AI-based medical and nursing integrated data management method and system, and storage medium

Abstract

The invention discloses an AI-based medical and health care integrated data management method and system and a storage medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: carrying out the data preprocessing and data standardization of collected multi-source medical and health care data, and obtaining a standardized medical data set; performing classification and semantic annotation on the data of the standardized medical data set by using a preset medical ontology, and constructing a structured medical knowledge graph; performing virtual mapping based on the structured medical knowledge graph by adopting a digital twinborn model to form a virtual medical twinborn body; based on the virtual medical twins, combining real-time monitoring data and historical patient flow data, through multi-dimensional feature extraction and pension mode recognition, dynamically optimizing and presetting a personalized medical scheme and a resource scheduling strategy, and outputting a medical care service optimization scheme; generating a target service list in combination with the analyzed doctor-patient dialogue text, and distributing the target service list to a preset pension service cooperation terminal; the method has the effect of improving the real-time performance and the accuracy of the old people medical care service strategy.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an AI-based integrated medical and nursing data management method, system, and program product. Background Art

[0002] As my country's aging population accelerates, the elderly population is characterized by "three highs and two lows": high rates of chronic disease, disability, and medical dependence, while family care functions are weakened and social service provision is inefficient. Currently, there is a disconnect between medical care and elderly care, with hospitals focusing on disease diagnosis and treatment, while elderly care institution management systems focus on daily life. The lack of cross-system data exchange can easily lead to patients having to submit repeated paper documents when seeking medical treatment, prolonging hospitalization processing times. Emergency response in home-based elderly care is also delayed, failing to meet the timeliness requirements for acute illness treatment.

[0003] Therefore, in view of the disconnection between the scheduling of medical and elderly care service resources and medical plans for the elderly, how to build a dynamic mapping relationship between medical data and elderly care services to achieve continuous services from disease treatment to health management? Although the existing technology has proposed a smart elderly care data platform that adopts a data middle-end architecture, it still has not solved the problem of the lack of coordination between real-time decision-making and offline analysis in the process of medical and elderly care services. That is, there is a defect that the real-time and accuracy of medical and elderly care service strategies for the elderly are still relatively low, and there is room for improvement. Summary of the Invention

[0004] In order to improve the real-time and accuracy of medical and nursing service strategies for the elderly, this application provides an AI-based integrated medical and nursing data management method, system and program product.

[0005] In the first aspect, the invention objectives of this application are achieved by adopting the following technical solutions: AI-based integrated medical and nursing data management methods include: Collect multi-source medical and nursing data based on AI, perform data preprocessing and data standardization on the multi-source medical and nursing data, and obtain a standardized medical data set; Using a preset medical ontology to classify and semantically annotate the data in the standardized medical dataset, a structured medical knowledge graph containing patient feature vectors and a time series chain of medical events is constructed; a digital twin model is used to perform virtual mapping based on the structured medical knowledge graph to form a virtual medical twin; Based on the virtual medical twin, real-time monitoring data and historical patient flow data are combined to dynamically optimize preset personalized medical plans and resource scheduling strategies through multi-dimensional feature extraction and elderly care model identification, and output medical and elderly care service optimization plans; According to the medical and nursing service optimization plan, combined with the parsed doctor-patient conversation text, the attention mechanism is used to generate a target service list and distribute it to the preset elderly care service collaborative terminal.

[0006] By adopting the above technical solutions, multi-source medical and nursing data include structured and unstructured data collected from electronic medical record systems, medical equipment, wearable devices and paper health records; medical and nursing service optimization solutions include resource allocation sub-strategies and patient management sub-strategies; nursing models include hospital nursing departments, nursing homes + medical care, and community home + nursing homes. The present application provides a multi-source heterogeneous data fusion and analysis method in the process of medical data and nursing services, that is, through AI-driven multi-source data collection (such as medical / wearable / paper archives) and intelligent preprocessing, it solves the problems of low efficiency in digitization of paper archives and insufficient data standardization in traditional nursing systems, resulting in high error rates when accessing data across systems, and improves the availability of medical and nursing data; and based on the semantic annotation technology of medical ontology, compared with the traditional keyword matching method, it improves the efficiency of constructing patient feature vectors and improves the accuracy of the temporal chain association of medical events, thereby helping to reduce the scheduling of medical and nursing service resources and medical methods for the elderly. The situation of disconnection between the case and the case; further, this application uses virtual medical twins to build a dynamic mapping relationship between medical data and elderly care services to achieve continuous services from disease treatment to health management; combined with real-time monitoring data of wearable monitoring devices and / or medical equipment for the elderly, it improves the timeliness of chronic disease warning, improves the response speed of elderly care and medical resource scheduling and the service matching accuracy of medical care service strategies, and improves the real-time and accuracy of medical care service strategies for the elderly; in order to further improve the real-time and service intelligence level in the medical care service process for the elderly, combined with the doctor-patient dialogue parsing technology driven by the attention mechanism (such as NLP technology to parse the doctor-patient dialogue text) to generate a target service list, which is allocated to the preset elderly care service collaborative terminal. The entire medical care service process forms an intelligent medical care service process closed loop based on AI drive + medical care service optimization plan + real-time monitoring of the user's physical condition, which is suitable for the medical care service needs of the elderly. The medical care service strategy of this technical solution has high real-time and accuracy.

[0007] In a preferred example of the present application, the multi-source medical and nursing data includes paper health records, physiological signals and medical data collected by wearable devices; and the data preprocessing includes: Using optical character recognition, we perform image binarization, noise filtering, and character segmentation on paper health records. We also use a medical term dictionary to perform entity recognition and convert unstructured text into standardized medical terminology. Perform Kalman filtering to denoise the physiological signals collected by wearable devices, use standardized methods to eliminate dimensional differences, and unify the data format in combination with medical device interface protocols; Based on the time series alignment algorithm, discrete medical data and continuous physiological monitoring data are temporally and spatially matched to build a data chain for the patient's entire life cycle; The isolation forest algorithm is used to detect abnormal data points in the multi-source medical and nursing data, and the sliding window method is used to fill the missing values of the sensor data.

[0008] By adopting the above-mentioned technical solutions, this application combines OCR technology (i.e., optical character recognition), Kalman filtering and time series alignment algorithm, which can effectively process paper health records, data collected by wearable devices and medical data, thereby achieving the diversity and comprehensiveness of data sources; using OCR technology and medical terminology dictionary for entity recognition, unstructured text is converted into standardized medical terminology; using standardized methods to eliminate dimensional differences, thereby improving the consistency and availability of data; based on the time series alignment algorithm, the temporal and spatial matching of physiological monitoring data and medical data is achieved, which helps to establish a data chain for the patient's entire life cycle; using the isolation forest algorithm to detect abnormal data points, and filling in missing values through the sliding window method, thereby ensuring the accuracy of the data during the data analysis process.

[0009] In a preferred example of this application, the steps of constructing a structured medical knowledge graph include: Using an integrated medical semantic network and clinical medical terminology, we perform multi-granular entity annotation on patient medical history, medication records, and lifestyle data to generate annotated datasets. The time series features of medical events are extracted based on the LSTM-CRF model, and a triple relationship network of disease-symptom-treatment plan is constructed through knowledge graph learning method. Introducing an attention mechanism to calculate the time weights of different medical events and dynamically generate the patient's health status vector; Based on the annotated data set and the triple relationship network, a structured medical knowledge graph including the patient health status vector is constructed, and the structured medical knowledge graph is stored in a graph database.

[0010] By adopting the above technical solution, we first use the integrated medical semantic network and clinical medical terminology set to perform multi-granular annotation of patient information to generate a high-quality annotated dataset; then, based on the LSTM-CRF model, we can effectively extract key temporal features from medical records, which helps to understand the patient's disease progression and treatment response; then, we introduce the attention mechanism to calculate the time weights of different medical events, which can dynamically generate patient health status vectors to reflect changes in the patient's health status; and store the structured medical knowledge graph in the graph database to facilitate users to quickly query and analyze.

[0011] In a preferred example of this application, the dynamic optimization of preset personalized medical plans and resource scheduling strategies, and output of medical and nursing service optimization plans, include: Based on the federated learning framework, we jointly analyze the patient data within the institution and regional medical big data to train a personalized medical plan recommendation model; Using the Q-learning algorithm to optimize resource scheduling strategies based on preset constraints, including the nursing skills matrix, equipment utilization threshold, and service response time; Use Monte Carlo tree search to simulate multi-scenario service paths and select the resource allocation solution with the lowest overall cost; Based on the trained personalized medical plan recommendation model, optimized resource scheduling strategy and selected resource allocation plan, the model parameters are continuously optimized using AI-driven reinforcement learning to obtain an optimized medical and nursing service plan that adapts to the health status migration rules of the elderly population.

[0012] By adopting the above technical solutions and jointly analyzing multi-source data based on the federated learning framework, the trained personalized medical plan recommendation model can formulate the most suitable treatment plan based on the specific situation of each patient; using the Q-learning algorithm and considering various constraints, such as the nursing skills matrix, equipment utilization rate and service response time, the resource scheduling strategy is optimized, which is conducive to improving service efficiency and service quality; simulating different service paths through Monte Carlo tree search and selecting the resource allocation plan with the lowest overall cost helps to reduce operating costs while ensuring service quality; using reinforcement learning AI to drive the continuous optimization of model parameters to adapt to the changing patterns of the health status of the elderly population.

[0013] In a preferred example of the present application, the target service list generation and allocation steps include: Use natural language processing technology to analyze doctor-patient conversation texts and extract key medical needs based on domain knowledge graphs; Generate a list of candidate service tasks based on the pointer network and calculate the service task priority through the classifier; The improved Hungarian algorithm is used to match the optimal caregiver and task combination in combination with task constraints, including geographical distance, skill matching, and task urgency coefficient. The service instruction is pushed through the preset terminal communication interface, and the service work order confirmation instruction corresponding to the service instruction is obtained to achieve two-way confirmation of the service task.

[0014] By adopting the above technical solution, natural language processing technology is used to parse the text of doctor-patient conversations, and key medical needs are extracted in combination with domain knowledge graphs to ensure the accuracy of service tasks; at the same time, a list of candidate service tasks is generated based on a pointer network, and the priority is calculated through a classifier, so that urgent and important service tasks can be handled in a timely manner; the improved Hungarian algorithm is used to combine constraints such as geographical distance, skill matching and task urgency coefficient to optimally match caregivers and service tasks; service instructions are pushed through the preset terminal communication interface and service work order confirmation instructions are obtained, realizing two-way confirmation of service tasks and ensuring the reliability of service execution.

[0015] In a preferred example of this application, the step of constructing a structured medical knowledge graph further includes: Based on the patient's medical imaging data, a 3D convolutional neural network is used to extract lesion features, which are then combined with the patient's electronic medical record text features to generate a multimodal fusion feature vector. A federated learning framework is used to combine data from several nursing homes. The LSTM-CRF model embeds a differential privacy mechanism in local model training. The differential privacy mechanism sets a privacy budget threshold that is the upper bound of the difference in output distributions between any two adjacent datasets. The weights of different modal data of the multi-source medical and nursing data are dynamically allocated through a graph attention network to construct an enhanced medical knowledge graph with cross-modal association capabilities.

[0016] By adopting the above technical solutions, a 3D convolutional neural network is used to extract lesion features based on the patient's medical imaging data, and then fused with the electronic medical record text features to generate a multimodal fusion feature vector, which improves the depth of understanding of the patient's condition; a federated learning framework is used to combine data from multiple nursing homes, and a differential privacy mechanism is embedded in local model training, which not only ensures data security but also promotes cross-institutional knowledge sharing; through the graph attention network, weights are dynamically assigned to data of different modalities, and an enhanced medical knowledge graph with cross-modal association capabilities is constructed, which helps to more comprehensively understand the patient's health status.

[0017] In a preferred embodiment of the present application, the dynamic optimization of preset personalized medical plans and resource scheduling strategies and output of medical and nursing service optimization plans also include: Construct a causal graph of medical events based on a structural causal model to identify key intervention nodes; Using the virtual medical twin to simulate the impact of different resource scheduling schemes on the health status of the elderly population, generating multidimensional evaluation indicators, including emergency response delay and complication rate; Predict the potential risks of the optimal resource scheduling strategy through counterfactual reasoning algorithm to obtain resource scheduling prediction risk information; The health benefit weight coefficient in the Q-learning algorithm is dynamically adjusted based on the resource scheduling prediction risk information.

[0018] By adopting the above technical solutions, a causal graph of medical events is constructed based on the structural causal model, which can identify key intervention nodes that affect patients' health and provide a basis for personalized treatment plans; using virtual medical twins to simulate the impact of different resource scheduling plans on the health status of the elderly population, and generate multi-dimensional evaluation indicators including emergency response delay and complication rate, which helps to optimize resource allocation; through the counterfactual reasoning algorithm, the potential risks of the optimal resource scheduling strategy are predicted to obtain resource scheduling prediction risk information; based on the resource scheduling prediction risk information, the health benefit weight coefficient in the Q-learning algorithm is dynamically adjusted to make the resource scheduling strategy more in line with the actual situation and improve the effectiveness of medical services.

[0019] In the second aspect, the invention objective of this application is achieved by adopting the following technical solutions: An AI-based integrated medical and nursing data management system, configured to execute the aforementioned AI-based integrated medical and nursing data management method, includes: Data collection module, which collects multi-source medical and nursing data based on AI technology; The data processing module is used to preprocess and standardize the collected multi-source medical and nursing data to generate a standardized medical data set; the classification and annotation module uses a preset medical ontology to classify and semantically annotate the data in the standardized medical data set to construct a structured medical knowledge graph containing patient feature vectors and medical event time series chains; A digital twin module, which uses a digital twin model to perform virtual mapping based on the structured medical knowledge graph to form a virtual medical twin; The solution optimization module combines real-time monitoring data with historical patient flow data, extracts multi-dimensional features, and identifies elderly care patterns to dynamically optimize preset personalized medical plans and resource scheduling strategies, ultimately outputting optimized medical and elderly care service plans. A service list generation and allocation module is used to generate a target service list based on the medical and nursing service optimization plan and the parsed doctor-patient conversation text using an attention mechanism, and allocate the target service list to a preset elderly care service collaborative terminal.

[0020] By adopting the above technical solutions, the data acquisition module collects medical and nursing data from multiple channels based on AI technology; the data processing module preprocesses and standardizes the original data to generate high-quality standardized medical data sets; the classification and labeling module uses preset medical ontology to classify and semantically label the data, and constructs a structured medical knowledge graph containing patient feature vectors and medical event time series chains; the digital twin module creates a virtual medical twin based on the structured medical knowledge graph to simulate and predict changes in patient health status under different circumstances; the solution optimization module combines real-time monitoring data and historical data, and dynamically optimizes personalized medical plans and resource scheduling strategies through feature extraction and pattern recognition to output the best medical and nursing service optimization plan; at the same time, the service list generation and allocation module generates a target service list according to the optimization plan, and allocates it to the elderly care service collaboration terminal, ensuring the efficient execution and two-way confirmation of service tasks.

[0021] In a third aspect, the invention objective of this application is achieved by adopting the following technical solutions: A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of the above-mentioned AI-based integrated medical and nursing data management method.

[0022] Fourthly, the invention objectives of this application are achieved by adopting the following technical solutions: A computer program product includes a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned AI-based integrated medical and nursing data management method.

[0023] In summary, this application includes at least one of the following beneficial technical effects: 1. Through virtual medical twins, by building a dynamic mapping relationship between medical data and elderly care services, continuous services from disease treatment to health management can be achieved. Combined with real-time monitoring data from wearable monitoring devices and / or medical devices for the elderly, this improves the effectiveness of chronic disease early warning, the speed of elderly care and medical resource scheduling, and the accuracy of service matching for medical and elderly care service strategies, thereby improving the real-time and accuracy of medical and elderly care service strategies for the elderly. 2. Constructing a causal map of medical events based on a structural causal model can identify key intervention nodes that affect patients' health and provide a basis for personalized treatment plans; using virtual medical twins to simulate the impact of different resource scheduling plans on the health status of the elderly population, generating multi-dimensional evaluation indicators including emergency response delays and complication rates, which helps to optimize resource allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a flow chart of an AI-based integrated medical and nursing data management method in one embodiment of the present application; Figure 2 This is a flowchart of step S3 in the AI-based integrated medical and nursing data management method in one embodiment of the present application. DETAILED DESCRIPTION

[0025] The present application is further described in detail below with reference to the accompanying drawings.

[0026] In one embodiment, if Figure 1 As shown, this application discloses an AI-based integrated medical and nursing data management method, which specifically includes the following steps: S1: Collect multi-source medical and nursing data based on AI, perform data preprocessing and data standardization on the multi-source medical and nursing data, and obtain a standardized medical data set.

[0027] In this embodiment, multi-source medical and nursing data includes paper health records (such as medical records and prescriptions), physiological signals collected by wearable devices (such as heart rate and blood oxygen), and medical data (such as CT images and ECG monitors); data standardization refers to converting heterogeneous data into a unified format (such as the HL7 FHIR standard).

[0028] Specifically, step S1 includes: S11: Perform image binarization, noise filtering, and character segmentation on paper health records based on optical character recognition, and perform entity recognition in combination with a medical term dictionary to convert unstructured text into standardized medical terminology.

[0029] Specifically, optical character recognition (OCR technology) uses an adaptive threshold algorithm to binarize images, noise filtering uses planting filtering combined with morphological opening operations, and the medical term dictionary is constructed based on the "Chinese Medical Subject Headings".

[0030] S12: Perform Kalman filtering to denoise the physiological signals collected by wearable devices, use standardized methods to eliminate dimensional differences, and unify the data format in combination with the medical device interface protocol.

[0031] Specifically, the Kalman filter parameters including the state transfer matrix and the observation matrix are standardized in combination with the Z-score normalization formula.

[0032] S13: Based on the time series alignment algorithm, discrete medical data and continuous physiological monitoring data are temporally and spatially matched to build a data chain for the patient's entire life cycle.

[0033] Specifically, the time series alignment algorithm uses dynamic time warping (DTW), and the window size is set to 10% of the data length; building a patient's full life cycle data chain means using the consultation time as the axis to associate outpatient records (such as interval ≤ 30 days) and hospitalization records (such as interval ≤ 7 days).

[0034] S14: Use the isolation forest algorithm to detect abnormal data points in multi-source medical and nursing data, and fill the missing values of sensor data through the sliding window method.

[0035] S2: Use the preset medical ontology to classify and semantically annotate the data of the standardized medical data set, and construct a structured medical knowledge graph that includes patient feature vectors and medical event time series chains; use the digital twin model to perform virtual mapping based on the structured medical knowledge graph to form a virtual medical twin.

[0036] In this embodiment, the preset medical ontology refers to a disease-symptom-treatment plan semantic network constructed based on a clinical medical terminology set (referring to the SNOMED CT clinical terminology set); the digital twin is a simulation model that maps the patient's health status in real time through a virtual model.

[0037] Specifically, step S2 includes: The steps to build a structured medical knowledge graph include: S21: Use an integrated medical semantic network and clinical medical terminology set to perform multi-granular entity annotation on patient medical history, medication records, and lifestyle data to generate annotated datasets.

[0038] In this embodiment, the integrated medical semantic network refers to a medical concept hierarchical network constructed based on standards such as SNOMED CT and ICD-11; the clinical medical terminology set includes a standardized dictionary of entities such as diseases, symptoms, drugs, and their relationships.

[0039] Specifically, the data sources are: disease diagnosis records in electronic medical records (EMRs); medication records (such as drug name, dosage, and frequency); lifestyle habit data (such as diet and exercise frequency, obtained through questionnaires); the entity labeling process is: first use the BERT-Med model for pre-training (parameters: learning rate 2e-5, batch size 16); then the rule engine supplements the labeling, that is, matching the generic name of the drug through regular expressions (such as "aspirin" → DRUG_001), and then outputting the data set in the JSON-LD format.

[0040] S22: Extract the temporal features of medical events based on the LSTM-CRF model, and construct a triple relationship network of disease-symptom-treatment plan through knowledge graph learning method.

[0041] In this embodiment, the LSTM-CRF model is a joint architecture of a bidirectional long short-term memory network and a conditional random field. The LSTM layer parameters of the LSTM-CRF model are: 128 hidden units, a dropout rate of 0.3, and the CRF layer parameters are a transfer matrix dimension of 128×128, and 50 iterations. The knowledge graph learning method is entity relationship embedding based on the TransE algorithm.

[0042] For example, the triple relationship network is constructed as follows: disease-symptom relationship example: (hypertension, clinical manifestations, headache); the treatment plan generation rule is: if (hypertension, drug resistance) is detected, then (adjust medication, replace ARB drugs) is triggered, and the graph database is optimized as follows: establish a B+ tree index for the node ID.

[0043] S23: Introduce the attention mechanism to calculate the time weights of different medical events and dynamically generate the patient health status vector.

[0044] In this embodiment, the attention mechanism is a multi-head attention mechanism, that is, the attention weights of different representation subspaces are calculated in parallel; the dynamic health state vector is a vector representation that integrates temporal features and static attributes.

[0045] Specifically, the attention calculation formula of the multi-head attention mechanism is: Among them, Q (Query) refers to the query matrix, which represents the target information that needs to be paid attention to (such as the current medical event), and is projected from the input data (such as the patient's historical health record) through linear transformation; K (Key) refers to the key matrix, which represents all the information to be compared (such as the timestamps and types of all medical events). The K matrix may contain medical event records in the past year (such as medication, examinations, and symptoms); V (Value) refers to the value matrix, which stores the actual data content related to the key. The V matrix may contain the specific values of each medical event (such as blood pressure value, drug dosage); d k =64 is the dimension scaling factor; the parameters are set to number of heads = 8; embedding dimension = 256; It is a dot product operation and softmax normalization, calculating the transposed matrix product of Q and K. The attention score matrix is obtained; Attention(Q, K, V) is the weighted value matrix.

[0046] S24: Based on the labeled dataset and triple relationship network, construct a structured medical knowledge graph containing the patient's health status vector, and store the structured medical knowledge graph in the graph database.

[0047] In this embodiment, the graph database is a NoSQL database that stores data in nodes and relationships; index optimization refers to accelerating graph traversal queries through spatial partitioning.

[0048] Furthermore, using Neo4j's APOC library to batch import data and index high-frequency query patterns (such as "find all medication records for a patient") helps shorten the response time of complex queries (involving three-layer relationship traversal), for example, reducing the response time from 1.2s to 300ms.

[0049] S3: Based on virtual medical twins, combined with real-time monitoring data and historical patient flow data, through multi-dimensional feature extraction and elderly care pattern recognition, it dynamically optimizes preset personalized medical plans and resource scheduling strategies, and outputs medical and elderly care service optimization plans.

[0050] In this embodiment, the virtual medical twin refers to a patient health status simulation model built based on digital twin technology; patient flow data refers to the patient's flow records between different medical scenarios (hospitals, communities, and families); and the elderly care model includes home-based elderly care, hospital elderly care department, elderly care institution + medical care combination, and community home + elderly care institution combination elderly care model.

[0051] Specifically, the entities (diseases, medications) in the structured medical knowledge graph are first associated with real-time sensor data (heart rate, blood sugar) through a graph database (Neo4j); for example, an abnormal patient body temperature (real-time data) → triggers the "fever → infection risk" association path in the knowledge graph; the multi-dimensional feature extraction includes time-frequency domain features and behavioral pattern features; time-frequency domain features refer to the statistical characteristics of signals in time and frequency; behavioral pattern features refer to living habit patterns extracted by activity recognition algorithms, and the patient's physiological feature extraction includes wavelet decomposition of wearable device data (such as accelerometer signals) to extract time-frequency domain features (energy, entropy); patient behavior pattern recognition includes using CNN models to identify behaviors such as falls and sleep.

[0052] In this embodiment, if Figure 2 As shown, the dynamic optimization in step S3 presets personalized medical plans and resource scheduling strategies, and outputs the medical and nursing service optimization plan, including: S31: Based on the federated learning framework, jointly analyze patient data within the institution and regional medical big data to train a personalized medical plan recommendation model.

[0053] In this embodiment, the federated learning framework includes horizontal federated learning and differential privacy. Horizontal federated learning refers to multiple participants sharing model parameters rather than original data, and differential privacy refers to adding noise to protect individual data privacy based on patient privacy protection requirements. The federated learning framework uses the FATE framework (version 1.10) to deploy federated learning nodes, and the logistic regression parameters are: learning rate 0.1, regularization coefficient 0.01); Gaussian noise parameter σ = 0.1, meeting the privacy budget of ε ≤ 0.5.

[0054] S32: Use the Q-learning algorithm to optimize the resource scheduling strategy based on preset constraints, which include the nursing skills matrix, equipment utilization rate threshold, and service response time.

[0055] In this embodiment, the Q-learning algorithm refers to a reinforcement learning algorithm based on the Bellman equation; the caregiver skill matrix refers to a multidimensional array that quantifies the caregiver's nursing ability; a Q-learning reward function is set, and the reward function parameters include response speed, cost-effectiveness, conflict rate, and health benefits. Corresponding weight coefficients are set for each parameter. The state vector dimension of the Q-learning algorithm is 10 (including the patient's heart rate, blood sugar, care need level, etc.); the action space is the caregiver allocation plan (for example, "Caregiver A is responsible for medication, and Caregiver B is responsible for turning the patient over").

[0056] S33: Simulate multi-scenario service paths through Monte Carlo tree search and select the resource allocation solution with the lowest overall cost.

[0057] In this embodiment, the Monte Carlo tree adopts the UCT algorithm, which is used for an upper confidence bound strategy that balances exploration and utilization; the comprehensive cost refers to the weighted sum of time cost, economic cost, and health risk.

[0058] Specifically, the exploration process of the Monte Carlo tree is as follows: 200 candidate paths are generated in each simulation (depth = 3 layers); the pruning strategy is to retain the top 10% of the paths with the cumulative rewards; the parameters are set as: exploration constant Where N is the total number of simulations. For example, in a device failure scenario, the optimal path is to select a backup device.

[0059] S34: Based on the trained personalized medical plan recommendation model, optimized resource scheduling strategy and selected resource allocation plan, the AI-driven continuous optimization of model parameters using reinforcement learning is used to obtain an optimized medical and nursing service plan that adapts to the health status migration patterns of the elderly population.

[0060] In this embodiment, the proximal policy optimization algorithm (PPO algorithm, where (baseline PPO parameters: γ = 0.99, λ = 0.95)) is adopted to balance the stability and efficiency of model training; the health state migration law refers to the dynamic evolution pattern of the health state of the elderly.

[0061] S4: Based on the medical and nursing service optimization plan and the parsed doctor-patient conversation text, the attention mechanism is used to generate a target service list and distribute it to the preset elderly care service collaborative terminal.

[0062] In this embodiment, the attention mechanism refers to a neural network module that focuses on key information through weighted distribution. Pre-configured elderly care service collaboration terminals include mobile app terminals, tablets, smartwatches, and WeChat mini-programs. Service instructions are pushed in real time via the WeChat mini-program Push API, supporting multi-channel access via SMS and voice calls. The Hungarian algorithm is a combinatorial optimization algorithm for solving the optimal matching problem on bipartite graphs. Doctor-patient conversation texts are parsed using the BERT model and / or NLP technology.

[0063] Specifically, step S4 includes: S41: Use natural language processing technology to analyze doctor-patient conversation texts and combine domain knowledge graphs to extract key medical needs.

[0064] In this embodiment, natural language processing (NLP) uses a deep learning model to perform text entity recognition; the domain knowledge graph contains a semantic network of medical entities such as diseases, symptoms, and treatment plans.

[0065] Specifically, the BERT-Med model is used for conversation intent recognition; when extracting key medical needs, an entity linking example is used: inputting "I have been feeling very dizzy recently" will link to the knowledge graph node "symptom_dizziness"; if "pain" and "medication" are mentioned in the doctor-patient conversation, it can be determined that the patient's need is to adjust the painkillers, and the urgency of this matter is level 2 (relatively important but not urgent); if "fall" is mentioned in the conversation, it can be determined that the patient needs emergency care, and the urgency of this matter is level 1 (very urgent and needs to be dealt with immediately).

[0066] S42: Generate a candidate service task list based on the pointer network, and calculate the service task priority through the classifier.

[0067] In this embodiment, the Pointer Network is a sequence generation model based on the attention mechanism; the classifier (such as the Softmax classifier) is a machine learning model for calculating task priorities.

[0068] Specifically, the input parameters of the pointer network during training are: triples in the knowledge graph (disease, symptoms, treatment plan); the output parameters are: a list of candidate service tasks (such as "measure blood pressure" and "adjust insulin dosage"), and the loss function is cross-entropy loss and sequence length penalty.

[0069] For example, the priority calculation formula is: set the feature dimensions as task type (weight 0.4), urgency (weight 0.3), and patient history preference (weight 0.3); the classifier formula is: P priority =σ1(W c [h type , h urgency , h preference ]), where σ1 is the Sigmoid function, h * are the feature vectors of task type, urgency and patient history preference, W c is the corresponding weight coefficient.

[0070] S43: Use the improved Hungarian algorithm combined with task constraints to match the optimal caregiver and task combination. Task constraints include geographical distance, skill matching, and task urgency coefficient.

[0071] In this embodiment, the constraints are quantified, that is, the geographical distance, skill matching and task urgency coefficient are converted into calculable parameters. For example, the geographical distance is matched according to a score of 1 kilometer. The lower the score, the closer the geographical location. The skill matching is calculated as a percentage, and the task urgency coefficient is converted from 1 to 10.

[0072] S44: Pushing the service instruction through the preset terminal communication interface and obtaining the service work order confirmation instruction corresponding to the service instruction to achieve two-way confirmation of the service task.

[0073] In this embodiment, the preset terminal communication interface refers to an MQTT-based IoT messaging channel, a WeChat mini-program Push API, or the like. The service ticket confirmation instruction refers to the confirmation message received after the service instruction is sent to the caregiver. During two-way confirmation, if the confirmation mechanism fails three times and manual intervention is required, the timeout period is set to 10 minutes (the specific duration and number of confirmations can be customized).

[0074] In one embodiment, in step S2, the step of constructing a structured medical knowledge graph further includes: S201: Based on the patient's medical imaging data, a 3D convolutional neural network is used to extract lesion features, which are combined with the patient's electronic medical record text features to generate a multimodal fusion feature vector.

[0075] In this embodiment, a 3D convolutional neural network (3D CNN) is used to extract three-dimensional spatial features of medical images (such as CT and MRI); multimodal fusion refers to the integration of feature vectors of different modalities (such as images and text); and the BERT word segmenter is used to segment the text of electronic medical records and extract clinical terms (such as "pulmonary nodules" and "hypertension").

[0076] Specifically, image feature extraction methods include using a pre-trained 3D ResNet-50 model to extract the shape and texture features of the lesion area; text feature extraction methods include extracting keywords (such as "chest pain" and "diabetes") through the TF-IDF algorithm, generating bag-of-words vectors, and calculating the weights of each modality feature through the attention mechanism. Finally, the feature vectors are fused. The fusion formula is: fusion =β1×F img +β2×F text Among them, F fusion is the fused feature vector; β1 and β2 are the weights calculated by the attention mechanism; F img is the image feature; F textis the text feature. The effectiveness of the attention mechanism is to solve the difference in modal importance through dynamic weight allocation (β1, β2): Acute conditions (such as intracerebral hemorrhage) rely more on imaging features (increased β1); Chronic diseases (such as diabetes) are more dependent on text features (β2 increases).

[0077] S202: A federated learning framework is used to combine data from several nursing homes. The LSTM-CRF model embeds a differential privacy mechanism in local model training. The differential privacy mechanism sets a privacy budget threshold that sets the upper bound of the difference in output distributions between any two adjacent data sets.

[0078] In this embodiment, the privacy budget threshold ∈*=0.5. The differential privacy injection method for the LSTM-CRF model is to add Gaussian noise during gradient updates. The differential privacy injection formula is: in, is the gradient after adding noise; g is the original gradient; σ is the standard deviation of the noise; Δf is the sensitivity; δ is the probability of privacy leakage; ∈ is the privacy budget, and the corresponding privacy budget threshold upper limit ∈*=0.5.

[0079] Specifically, in order to achieve a balance between privacy protection and model performance, this application adds noise "N(0, σ 2 )” prevents individual data from being reversely deduced, satisfies (∈, δ)-differential privacy, and attackers cannot infer the existence of a certain record by observing the model output; and realizes adaptive adjustment of noise intensity, namely through sensitivity (Δf): the maximum change in gradient update (such as local data difference in federated learning); privacy budget (∈): controls the strength of privacy protection (the smaller ∈, the greater the noise, and the stronger the privacy protection); risk parameter (δ): the privacy upper limit (usually set to 10^-5).

[0080] S203: Dynamically assign weights of different modalities of multi-source medical and nursing data through a graph attention network to construct an enhanced medical knowledge graph with cross-modal association capabilities.

[0081] In this embodiment, the JanusGraph graph database is used to store the enhanced medical knowledge graph.

[0082] Specifically, the steps of graph initialization include: Node types: patient, disease, symptom, imaging feature, drug.

[0083] Edge types: diagnostic association (disease → symptom), therapeutic association (drug → disease), and imaging association (imaging feature → disease).

[0084] Attention weight calculation: assign attention heads to each modality data: Where i is the node identifier; l is the layer number identifier in the multi-layer perceptron (MLP); represents the hidden state vector of node i in layer l; W (l) Represents the weight matrix, dimension D (l+1) ×D (l) , D (l+1) is the dimension of the hidden state of the next layer, D (l) is the dimension of the hidden state of the current layer (e.g., 256, 128), which is used to linearly transform node features and map them to a higher-dimensional space to capture complex relationships; Represents the attention weight, which indicates the degree of attention of the kth attention head to node i, and is normalized by Softmax to satisfy μ is the activation function (such as LeakyReLU, ReLU); K is the number of attention heads, and in this embodiment, K=8.

[0085] Examples of dynamic weights include: image feature weight: 0.4 (acute symptoms); text feature weight: 0.3 (chronic disease history); and structured data weight: 0.3 (laboratory indicators).

[0086] In one embodiment, in step S3, dynamically optimizing the preset personalized medical plan and resource scheduling strategy and outputting the medical and nursing service optimization plan also includes: S301: Construct a causal graph of medical events based on the structural causal model and identify key intervention nodes.

[0087] In this embodiment, a structural causal model (SCM) refers to a mathematical model that describes the causal relationship between variables through a directed acyclic graph (DAG); a key intervention node refers to an intervention point that plays a decisive role in the development of a medical event (such as drug use, timing of surgery).

[0088] Specifically, the medical event causal graph uses the PC algorithm (Peter-Clark algorithm) to infer causal relationships from the data when constructing it (significance level α=0.05).

[0089] Example of a cause-effect diagram: Identification of key intervention nodes: Calculate the node intervention effect through do-calculus and screen the nodes (such as drug A dose adjustment) that have the greatest impact on endpoint events (such as heart failure).

[0090] S302: Use virtual medical twins to simulate the impact of different resource scheduling schemes on the health status of the elderly population and generate multi-dimensional evaluation indicators, including emergency response delay and complication rate.

[0091] In this embodiment, the dimensional evaluation index is a composite index that quantifies the resource scheduling effect (such as emergency response time and complication rate).

[0092] Specifically, AnyLogic or OpenModelica is used to establish a physiological model that includes the cardiovascular system and metabolic modules, that is, to implement the modeling operation of the twin model. For example, the twin of patient A includes a heart pumping function model (Windkessel model) and a blood glucose metabolism model (glucose-insulin dynamics model).

[0093] The simulation scenario design includes: Scenario 1: Usual care (blood glucose monitoring every 4 hours, daily insulin injections).

[0094] Scenario 2: Optimize scheduling (dynamically adjust monitoring frequency and prioritize nocturnal hypoglycemia events).

[0095] Output indicators: Emergency response delay: the time from abnormality detection to the arrival of medical personnel (simulation results: scenario 1 = 45 minutes, scenario 2 = 18 minutes).

[0096] Complication rate: Hypoglycemic event rate (scenario 1 = 12%, scenario 2 = 4%).

[0097] S303: Predicting potential risks of the optimal resource scheduling strategy through a counterfactual reasoning algorithm to obtain resource scheduling prediction risk information.

[0098] In this embodiment, counterfactual reasoning refers to evaluating strategy risks by assuming “what would happen if the intervention measures were different”; potential risks refer to the negative consequences that may be caused by resource scheduling (such as excessive examinations causing patient anxiety).

[0099] Specifically, DAG-GNN (Causal Graph Neural Network) is used to generate counterfactual samples (refer to the paper "Causal Effect Inference with Deep Latent-Variable Models"). The input parameters are: the original medical event sequence and the resource scheduling strategy; the output parameter is: the counterfactual event chain (such as "If the insulin dose is reduced → the risk of hypoglycemia increases by 20%"). Then, risk assessment is performed, for example: The formula for calculating risk probability is: P(risk|strategy) = total number of counterfactual samples / number of risk occurrences in the counterfactual event chain, combined with the risk threshold setting: if P(risk)>0.15, it is marked as a high-risk strategy.

[0100] S304: Dynamically adjust the health benefit weight coefficient in the Q-learning algorithm based on the resource scheduling prediction risk information.

[0101] Specifically, the health benefit weight refers to a quantitative indicator that measures the improvement of patient health caused by resource scheduling; the state space vector includes the patient's physiological indicators (such as blood sugar, blood pressure), resource utilization (such as the number of caregivers), and time window (such as the past 24 hours). For example, st = [blood sugar t ,blood pressure t , nursing staff vacancy rate t ,time t ].

[0102] The actions in the action space and reward function are resource scheduling decisions (e.g., “dispatch nurse A to ward 2”, “cancel nighttime blood sugar monitoring”); the reward function is: R(s,a)=ω h ×H(s′)-ω c ×C(a)-ω r ×R(a); where ω h 、ω c 、ω r is the weight coefficient; H(s′) represents the health benefit (such as the reduction in the incidence of complications); C(a) represents the resource consumption cost; R(a) represents the risk coefficient (based on the result of step S303); adjust ω according to the risk prediction result r (If the risk is high r ×2).

[0103] Using the ε-greedy strategy to balance exploration and exploitation (ε = 0.1), the iterative update formula for strategy optimization is:

[0104] Q(s, a) represents the expected total reward (i.e., Q value) of taking action a in state s, state s (for example, the patient's blood sugar is 15mmol / L, caregiver A is idle, and caregiver B is handling other tasks), and action a (dispatching caregiver A to inject insulin for the patient); α X is the learning rate, which controls the update amplitude of the old Q value from the new experience, and its value range is [0, 1]. X ≈1: Fully trust new experiences and quickly update Q values (suitable for the initial exploration phase); α X ≈0: Almost ignore new experience and update conservatively (suitable for the stable period of strategy); r represents the immediate reward, which is the reward obtained immediately after executing action a. The reward can be positive or negative; γ represents the discount factor, which balances the weight of immediate reward and future reward. The value range is [0, 1]; Indicates: In the next state s ′ In the example, all possible actions a ′The maximum Q value represents the prediction of the optimal future path and guides the current action selection. The termination condition is that the average reward change rate for 10 consecutive episodes is less than 1%. Through continuous trial-and-error optimization of the Q value, the intelligent agent (such as the scheduling system) learns to select the optimal action a in state s.

[0105] It should be understood that the serial numbers of the steps in the above embodiments do not imply the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0106] In one embodiment, an AI-based integrated medical and nursing data management system is provided, which corresponds to the AI-based integrated medical and nursing data management method in the above embodiment.

[0107] The AI-based integrated medical and nursing data management system includes a data acquisition module, a data processing module, a classification and labeling module, a digital twin module, a solution optimization module, and a service list generation and allocation module. Detailed descriptions of each functional module are as follows: Data collection module, which collects multi-source medical and nursing data based on AI technology; The data processing module is used to preprocess and standardize the collected multi-source medical and nursing data to generate a standardized medical data set. The classification and annotation module uses a preset medical ontology to classify and semantically annotate the data in the standardized medical data set, and construct a structured medical knowledge graph containing patient feature vectors and medical event time series chains. The digital twin module uses a digital twin model to perform virtual mapping based on a structured medical knowledge graph to form a virtual medical twin; The solution optimization module combines real-time monitoring data with historical patient flow data, extracts multi-dimensional features, and identifies elderly care patterns to dynamically optimize preset personalized medical plans and resource scheduling strategies, ultimately outputting optimized medical and elderly care service plans. The service list generation and allocation module is used to generate a target service list based on the medical and nursing service optimization plan and the parsed doctor-patient conversation text using the attention mechanism, and then allocate the target service list to the preset elderly care service collaborative terminal.

[0108] For the specific limitations of the AI-based integrated medical and nursing data management system, please refer to the limitations of the AI-based integrated medical and nursing data management method above, which will not be repeated here; the various modules in the above-mentioned AI-based integrated medical and nursing data management system can be implemented in whole or in part through software, hardware and their combination; the above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above modules.

[0109] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: S1: Collect multi-source medical and nursing data based on AI, perform data preprocessing and data standardization on the multi-source medical and nursing data, and obtain a standardized medical data set.

[0110] S2: Use the preset medical ontology to classify and semantically annotate the data of the standardized medical data set, and construct a structured medical knowledge graph that includes patient feature vectors and medical event time series chains; use the digital twin model to perform virtual mapping based on the structured medical knowledge graph to form a virtual medical twin.

[0111] S3: Based on virtual medical twins, combined with real-time monitoring data and historical patient flow data, through multi-dimensional feature extraction and elderly care pattern recognition, it dynamically optimizes preset personalized medical plans and resource scheduling strategies, and outputs medical and elderly care service optimization plans.

[0112] S4: Based on the medical and nursing service optimization plan and the parsed doctor-patient conversation text, the attention mechanism is used to generate a target service list and distribute it to the preset elderly care service collaborative terminal.

[0113] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0114] In one embodiment, in particular, according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the AI-based medical and nursing integrated data management method. In such an embodiment, the computer program can be downloaded and installed from a network through a communication module, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), the various functions defined in the present invention are performed.

[0115] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0116] The above-described 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, it should be understood by those skilled in the art that the technical solutions described in the aforementioned embodiments may still be modified, or some of the features thereof may be replaced by equivalents. 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 various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. The AI-based integrated medical and nursing data management method is characterized by: include: Collect multi-source medical and nursing data based on AI, perform data preprocessing and data standardization on the multi-source medical and nursing data, and obtain a standardized medical data set; Using a preset medical ontology to classify and semantically annotate the data in the standardized medical dataset, a structured medical knowledge graph containing patient feature vectors and a time series chain of medical events is constructed; a digital twin model is used to perform virtual mapping based on the structured medical knowledge graph to form a virtual medical twin; Based on the virtual medical twin, real-time monitoring data and historical patient flow data are combined to dynamically optimize preset personalized medical plans and resource scheduling strategies through multi-dimensional feature extraction and elderly care model identification, and output medical and elderly care service optimization plans; According to the medical and nursing service optimization plan, combined with the parsed doctor-patient conversation text, the attention mechanism is used to generate a target service list and distribute it to the preset elderly care service collaborative terminal.

2. The AI-based integrated medical and nursing data management method according to claim 1 is characterized in that: The multi-source medical and nursing data includes paper health records, physiological signals and medical data collected by wearable devices; the data preprocessing includes: Using optical character recognition, we perform image binarization, noise filtering, and character segmentation on paper health records. We also use a medical term dictionary to perform entity recognition and convert unstructured text into standardized medical terminology. Perform Kalman filtering to denoise the physiological signals collected by wearable devices, use standardized methods to eliminate dimensional differences, and unify the data format in combination with medical device interface protocols; Based on the time series alignment algorithm, discrete medical data and continuous physiological monitoring data are temporally and spatially matched to build a data chain for the patient's entire life cycle; The isolation forest algorithm is used to detect abnormal data points in the multi-source medical and nursing data, and the sliding window method is used to fill the missing values of the sensor data.

3. The AI-based integrated medical and nursing data management method according to claim 1 is characterized in that: The steps of constructing a structured medical knowledge graph include: Using an integrated medical semantic network and clinical medical terminology, we perform multi-granular entity annotation on patient medical history, medication records, and lifestyle data to generate annotated datasets. The time series features of medical events are extracted based on the LSTM-CRF model, and a triple relationship network of disease-symptom-treatment plan is constructed through knowledge graph learning method. Introducing an attention mechanism to calculate the time weights of different medical events and dynamically generate the patient's health status vector; Based on the annotated data set and the triple relationship network, a structured medical knowledge graph including the patient health status vector is constructed, and the structured medical knowledge graph is stored in a graph database.

4. The AI-based integrated medical and nursing data management method according to claim 1 is characterized in that: The dynamic optimization presets personalized medical plans and resource scheduling strategies, and outputs medical and nursing service optimization plans, including: Based on the federated learning framework, we jointly analyze the patient data within the institution and regional medical big data to train a personalized medical plan recommendation model; Using the Q-learning algorithm to optimize resource scheduling strategies based on preset constraints, including the nursing skills matrix, equipment utilization threshold, and service response time; Use Monte Carlo tree search to simulate multi-scenario service paths and select the resource allocation solution with the lowest overall cost; Based on the trained personalized medical plan recommendation model, optimized resource scheduling strategy and selected resource allocation plan, the model parameters are continuously optimized using AI-driven reinforcement learning to obtain an optimized medical and nursing service plan that adapts to the health status migration rules of the elderly population.

5. The AI-based integrated medical and nursing data management method according to claim 1 or 3 is characterized in that: The target service list generation and allocation steps include: Use natural language processing technology to analyze doctor-patient conversation texts and extract key medical needs based on domain knowledge graphs; Generate a list of candidate service tasks based on the pointer network and calculate the service task priority through the classifier; The improved Hungarian algorithm is used to match the optimal caregiver and task combination in combination with task constraints, including geographical distance, skill matching, and task urgency coefficient. The service instruction is pushed through the preset terminal communication interface, and the service work order confirmation instruction corresponding to the service instruction is obtained to achieve two-way confirmation of the service task.

6. The AI-based integrated medical and nursing data management method according to claim 3 is characterized in that: The step of constructing a structured medical knowledge graph further includes: Based on the patient's medical imaging data, a 3D convolutional neural network is used to extract lesion features, which are then combined with the patient's electronic medical record text features to generate a multimodal fusion feature vector. A federated learning framework is used to combine data from several nursing homes. The LSTM-CRF model embeds a differential privacy mechanism in local model training. The differential privacy mechanism sets a privacy budget threshold that is the upper bound of the difference in output distributions between any two adjacent datasets. The weights of different modal data of the multi-source medical and nursing data are dynamically allocated through a graph attention network to construct an enhanced medical knowledge graph with cross-modal association capabilities.

7. The AI-based integrated medical and nursing data management method according to claim 4 is characterized in that: The dynamic optimization of preset personalized medical plans and resource scheduling strategies and output of medical and nursing service optimization plans also includes: Construct a causal graph of medical events based on a structural causal model to identify key intervention nodes; Using the virtual medical twin to simulate the impact of different resource scheduling schemes on the health status of the elderly population, generating multidimensional evaluation indicators, including emergency response delay and complication rate; Predict the potential risks of the optimal resource scheduling strategy through counterfactual reasoning algorithm to obtain resource scheduling prediction risk information; The health benefit weight coefficient in the Q-learning algorithm is dynamically adjusted based on the resource scheduling prediction risk information.

8. AI-based integrated medical and nursing data management system, characterized by: A system for executing the AI-based integrated medical and nursing data management method according to any one of claims 1 to 7, comprising: Data collection module, which collects multi-source medical and nursing data based on AI technology; The data processing module is used to preprocess and standardize the collected multi-source medical and nursing data to generate a standardized medical data set; A classification and annotation module uses a preset medical ontology to classify and semantically annotate the data in the standardized medical dataset, and constructs a structured medical knowledge graph containing patient feature vectors and a time series chain of medical events; A digital twin module, which uses a digital twin model to perform virtual mapping based on the structured medical knowledge graph to form a virtual medical twin; The solution optimization module combines real-time monitoring data with historical patient flow data, extracts multi-dimensional features, and identifies elderly care patterns to dynamically optimize preset personalized medical plans and resource scheduling strategies, ultimately outputting optimized medical and elderly care service plans. A service list generation and allocation module is used to generate a target service list based on the medical and nursing service optimization plan and the parsed doctor-patient conversation text using an attention mechanism, and allocate the target service list to a preset elderly care service collaborative terminal.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the AI-based integrated medical and nursing data management method as described in any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the AI-based integrated medical and nursing data management method as described in any one of claims 1 to 7 are implemented.

Images