time series data, the following formula can be used for decomposition:
[0117] X(t)=T(t)+S(t)+R(t);
[0118] Among them, X(t) is the original time series, T(t) is the trend component, S(t) is the seasonal component, and R(t) is the residual.
[0119] Next, these multi-dimensional medical data streams are cleaned and standardized, anomaly detection algorithms (such as isolation forests) are used to identify and process outliers, and Z-score standardization and other methods are used to convert data of different scales to the same standard, and finally a standardized medical data set is obtained. The Z-score standardization formula is as follows:
[0120] ;
[0121] Among them, z is the standardized value, x is the original value, is the mean, is the standard deviation.
[0122] Performing time series correlation analysis on normalized medical data sets is a key step in mining potential associations between medical events. This process uses time series analysis techniques, such as cross-correlation analysis or Granger causality test, to explore the temporal dependencies between different medical events. The cross-correlation function can be expressed as:
[0123] ;
[0124] in, is the hysteresis The cross-correlation coefficient, and are two time series, and is their mean, and n is the total length of the time series.
[0125] Building a medical decision tree based on the medical event association chain is an important step in converting the results of time series analysis into actionable decisions. This process uses a decision tree algorithm, such as ID3 or C4.5, to extract decision rules from the medical event association chain. The information gain of the decision tree can be calculated using the following formula:
[0126] ;
[0127] in, is the information gain of feature a, is the entropy of the data set T, It is the conditional entropy of feature a under given conditions.
[0128] Cross-validation and pruning optimization of the initial medical decision rule set is the key to improving the accuracy and generalization ability of the decision tree. During the pruning process, the cost complexity function can be expressed as:
[0129] ;
[0130] in, is the cost complexity, Is a leaf node The number of samples in Is a leaf node The entropy of is the number of leaf nodes in the tree, is the complexity parameter.
[0131] Mining medical resource utilization patterns based on streamlined medical decision-making rules is an important step in optimizing resource allocation. This step uses association rule mining algorithms, such as Apriori or FP-Growth, to discover potential patterns of resource utilization from large amounts of medical data. The support and confidence of association rules can be calculated using the following formula:
[0132] ;
[0133] ;
[0134] Among them, A is the first matter, B is the second matter, is the total number of transactions, Indicates the frequency of occurrence of item A and item B at the same time, It indicates the frequency of occurrence of item A, support measures the frequency of occurrence of the rule in the data set, and confidence measures the reliability of the rule. Cluster analysis of the resource allocation feature graph is a step to further summarize and generalize the resource utilization pattern. This process uses clustering algorithms, such as K-means or hierarchical clustering, to group similar resource allocation patterns into one category.
[0135] The purpose of constructing a multi-objective optimization function based on the medical resource utilization efficiency evaluation index is to find a balance between multiple objectives. The multi-objective optimization problem can be expressed as:
[0136]
[0137]
[0138] in, No. Utilization rate of resources; :Total available resources; : Waiting time of the jth patient; :No. medical quality indicators; :Total medical costs; B:Budget constraints; , , , : The weight of each goal; : Decision variable, indicating the assignment to the The number of resources; :No. The maximum available amount of a resource; : Maximum allowed waiting time; :Minimum medical quality requirement index value.
[0139] Sensitivity analysis of resource allocation strategy sets is an important step in evaluating the stability and adaptability of strategies. The Sobol index in global sensitivity analysis can be expressed as:
[0140] ;
[0141] in, It is The Sobol index of the parameters, is the output variance caused by the ith parameter, is the total output variance.
[0142] The purpose of constructing a multi-objective optimization function based on the evaluation index of medical resource utilization efficiency is to find a balance between multiple objectives. This function takes into account multiple objectives such as resource utilization efficiency, patient satisfaction, and medical quality, and uses a multi-objective optimization algorithm such as NSGA-II (Non-dominated Sorting Genetic Algorithm II) to obtain a resource allocation strategy set. This strategy set contains the optimal resource allocation solutions under different circumstances.
[0143] Sensitivity analysis of the resource allocation strategy set is an important step in evaluating the stability and adaptability of the strategy. This process uses local sensitivity analysis or global sensitivity analysis methods, such as the Morris method or the Sobol index, to change the values of different parameters and observe the impact of these changes on the strategy effect, thereby obtaining a set of key influencing factors. These factors are the variables that most significantly affect the effect of resource allocation and need to be focused on and controlled in actual operations. Based on the set of key influencing factors, the parameters of the virtual medical twin are tuned. This step uses optimization algorithms, such as gradient descent or genetic algorithms, to adjust the various parameters of the twin so that it can more accurately reflect and predict the behavior of the actual medical system. Through this process, medical analysis data is obtained, which not only contains in-depth analysis results of the current medical system, but also includes predictions and optimization suggestions for future situations.
[0144] For example, the virtual medical twin of the emergency department of a tertiary hospital contains detailed operational data for the past year. First, the data is decomposed in time and space, and the data is split into time scales such as hours, days, and weeks, as well as spatial dimensions such as different consulting rooms and examination rooms. After data cleaning and standardization, a normalized data set containing 10 key indicators (such as patient waiting time, doctor workload, equipment utilization rate, etc.) is obtained. Through time series correlation analysis, it is found that there is a 2-hour lag effect between patient arrival rate and average waiting time, that is, the arrival rate at the current moment will affect the waiting time after 2 hours. Based on this association, a decision tree is constructed to predict resource demand during peak hours. After cross-validation and pruning, the number of nodes in the decision tree is reduced from 50 to 15, but the accuracy rate only drops from 85% to 83%. Through the analysis of these rules, it is found that between 9 and 11 am on Monday morning, adding a triage nurse can reduce the average waiting time from 40 minutes to 25 minutes. The resource allocation patterns were clustered using the K-means algorithm, with the k value set to 3, and three typical resource utilization patterns were obtained: efficient, balanced, and overloaded. Based on these patterns, a multi-objective optimization function considering resource utilization, patient waiting time, and medical quality was constructed. A strategy set containing 50 optimal resource allocation solutions under different circumstances was obtained by solving the problem using the NSGA-II algorithm. Sensitivity analysis showed that the number of triage nurses and the working hours of doctors were the two most critical factors affecting system performance, with sensitivity indexes of 0.6 and 0.5, respectively. Finally, based on these analysis results, the parameters of the virtual medical twin were tuned to enable it to accurately predict system performance under different circumstances and provide real-time suggestions for resource allocation. For example, during the simulated peak period of influenza, the twin suggested adding two doctors and one nurse to the original basis, which was expected to reduce the average waiting time of patients from 90 minutes to 45 minutes, while reducing the average workload of doctors from 85% to 70%. These medical analysis data provide valuable decision support for hospital managers, helping them to better cope with various complex medical scenarios.
[0145] In a specific embodiment, the process of executing step S105 may specifically include the following steps:
[0146] (1) Perform multi-dimensional decomposition of medical analysis data to obtain patient feature vectors and hospital resource status matrices;
[0147] (2) Perform cluster analysis on the patient feature vectors to obtain patient grouping data, and build personalized risk assessment indicators based on the patient grouping data to obtain the patient risk distribution map;
[0148] (3) Perform dynamic load analysis on the hospital resource status matrix to obtain a resource utilization curve, and perform peak prediction based on the resource utilization curve to obtain a resource demand forecast table;
[0149] (4) Cross-match the patient risk distribution map and resource demand forecast table to obtain the initial resource allocation plan, and build a multi-constraint optimization rule set based on the initial resource allocation plan to obtain the resource scheduling strategy;
[0150] (5) Conduct discrete event simulation on the resource scheduling strategy to obtain the strategy evaluation results, and iteratively optimize the resource scheduling strategy based on the strategy evaluation results to obtain the optimized resource allocation plan;
[0151] (6) Correlate the optimized resource allocation plan with the patient's individual data to obtain personalized medical plans and resource optimization strategies.
[0152] Specifically, high-dimensional data is converted into low-dimensional representation using techniques such as principal component analysis (PCA) or tensor decomposition. For patient data, the PCA method can extract key features and obtain patient feature vectors. At the same time, a similar method is applied to hospital resource data to obtain a hospital resource status matrix. These two results reflect the core characteristics of patients and the dynamic status of hospital resources, respectively. Cluster analysis is performed on patient feature vectors. The commonly used K-means algorithm groups similar patient features into a group through iterative calculations to obtain patient grouping data. Based on these groups, personalized risk assessment indicators are constructed. Logistic regression models can be used to calculate the probability of each patient belonging to different risk levels, thereby obtaining a patient risk distribution map. This distribution map intuitively shows the health risk levels of different patient groups. Dynamic load analysis is performed on the hospital resource status matrix. Time series analysis methods, such as the autoregressive integrated moving average model (ARIMA), are used to analyze the dynamic trend of resource use and obtain a resource utilization curve. Based on this curve, peak detection algorithms and time series prediction models, such as Prophet, are used to predict resource demand in the future and obtain a resource demand forecast table. This forecast table provides forward-looking guidance for the rational allocation of resources.
[0153] Cross-matching the patient risk distribution map and the resource demand forecast table is a key step. This process uses heuristic algorithms or linear programming methods, such as the simplex method, to optimally match the patient's needs with the hospital's resource supply and obtain an initial resource allocation plan. Based on this plan, a multi-constraint optimization rule set is constructed, taking into account multiple constraints, such as medical quality, cost-effectiveness, and patient satisfaction. Mathematical programming methods, such as integer programming, are used to obtain a resource scheduling strategy. This strategy not only meets the basic needs of resource allocation, but also seeks a balance between multiple goals. Discrete event simulation of resource scheduling strategies is an important step in verifying the effectiveness of strategies. This process uses discrete event simulation software to simulate the execution effect of strategies under different circumstances and obtain strategy evaluation results. These results reflect the performance of strategies in various possible situations and provide a basis for further optimization. Based on the strategy evaluation results, the resource scheduling strategy is iteratively optimized. This step uses heuristic algorithms, such as genetic algorithms or simulated annealing algorithms, to continuously adjust and improve the strategy and finally obtain an optimized resource allocation plan. This solution has been optimized and verified for many rounds and can achieve good results in practical applications.
[0154] The optimized resource allocation plan is associated with the individual patient data. This process uses association rule mining algorithms, such as the Apriori algorithm, to combine the resource allocation plan with the specific patient needs to obtain personalized medical plans and resource optimization strategies. This final output not only takes into account the optimal allocation of overall resources, but also provides personalized treatment recommendations for each patient's specific situation.
[0155] For example, the cardiovascular department of a tertiary hospital uses this method to optimize its medical services. First, the medical data of the past year was decomposed in multiple dimensions to obtain feature vectors containing 10,000 patients (each vector contains 50 features, such as age, blood pressure, blood lipids, etc.) and a state matrix reflecting the use of 30 key resources in the cardiovascular department. Through the K-means clustering algorithm, the patients were divided into five main groups, such as hypertension group, coronary heart disease group, etc. Based on these groups, a risk assessment model was constructed to obtain the average risk score of each group of patients (1-10 points). At the same time, the resource state matrix was analyzed in time series, and the resource demand for the next week was predicted using the ARIMA model. It was found that there would be a peak in the use of examination equipment from 9 to 11 am on Monday, and the expected utilization rate would reach 95%. The system matched the patient risk distribution with the resource demand forecast and initially formulated a resource allocation plan, such as recommending that high-risk patients (risk score>8) be given priority to see a doctor during the period with low resource utilization (such as Tuesday afternoon from 2 to 4, with an expected utilization rate of 60%). Based on this initial plan, a multi-objective optimization model considering medical quality, waiting time and cost was constructed, and a better resource scheduling strategy was obtained through integer programming. This strategy was verified by discrete event simulation, and the simulation results showed that the new strategy can reduce the average patient waiting time from 45 minutes to 30 minutes, while reducing the average workload of doctors from 85% to 75%.
[0156] After multiple rounds of iterative optimization, the system finally generated an optimized resource allocation plan. This plan not only optimizes the allocation of overall resources, but also generates personalized medical advice for each patient. For example, for a 65-year-old hypertensive patient (risk score 9.2), the system recommends a routine check-up at 10 a.m. on Wednesday (estimated device utilization rate 70%), and arranges a 15-minute consultation with a specialist at 2 p.m. on the same day, and recommends a personalized drug treatment plan.
[0157] In a specific embodiment, the process of executing step S106 may specifically include the following steps:
[0158] (1) Perform wavelet transform on the pre-collected patient flow data to obtain multi-scale time series decomposition results;
[0159] (2) Perform autoregressive analysis on the multi-scale time series decomposition results to obtain short-term prediction values of patient flow;
[0160] (3) Calculate the resource utilization rate of each department based on the short-term predicted value of patient flow, obtain a dynamic resource load distribution map, and perform hot spot analysis on the dynamic resource load distribution map to obtain a resource bottleneck identification report;
[0161] (4) Generate a resource dynamic allocation instruction set based on the resource bottleneck identification report to obtain a resource allocation sub-strategy;
[0162] (5) Perform queue theory analysis on the pre-collected patient flow data to obtain the patient waiting time distribution function, and construct triage priority rules based on the patient waiting time distribution function to obtain an intelligent triage solution;
[0163] (6) Combine the intelligent triage plan with the patient's individual characteristic data to obtain a personalized treatment path;
[0164] (7) Construct a patient guidance mechanism based on the personalized treatment pathway and obtain patient management sub-strategies;
[0165] (8) Co-optimize the resource allocation sub-strategy and the patient management sub-strategy to obtain a service process reconstruction plan, and build service quality monitoring indicators based on the service process reconstruction plan to obtain a service optimization sub-strategy;
[0166] (9) Integrate the resource allocation sub-strategy, patient management sub-strategy, and service optimization sub-strategy to obtain the target management strategy.
[0167] Specifically, wavelet transform of pre-collected patient flow data is a key step in achieving multi-scale time series decomposition. Wavelet transform can decompose time series data into components of different frequencies, effectively capturing various periodic patterns and emergencies in patient flow data. This process uses the discrete wavelet transform (DWT) algorithm to decompose the original patient flow data into high-frequency detail coefficients and low-frequency approximate coefficients to obtain multi-scale time series decomposition results. These decomposition results reflect the changing characteristics of patient flow at different time scales, providing a rich information basis for subsequent analysis. Autoregressive analysis is performed on the multi-scale time series decomposition results, and this process uses the autoregressive integrated moving average (ARIMA) model. The ARIMA model predicts patient flow in the short term in the future by analyzing the autocorrelation and moving average characteristics of the time series. The parameters of the model are determined by the maximum likelihood estimation method to obtain the short-term prediction value of patient flow. This prediction value reflects the changing trend of patient flow in the future (such as the next 24 hours), providing forward-looking guidance for resource allocation.
[0168] The resource utilization rate of each department is calculated based on the short-term forecast of patient flow. This step compares the predicted patient flow with the resource capacity of each department, calculates the resource utilization rate of different time periods and departments, and obtains a dynamic resource load distribution map. This distribution map intuitively shows the resource pressure of each department of the hospital at different time points. Hot spot analysis is performed on the dynamic resource load distribution map. Spatial statistical methods such as Getis-Ord Gi* statistics are used to identify areas and time periods with abnormally high or low resource utilization, and a resource bottleneck identification report is obtained. This report lists potential resource bottleneck points in detail and provides clear goals for subsequent resource allocation. Based on the resource bottleneck identification report, a set of resource dynamic allocation instructions is generated. This process uses heuristic algorithms or optimization algorithms, such as genetic algorithms, to develop a series of resource allocation plans and obtain resource allocation sub-strategies. This sub-strategy contains specific resource allocation instructions, such as increasing the number of medical staff in a department at a specific time or adjusting the equipment usage plan.
[0169] At the same time, the pre-collected patient flow data is analyzed by queue theory. In this step, the M / M / c queuing model is used to analyze the distribution characteristics of patient waiting time and obtain the patient waiting time distribution function. This function describes the probability distribution of patient waiting time under different circumstances and provides a theoretical basis for optimizing patient management. Based on the patient waiting time distribution function, the triage priority rule is constructed. This process takes into account factors such as the patient's urgency, waiting time, and resource availability to obtain an intelligent triage solution. This solution can dynamically adjust the patient's treatment order and balance the efficiency of medical resource utilization and the patient's waiting experience. The intelligent triage solution is combined with the patient's individual characteristic data. In this step, machine learning algorithms such as decision trees or random forests are used to combine the triage rules with the patient's specific situation to obtain a personalized treatment path. This path tailors the optimal medical process for each patient, including appointment time, examination order, etc. A patient guidance mechanism is constructed based on the personalized treatment path. This process constructs a set of intelligent navigation and reminder systems to obtain a patient management sub-strategy. This sub-strategy can guide patients in real time and optimize their medical experience.
[0170] The resource allocation sub-strategy and the patient management sub-strategy are collaboratively optimized. This step uses multi-objective optimization algorithms, such as Pareto optimization, to seek a balance between resource efficiency and patient experience, and obtains a service process reconstruction solution. This solution comprehensively considers resource utilization and patient needs and comprehensively optimizes the medical service process. Service quality monitoring indicators are constructed based on the service process reconstruction solution. This process defines a series of key performance indicators (KPIs), such as patient satisfaction and average waiting time, to obtain a service optimization sub-strategy. This sub-strategy provides a complete set of service quality evaluation and improvement mechanisms. Integrate the resource allocation sub-strategy, patient management sub-strategy, and service optimization sub-strategy. This step uses system integration technology to organically combine the three sub-strategies to form a unified management framework and obtain a target management strategy. This target management strategy is a comprehensive, dynamic, and intelligent medical management solution that can respond to changes in patient flow in real time, optimize resource allocation, and improve service quality.
[0171] For example, the emergency department of a general hospital is using this method to optimize its operations. First, the patient flow data of the past year is subjected to wavelet transformation to decompose the data into high-frequency (hourly), medium-frequency (daily), and low-frequency (weekly) components. Through autoregression analysis, the patient flow in the next 24 hours is predicted, and it is found that there will be a peak in the morning from 8 to 10 on Monday, and the number of patients is expected to increase by 30% compared with usual. Based on this prediction, the resource utilization rate of each area of the emergency department is calculated, and it is found that during the predicted peak period, the resource utilization rate of the triage area and the internal medicine clinic will exceed 90%, forming a potential bottleneck.
[0172] In response to this situation, the resource allocation sub-strategy recommends adding 2 triage nurses and 1 internal medicine physician on Monday morning, and opening 2 additional clinics. At the same time, through queue theory analysis, it was found that the average waiting time for patients during peak hours may exceed 60 minutes. Based on this, the patient management sub-strategy constructed a dynamic triage plan, which divides patients into three levels according to their urgency and generates a personalized treatment path for each patient. For example, for a 65-year-old patient with chest pain, the system recommends that he or she give priority to an electrocardiogram and then immediately arrange for a physician to diagnose.
[0173] The service optimization sub-strategy has built a set of monitoring indicators, including the average waiting time of patients, the utilization rate of doctors' treatment time, etc., and set corresponding target values. Through the coordinated implementation of these strategies, it is expected that the average waiting time during peak hours can be reduced from 60 minutes to 30 minutes, and the average workload of doctors can be reduced from the original 95% to 85%, which greatly improves the operational efficiency and service quality of the emergency department.
[0174] The above describes the medical data management method based on artificial intelligence in the embodiment of the present application. The following describes the medical data management system based on artificial intelligence in the embodiment of the present application. Figure 2 In the embodiments of the present application, an embodiment of a medical data management system based on artificial intelligence includes:
[0175] The processing module 201 is used to pre-process the multi-source medical data collected from the electronic medical record system, medical equipment and wearable devices to obtain a standardized medical data set;
[0176] The classification module 202 is used to intelligently classify and label the standardized medical data set to obtain structured medical knowledge data;
[0177] A construction module 203 is used to construct a digital twin based on the structured medical knowledge data to obtain a virtual medical twin;
[0178] An extraction module 204 is used to perform data mining and feature extraction on the virtual medical twin to obtain medical analysis data;
[0179] The analysis module 205 is used to comprehensively analyze the real-time collected patient individual data and hospital operation data based on the medical analysis data to obtain personalized medical solutions and resource optimization strategies;
[0180] The optimization module 206 is used to optimize the personalized medical plan and resource optimization strategy through the pre-collected patient flow data to obtain a target management strategy, wherein the target management strategy includes: a resource allocation sub-strategy, a patient management sub-strategy and a service optimization sub-strategy.
[0181] Through the collaborative cooperation of the above components, through the preprocessing and standardization of multi-source medical data, the effective integration of heterogeneous data is achieved, the problem of information islands in traditional systems is solved, and a solid data foundation is laid for subsequent intelligent analysis. Secondly, by intelligently classifying and labeling standardized medical data sets, structured medical knowledge data is obtained, which not only improves the comprehensibility and availability of data, but also provides more comprehensive and accurate information support for medical decision-making. The virtual medical twin constructed based on structured medical knowledge data provides a comprehensive and dynamic digital mapping for medical management, enabling medical institutions to perform various simulations and predictions in a virtual environment, greatly reducing decision-making risks. Data mining and feature extraction of virtual medical twins, the medical analysis data obtained provide a scientific basis for the formulation of personalized medical plans and the generation of resource optimization strategies. This decision support based on deep data analysis significantly improves the accuracy of medical services and the efficiency of resource utilization. By comprehensively analyzing the real-time collected patient individual data and hospital operation data, this method can generate truly personalized medical plans to meet the needs of precision medicine, and can also dynamically adjust resource allocation according to real-time conditions, improving the operational efficiency of hospitals. In particular, personalized medical plans and resource optimization strategies are further optimized through pre-collected patient flow data. The resulting target management strategy covers multiple aspects such as resource allocation, patient management, and service optimization, providing a comprehensive intelligent solution for hospital management. This dynamic and intelligent management method can effectively cope with the complexity and variability of the medical environment, and significantly improve the quality of medical services and patient satisfaction. In addition, the application of this method can also reduce medical costs, reduce the waste of medical resources, and create greater economic benefits for medical institutions. This method involves the processing of a large amount of sensitive medical data. During the construction and implementation process, data security and privacy protection issues are fully considered. Multiple security measures such as data desensitization and access control are adopted, which can effectively protect patient privacy and improve the efficiency and accuracy of medical data management based on artificial intelligence.
[0182] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the artificial intelligence-based medical data management method.
[0183] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0184] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A medical data management method based on artificial intelligence, characterized in that: The medical data management method based on artificial intelligence includes: Preprocess multi-source medical data collected from electronic medical record systems, medical devices and wearable devices to obtain standardized medical data sets; Intelligently classify and label standardized medical data sets to obtain structured medical knowledge data; Building a digital twin based on the structured medical knowledge data to obtain a virtual medical twin; Performing data mining and feature extraction on the virtual medical twin to obtain medical analysis data; Based on the medical analysis data, a comprehensive analysis is performed on the real-time collected individual patient data and hospital operation data to obtain a personalized medical plan and a resource optimization strategy, which specifically includes: performing multi-dimensional decomposition on the medical analysis data to obtain a patient feature vector and a hospital resource status matrix; performing cluster analysis on the patient feature vector to obtain patient grouping data, and constructing a personalized risk assessment index based on the patient grouping data to obtain a patient risk distribution map; performing dynamic load analysis on the hospital resource status matrix to obtain a resource utilization curve, and performing peak prediction based on the resource utilization curve to obtain a resource demand prediction table; cross-matching the patient risk distribution map and the resource demand prediction table to obtain an initial resource allocation plan, and constructing a multi-constraint optimization rule set based on the initial resource allocation plan to obtain a resource scheduling strategy; performing discrete event simulation on the resource scheduling strategy to obtain a strategy evaluation result, and iteratively optimizing the resource scheduling strategy based on the strategy evaluation result to obtain an optimized resource allocation plan; correlating the optimized resource allocation plan with the individual patient data to obtain the personalized medical plan and resource optimization strategy; The personalized medical plan and resource optimization strategy are optimized by pre-collected patient flow data to obtain a target management strategy, which specifically includes: performing wavelet transform on the pre-collected patient flow data to obtain a multi-scale time series decomposition result; performing autoregressive analysis on the multi-scale time series decomposition result to obtain a short-term prediction value of patient flow; calculating the resource utilization rate of each department based on the short-term prediction value of patient flow to obtain a dynamic resource load distribution diagram, and performing hot spot analysis on the dynamic resource load distribution diagram to obtain a resource bottleneck identification report; generating a resource dynamic allocation instruction set based on the resource bottleneck identification report to obtain a resource allocation sub-strategy; performing pre-collected patient flow data on the patient flow data to obtain a resource allocation sub-strategy; performing autoregressive analysis on the multi-scale time series decomposition result to obtain a short-term prediction value of patient flow; calculating the resource utilization rate of each department based on the short-term prediction value of patient flow to obtain a dynamic resource load distribution diagram, and performing hot spot analysis on the dynamic resource load distribution diagram to obtain a resource bottleneck identification report; generating a resource dynamic allocation instruction set based on the resource bottleneck identification report to obtain a resource allocation sub-strategy; performing autoregressive analysis on the pre-collected patient flow data to obtain a resource allocation sub-strategy; Conduct queue theory analysis to obtain a patient waiting time distribution function, and construct triage priority rules based on the patient waiting time distribution function to obtain an intelligent triage plan; combine the intelligent triage plan with the patient's individual characteristic data to obtain a personalized treatment path; construct a patient guidance mechanism based on the personalized treatment path to obtain a patient management sub-strategy; collaboratively optimize the resource allocation sub-strategy and the patient management sub-strategy to obtain a service process reconstruction plan, and construct service quality monitoring indicators based on the service process reconstruction plan to obtain a service optimization sub-strategy; integrate the resource allocation sub-strategy, patient management sub-strategy and service optimization sub-strategy to obtain the target management strategy.
2. The medical data management method based on artificial intelligence according to claim 1, characterized in that: The multi-source medical data collected from the electronic medical record system, medical equipment and wearable devices are preprocessed to obtain a standardized medical data set, including: Performing data cleaning on the multi-source medical data to obtain initial cleaned data, and performing format consistency processing on the initial cleaned data to obtain unified format data; Performing data desensitization processing on the unified format data to obtain privacy-preserving data, and performing time series alignment on the privacy-preserving data to obtain time-consistent data; Performing interpolation processing on missing values in the time-consistent data to obtain a complete data set, and performing outlier detection and processing on the complete data set to obtain outlier-corrected data; Performing data normalization processing on the abnormal correction data to obtain normalized data, and performing dimension reduction processing on the normalized data through a principal component analysis algorithm to obtain dimension reduced data; Segmenting and encoding the dimension-reduced data to obtain encoded data; Metadata annotation and index creation are performed on the coded data to obtain the standardized medical data set.
3. The medical data management method based on artificial intelligence according to claim 1, characterized in that: The standardized medical data set is intelligently classified and labeled to obtain structured medical knowledge data, including: Performing multi-dimensional feature extraction on the standardized medical data set to obtain a medical feature vector set, and performing unsupervised cluster analysis on the medical feature vector set to obtain initial medical data categories; Based on a preset medical ontology, semantic mapping is performed on the initial medical data categories to obtain semantically enhanced medical data, and multimodal relationship extraction is performed on the semantically enhanced medical data to obtain a medical entity association graph; Performing medical terminology standardization processing on the medical entity association graph to obtain a standardized medical vocabulary network, and constructing a medical event time series chain based on the standardized medical vocabulary network to obtain medical time series association data; Performing pattern recognition and mining on the medical time series correlation data to obtain a medical management model library; Based on the medical management model library, a medical decision-making process is simulated and evaluated to obtain a decision-making impact indicator set; Performing data fusion on the decision-influencing indicator set and the medical time series associated data to obtain a fused medical knowledge graph; The fused medical knowledge graph is structured and indexed to obtain the structured medical knowledge data.
4. The medical data management method based on artificial intelligence according to claim 1, characterized in that: The step of constructing a digital twin based on the structured medical knowledge data to obtain a virtual medical twin includes: Performing semantic analysis on the structured medical knowledge data to obtain a medical entity relationship network, and constructing a three-dimensional space model based on the medical entity relationship network to obtain an initial virtual medical environment; Performing dynamic parameter mapping on the initial virtual medical environment to obtain an interactive medical scene; Accessing the interactive medical scene with real-time data stream to obtain a data-driven virtual medical model; Establishing multi-dimensional evaluation indicators based on the data-driven virtual medical model to obtain a medical twin evaluation indicator set, and performing simulation verification on the medical twin evaluation indicator set to obtain a twin operation rule set; Digitally mapping the medical process based on the twin operation rule set to obtain virtual medical operation data; Multi-scenario testing and optimization are performed on the virtual medical operation data to obtain the virtual medical twin.
5. The medical data management method based on artificial intelligence according to claim 1, characterized in that: The data mining and feature extraction of the virtual medical twin to obtain medical analysis data includes: Performing spatiotemporal data decomposition on the virtual medical twin to obtain a multi-dimensional medical data stream, and performing data cleaning and standardization processing on the multi-dimensional medical data stream to obtain a normalized medical data set; Performing time series correlation analysis on the standardized medical data set to obtain a medical event association chain; Building a medical decision tree based on the medical event association chain to obtain an initial medical decision rule set; Cross-validating and pruning the initial medical decision rule set to obtain streamlined medical decision rules; Mining the medical resource utilization pattern based on the streamlined medical decision-making rules to obtain a resource allocation feature map; Performing cluster analysis on the resource allocation characteristic graph to obtain a medical resource utilization efficiency evaluation index; Constructing a multi-objective optimization function based on the medical resource utilization efficiency evaluation index to obtain a resource allocation strategy set; Performing sensitivity analysis on the resource allocation strategy set to obtain a set of key influencing factors; The parameters of the virtual medical twin are tuned based on the set of key influencing factors to obtain the medical analysis data.
6. A medical data management system based on artificial intelligence, characterized in that: Used to execute the artificial intelligence-based medical data management method as described in any one of claims 1 to 5, the artificial intelligence-based medical data management system comprising: A processing module, used to pre-process multi-source medical data collected from electronic medical record systems, medical devices and wearable devices to obtain a standardized medical data set; The classification module is used to intelligently classify and label standardized medical data sets to obtain structured medical knowledge data; A construction module, used to construct a digital twin based on the structured medical knowledge data to obtain a virtual medical twin; An extraction module, used to perform data mining and feature extraction on the virtual medical twin to obtain medical analysis data; An analysis module is used to conduct a comprehensive analysis of the real-time collected patient individual data and hospital operation data based on the medical analysis data to obtain a personalized medical plan and a resource optimization strategy, specifically including: performing a multi-dimensional decomposition on the medical analysis data to obtain a patient feature vector and a hospital resource status matrix; performing a cluster analysis on the patient feature vector to obtain patient grouping data, and constructing a personalized risk assessment index based on the patient grouping data to obtain a patient risk distribution map; performing a dynamic load analysis on the hospital resource status matrix to obtain a resource utilization curve, and performing a peak prediction based on the resource utilization curve to obtain a resource demand prediction table; cross-matching the patient risk distribution map and the resource demand prediction table to obtain an initial resource allocation plan, and constructing a multi-constraint optimization rule set based on the initial resource allocation plan to obtain a resource scheduling strategy; performing a discrete event simulation on the resource scheduling strategy to obtain a strategy evaluation result, and iteratively optimizing the resource scheduling strategy based on the strategy evaluation result to obtain an optimized resource allocation plan; performing an association analysis on the optimized resource allocation plan and the patient individual data to obtain the personalized medical plan and resource optimization strategy; The optimization module is used to optimize the personalized medical plan and resource optimization strategy through the pre-collected patient flow data to obtain the target management strategy, wherein the target management strategy includes: resource allocation sub-strategy, patient management sub-strategy and service optimization sub-strategy, specifically including: performing wavelet transform on the pre-collected patient flow data to obtain a multi-scale time series decomposition result; performing autoregressive analysis on the multi-scale time series decomposition result to obtain a short-term prediction value of patient flow; calculating the resource utilization rate of each department based on the short-term prediction value of patient flow to obtain a dynamic resource load distribution diagram, and performing hot spot analysis on the dynamic resource load distribution diagram to obtain a resource bottleneck identification report; generating a resource dynamic allocation instruction based on the resource bottleneck identification report set to obtain a resource allocation sub-strategy; perform queue theory analysis on the pre-collected patient flow data to obtain a patient waiting time distribution function, and build a triage priority rule based on the patient waiting time distribution function to obtain an intelligent triage plan; combine the intelligent triage plan with the patient's individual characteristic data to obtain a personalized treatment path; build a patient guidance mechanism based on the personalized treatment path to obtain a patient management sub-strategy; collaboratively optimize the resource allocation sub-strategy and the patient management sub-strategy to obtain a service process reconstruction plan, and build a service quality monitoring indicator based on the service process reconstruction plan to obtain a service optimization sub-strategy; integrate the resource allocation sub-strategy, the patient management sub-strategy and the service optimization sub-strategy to obtain the target management strategy.
7. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the artificial intelligence-based medical data management method as described in any one of claims 1-5 is implemented.
Citation Information
Patent Citations
Digital twinborn health management system
CN113035353A
Medical service automation process optimization method
CN117894421A
Pension service optimization method and system based on big data
CN118035870A
Service package generation method, apparatus and device based on patient data, and storage medium
WO2023240837A1
