Orthopedic patient-oriented nursing assistance scheme generation method and system

Abstract

The invention relates to the technical field of intelligent medical treatment and personalized nursing assistance, and particularly discloses a method and a system for generating a nursing assistance scheme for orthopedic patients. The method comprises the following steps: acquiring multi-modal data such as an electronic medical record, a medical image and rehabilitation monitoring, and constructing a semantic structure body; extracting a key rehabilitation state by using a graph neural network and time sequence modeling and generating a label graph; semantic matching and limiting condition extraction are carried out in combination with a nursing target library, and a target-constraint pair set is constructed; reasoning a nursing path based on a path generation model, and performing scoring modeling and weight updating in combination with execution feedback; and finally, cross-patient optimization and closed-loop semantic updating of the path model are realized. According to the method, the individuation and dynamic adaptive capacity of nursing path generation can be improved, and the intelligent level of nursing aid decision making is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical information processing and intelligent auxiliary decision-making, and in particular to a method and system for generating a nursing assistance plan for orthopedic patients. Background Art

[0002] With the continuous advancement of medical information technology and the development of rehabilitation medicine, developing efficient and accurate personalized nursing assistance plans for orthopedic patients has become a research priority in the field of clinical rehabilitation. Existing nursing plans often rely on manual experience and static templates, failing to fully integrate multimodal information such as patients' electronic medical records, medical imaging, and rehabilitation monitoring data. They also lack dynamic perception and intelligent response capabilities for key recovery stages, behavioral risk factors, and pathway execution feedback. This results in poor adaptability of nursing pathways, long update cycles, and difficulty migrating strategies.

[0003] In recent years, graph neural networks and semantic modeling technologies have demonstrated significant advantages in scenarios such as medical knowledge graphs, intelligent recommendations, and pathway generation. However, their integration and practical application in dynamic planning of orthopedic nursing pathways and cross-patient strategy optimization remain gaps. Furthermore, how to structure and model feedback data from pathway execution and then feed it back into the pathway generation model for adaptive optimization is a key issue that traditional nursing systems struggle to address. Therefore, an innovative technical approach that integrates semantic structure modeling, pathway scoring feedback, and a strategy migration mechanism is urgently needed to enable the intelligent generation and continuous evolution of nursing assistance solutions. Summary of the Invention

[0004] The present invention provides a method and system for generating nursing assistance plans for orthopedic patients, so as to solve the problem of how to construct a data semantic structure based on the multimodal diagnosis and treatment data of orthopedic patients, and realize the dynamic generation of personalized nursing pathways and cross-patient strategy optimization through graph neural networks and path scoring feedback mechanisms, thereby improving the intelligence level and adaptability of nursing assistance decision-making.

[0005] In order to solve the above technical problems, the present invention provides a method for generating a nursing assistance plan for orthopedic patients, comprising:

[0006] Obtain the patient's electronic medical record data, medical imaging data, and rehabilitation monitoring data, unify the format and remove noise, generate a nested structure, and then construct a data semantic structure to form an initial graph node set;

[0007] Performing graph neural network modeling and temporal state modeling on the data semantic structure, extracting key recovery stages and behavioral risk factors, constructing a rehabilitation state label graph, and performing semantic classification and multi-dimensional label assignment to form a structured label set;

[0008] According to the rehabilitation status label graph and the structured label set, semantic matching with the nursing goal knowledge base and extraction of rehabilitation restriction conditions are performed, the goal-constraint mapping relationship is screened and encoded, and a goal-constraint pair set is generated;

[0009] Obtain the target-constraint pair set, perform phased nursing pathway reasoning, serialization processing, and intervention priority label embedding to construct a sorted nursing pathway sequence;

[0010] Modeling the rehabilitation feedback data during the execution of the ordered nursing pathway sequence, using the modeling results as input for association analysis and scoring modeling, generating a pathway scoring sequence for weight adjustment of pathway nodes, and forming a pathway structure with updated weights;

[0011] Based on the pathway scoring sequence and the pathway structure after weight update, the pathway generation model parameter weights are updated, cross-patient pathway strategy learning is performed, and the data semantic structure and rehabilitation status label map are synchronously updated.

[0012] Furthermore, the steps of obtaining the patient's electronic medical record data, medical imaging data, and rehabilitation monitoring data, performing format unification and noise removal processing, generating a nested structure, and constructing a data semantic structure to form an initial graph node set include:

[0013] Obtain the patient's electronic medical record data, medical imaging data, and rehabilitation monitoring data, unify the format, and remove noise to obtain multimodal structured data;

[0014] Performing feature alignment and modality nesting modeling on the multimodal structured data to generate a nested structure in a unified format;

[0015] A data semantic structure is constructed based on the nested structure, and an initial graph node set is formed.

[0016] Furthermore, the steps of performing graph neural network modeling and temporal state modeling on the data semantic structure, extracting key recovery stages and behavioral risk factors, constructing a rehabilitation state label graph, and performing semantic classification and multi-dimensional label assignment to form a structured label set include:

[0017] Performing graph neural network modeling and time series state modeling on the data semantic structure to generate a rehabilitation status prediction result;

[0018] Extracting key recovery stages and behavioral risk factors from the recovery status prediction results to construct a recovery status label map;

[0019] The rehabilitation status label graph is semantically classified and multi-dimensionally labeled to form a structured label set.

[0020] Furthermore, the steps of performing semantic matching with a nursing goal knowledge base and extracting rehabilitation restriction conditions based on the rehabilitation status label graph and the structured label set include:

[0021] Obtaining a goal representation in a nursing goal knowledge base and performing semantic matching with the rehabilitation status label graph;

[0022] Based on the semantic matching results, rehabilitation constraints related to the goal are extracted from the structured tag set to construct a goal-constraint mapping relationship.

[0023] Furthermore, the step of screening and encoding the target-constraint mapping relationship to generate a target-constraint pair set includes:

[0024] The target-constraint mapping relationship is screened and encoded to generate a target-constraint pair set as input for generating the nursing pathway.

[0025] Furthermore, the steps of obtaining the target-constraint pair set, performing staged nursing pathway reasoning, serialization processing, and intervention priority label embedding include:

[0026] Obtaining the target-constraint pair set and inputting it into a pathway generation model to perform phased nursing pathway reasoning;

[0027] Serializing the nursing pathway reasoning results and setting intervention priority labels in combination with risk factors;

[0028] The intervention priority labels are embedded into the pathway candidate sequence to generate a ranked care pathway sequence.

[0029] Furthermore, the steps of modeling the rehabilitation feedback data during the execution of the sorted nursing pathway sequence and using the modeling results as input for association analysis and scoring modeling include:

[0030] Obtaining rehabilitation feedback data during the execution of the sequenced nursing pathway, and modeling behavioral trajectories and subjective indicators;

[0031] Conduct correlation analysis and score modeling on the modeled behavioral trajectory and subjective indicator results to generate a path score sequence;

[0032] The weights of the path nodes are adjusted based on the path scoring sequence to form a path structure with updated weights.

[0033] Furthermore, the step of updating the pathway generation model parameter weights based on the pathway score sequence and the pathway structure after weight update and performing cross-patient pathway strategy learning includes:

[0034] Obtaining the path scoring sequence and the path structure after weight update, and updating the parameter weights of the path generation model;

[0035] The updated pathway generation model is embedded in the knowledge transfer framework for cross-patient pathway strategy learning.

[0036] Furthermore, the step of synchronously updating the data semantic structure and the rehabilitation status label map includes:

[0037] Based on the cross-patient pathway strategy learning results, the data semantic structure and the rehabilitation status label graph are synchronously updated to complete the closed-loop update of the semantic structure.

[0038] Furthermore, a system for generating a nursing assistance plan for orthopedic patients is characterized by comprising:

[0039] The data processing module is used to obtain the patient's electronic medical record data, medical imaging data, and rehabilitation monitoring data, unify the format and remove noise, generate a nested structure, and then construct a data semantic structure to form an initial graph node set;

[0040] A state recognition module is used to perform graph neural network modeling and temporal state modeling on the data semantic structure, extract key recovery stages and behavioral risk factors, construct a rehabilitation state label map, and perform semantic classification and multi-dimensional label assignment;

[0041] A target matching module is used to perform semantic matching with the nursing target knowledge base and extract rehabilitation constraint conditions based on the rehabilitation status label graph and the structured label set, screen and encode the target-constraint mapping relationship, and generate a target-constraint pair set;

[0042] A pathway generation module is used to obtain the set of goal-constraint pairs, perform phased nursing pathway reasoning, serialization processing, and intervention priority label embedding to construct a sequence of ranked nursing pathways;

[0043] A feedback modeling module is used to model the rehabilitation feedback data during the execution of the sequenced nursing pathway, use the modeling results as input to perform association analysis and scoring modeling, generate a pathway scoring sequence, adjust the weights of pathway nodes, and form a pathway structure with updated weights;

[0044] The optimization update module is used to update the path generation model parameter weights based on the path scoring sequence and the path structure after the updated weights, perform cross-patient path strategy learning, and synchronously update the data semantic structure and rehabilitation status label map.

[0045] The following are its main beneficial effects:

[0046] (1) The present invention constructs a multimodal semantic structure that integrates electronic medical records, medical images, and rehabilitation monitoring data. It adopts a unified format and nested modeling approach to achieve standardized and structured expression of heterogeneous medical data, effectively solving the problems of weak data comprehension and low information integration in traditional nursing pathway design.

[0047] (2) The present invention jointly constructs a rehabilitation status prediction map through graph neural network and time series modeling, and generates a semantic label structure, which realizes the accurate identification of the patient's recovery stage and the automatic extraction of behavioral risk factors. Compared with the method relying on manual evaluation, the prediction accuracy is higher and the label classification is more detailed, which significantly enhances the data support capability of nursing decision-making.

[0048] (3) The present invention introduces a path feedback scoring modeling mechanism and a weight adjustment strategy to construct a closed-loop structure of "scoring-pathway-model". It combines the cross-patient transfer learning framework to dynamically optimize the parameters of the path generation model, solving the problem of static configuration of traditional nursing plans and difficulty in adapting to individual differences of patients, and realizing adaptive iterative updates of the path plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A flowchart of a method for generating a nursing assistance plan for orthopedic patients provided in an embodiment of the present application;

[0050] Figure 2 This is a structural block diagram of a system for generating a nursing assistance plan for orthopedic patients provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] Example 1: Reference Figure 1 , is a flow chart of a method for generating a nursing assistance plan for orthopedic patients provided by an embodiment of the present invention. The process may include at least steps S100-S600:

[0052] S100, obtain the patient's electronic medical record data, medical imaging data and rehabilitation monitoring data, perform format unification and noise removal processing, generate a nested structure and then construct a data semantic structure to form an initial graph node set.

[0053] S200. Perform graph neural network modeling and temporal state modeling on the data semantic structure, extract key recovery stages and behavioral risk factors, construct a rehabilitation state label graph, and perform semantic classification and multi-dimensional label assignment to form a structured label set.

[0054] S300 , performing semantic matching with a nursing goal knowledge base and extracting rehabilitation restriction conditions based on the rehabilitation status label graph and the structured label set, screening and encoding goal-constraint mapping relationships, and generating a goal-constraint pair set.

[0055] S400 , obtaining the target-constraint pair set, performing staged nursing pathway reasoning, serialization processing, and intervention priority label embedding, and constructing a sorted nursing pathway sequence.

[0056] S500: Model the rehabilitation feedback data during the execution of the sorted nursing pathway sequence, use the modeling results as input for association analysis and scoring modeling, generate a pathway scoring sequence, adjust the weights of pathway nodes, and form a pathway structure with updated weights.

[0057] S600: Based on the pathway scoring sequence and the pathway structure after weight update, update the pathway generation model parameter weights, perform cross-patient pathway strategy learning, and synchronously update the data semantic structure and rehabilitation status label map.

[0058] Step S100 at least includes steps S110-S130:

[0059] S110: Obtain the patient's electronic medical record data, medical imaging data, and rehabilitation monitoring data, perform format unification and noise removal processing, and obtain multimodal structured data.

[0060] In this step, the system first performs unified collection and standardization of multi-source heterogeneous raw data, specifically including the acquisition of the following three data subsets:

[0061] The first step is to obtain the patient's electronic medical record data, which comes from the clinical information system and includes medical records, postoperative observation records, attending physician evaluation summaries, medication regimens, and medical history. This data may be in the form of structured tables, semi-structured forms, or unstructured free text. The system must uniformly extract core fields and construct a field comparison table based on the term ontology mapping table.

[0062] The second step is to acquire the patient's medical imaging data, including preoperative CT and MRI data and postoperative follow-up imaging data, from the PACS system. Due to differences in imaging sources and formats, the system must standardize image formats (for example, to DICOM 3.0 or NIfTI standards) and perform spatial registration of image sequences to address the issue of insufficient image alignment accuracy between different acquisition devices and at different time points.

[0063] The third step is to obtain patient rehabilitation monitoring data, including gait data collected by wearable devices, joint range of motion measurements, movement frequency records, heart rate change sequences, and behavioral trajectory data recorded by ward monitoring systems. This type of data has time series characteristics and environmental sensitivity, so the system needs to standardize sampling rates and timestamp alignment formats, and correct for sensor accuracy across different source devices.

[0064] After completing the collection of the above three types of data, the system performs format unification operations on the electronic medical record data, medical imaging data and rehabilitation monitoring data, specifically including: using a data standardization engine to clean the terms, units and formats in the text data; using a gradient edge preservation-based noise reduction algorithm to remove artifacts and scanning afterimages from the image data; performing linear interpolation and wavelet filtering on the time series data to remove interference bands during the collection process, ensuring the temporal continuity of physiological indicators and movement trajectories.

[0065] Finally, the cleaned and formatted data is integrated into a multimodal structured dataset. This multimodal structured data serves as the input for modal nesting modeling and alignment in S120. It must maintain index consistency, time synchronization, and identification uniformity across modalities, providing data support for the subsequent generation of a unified format nested structure.

[0066] S120: Perform feature alignment and modal nesting modeling on the multimodal structured data to generate a nested structure in a unified format.

[0067] After standardizing the multimodal structured data, the system enters the feature alignment and nested modeling phase. This step uses the multimodal structured data as input and performs semantic, temporal, and spatial multi-dimensional alignment operations to generate a nested structure in a unified format, laying the foundation for the subsequent construction of the data semantic structure.

[0068] Specifically, a semantic alignment operation is performed first. Based on the constructed medical terminology knowledge graph and semantic embedding model, the system performs semantic mapping on the text fields in the electronic medical record data and the behavioral descriptions in the rehabilitation monitoring data. The system introduces a multilingual embedding model and an ontology comparison table to uniformly map terms such as "lower limb weight-bearing training" and "knee extension exercise" to standardized codes. For the annotated areas in the medical imaging data (such as fracture lines, implant positions, etc.), they are converted into semantically labeled nested nodes in combination with medical image segmentation labels to ensure that the three types of modal data are consistent at the semantic level.

[0069] Secondly, the system performs time alignment. Because the rehabilitation monitoring data exhibits distinct time series characteristics, the system must synchronize it with the time points marked in the electronic medical record for the postoperative rehabilitation phase. Furthermore, postoperative imaging data and monitoring indicators must be aligned to a daily or hourly timeline. The system constructs a unified time index structure that maps the acquisition time of imaging data to the acquisition time of behavioral data, using a unified timestamp as the anchor point for nested alignment.

[0070] Next, the system performs spatial alignment. The spatial coordinate system in the medical image must be mapped to the spatial dimensions of the motion trajectory in rehabilitation monitoring. Using registration point conversion technology, the system maps the behavioral path from the spatial coordinate system to the image space, forming a nested association between motion and anatomical structure. For example, a one-to-one spatial hierarchical relationship is established between the gait trajectory and the lower limb skeletal model for nested modeling.

[0071] After completing the three alignments described above, the system uses a nested structure modeling module to construct a hierarchical nested model of the electronic medical record data, medical imaging data, and rehabilitation monitoring data based on semantic, temporal, and spatial layers. This generated nested structure uses a graph-based data organization format, with the patient as the root node and child nodes such as diagnostic information, time nodes, rehabilitation action sequences, and associated anatomical regions nested sequentially, forming a unified, multimodal nested structure.

[0072] This unified nested structure serves as the core input for S130 to construct the data semantic structure. Its node attributes, structural relationships, and nesting depth will directly affect the accuracy of subsequent graph neural network modeling and label reasoning.

[0073] S130. Construct a data semantic structure based on the nested structure and form an initial graph node set.

[0074] In this step, the system performs a data semantic structure construction operation based on the nested structure in the unified format, with the goal of generating graph structure information for subsequent state recognition and label generation, and outputting an initial graph node set to provide input support for S200.

[0075] Specifically, the system first parses the nested structure into a set of graph nodes and a set of relationships between them. Each nested field is converted into a semantic node, which contains information such as its corresponding modal source, semantic label, time index, spatial anchor, and so on. For example, the information "knee extension recovered to 75%" on the third day after surgery will be broken down into four nodes: "action type," "time point," "body part," and "numerical label," and directed edges will be established within the semantic structure.

[0076] Furthermore, while generating node sets, the system also introduces node-level labels and feature identification fields to form the initial structure for graph neural network propagation. For text nodes, word vector features are introduced; for image nodes, image slice numbers and semantic segmentation labels are retained; and for time series nodes, normalized behavior frequency and change trend features are added.

[0077] The system then establishes an initial set of graph edges based on the graph's topology and performs structural corrections. Correction rules include: if two nodes originate from the same time point, a temporal parallel edge is established; if two nodes represent the same anatomical region, a spatial nested edge is established; and if two nodes share a related verb in the medical record description, a semantic causal edge is established. The resulting data semantic structure is a multi-edge relational graph, capable of multimodal label fusion, cross-modal semantic sharing, and a clearly structured graph neural architecture.

[0078] The initial graph node set formed by the data semantic structure is used as the input node of the graph neural network in S210 to realize rehabilitation status prediction; it is used as the basic input structure for extracting key recovery stages and behavioral risk factors in S220; it serves as the node source for semantic matching of nursing goals in the S300 module; and it receives feedback iterative results updates in the S600 module to realize graph closed-loop optimization.

[0079] This implementation step effectively realizes the semantic fusion and spatiotemporal alignment of electronic medical records, imaging data, and rehabilitation behavior data through the unified processing and nested structure construction of multimodal structured data. The constructed data semantic structure has complete semantic representation capabilities, graph network scalability, and label adaptation capabilities, providing a structurally consistent and attribute-unified input foundation for subsequent state prediction, target matching, and path generation, ensuring that the entire nursing assistance plan generation process has a high degree of information closure and evolution capabilities.

[0080] Step S200 at least includes steps S210-S230:

[0081] S210: Perform graph neural network modeling and time series state modeling on the data semantic structure to generate a rehabilitation status prediction result.

[0082] This step uses the data semantic structure constructed in S130 as the input data structure to perform graph neural network modeling and temporal state modeling operations. The goal is to generate a rehabilitation status prediction result with stage-specific indicative significance based on the graph topological relationship between semantic nodes and the temporal evolution trend of behavioral states.

[0083] Specifically, the system first initializes the graph neural structure of the graph nodes in the data semantic structure. Based on the node type (text, image, behavior, etc.), the system loads the corresponding node feature vector, constructs a set of node attributes, and establishes edge sets corresponding to the semantic, temporal, and spatial relationships between nodes.

[0084] The system then introduces graph convolutional units and time series modeling components to the node graph structure, constructing a multi-layer graph propagation network. The graph neural network component primarily leverages the adjacency relationships within the graph structure to transfer feature information between nodes, capturing the spatial topology within the structure. The time series state modeling component constructs evolution paths for time series nodes, combining historical node states with predicted time point locations to dynamically generate state evolution trends.

[0085] During this process, the system further introduces multiple prediction target vectors, including physiological state change trends, functional recovery stage status labels, behavioral intervention response predictions, etc., to assist in the training and output prediction of the supervised graph model.

[0086] Ultimately, the system outputs a recovery status prediction, specifically a predicted state probability distribution for each target tag node, a recommended recovery stage label, and a potential abnormal behavior risk score. These prediction outputs are directly transmitted to S220 for recovery stage extraction and risk factor identification.

[0087] S220: Extract key recovery stages and behavioral risk factors from the rehabilitation status prediction results, and construct a rehabilitation status label map.

[0088] This step is based on the rehabilitation status prediction results generated in S210, completes the extraction of rehabilitation stage labels and the identification of behavioral risk factors, and further constructs a rehabilitation status label graph to express the comprehensive status of each patient at different time nodes, anatomical locations and behavioral types.

[0089] First, the system statistically aggregates the predicted distributions of various nodes in the rehabilitation status prediction results. Nodes that exhibit significant change trends (such as knee range of motion and gait stability) are classified as "key recovery indicators" based on a set threshold. Based on the predicted change trends, the system then assigns these nodes to predefined rehabilitation phases (e.g., acute postoperative phase, stabilization phase, and enhanced rehabilitation phase).

[0090] Secondly, the system detects abnormal patterns in behavioral monitoring nodes. If a behavior (such as repetitive weight-bearing or asymmetrical posture) consistently deviates from normal behavior during the predicted state, the system identifies it as a potential behavioral risk factor. Combined with information from adjacent structural nodes, the system further analyzes the potential physical impact or postoperative risk.

[0091] In constructing the recovery status label graph, the system uses the output nodes of the graph neural network as a foundation to reconstruct a semantic graph structure centered on stage labels, risk factors, and node context. Nodes in this graph are connected by stage affiliation, risk mapping, and temporal evolution, forming a complete label graph structure.

[0092] The rehabilitation status label graph is not only the input structure of the semantic assignment in S230, but also serves as the basic semantic object set for target semantic matching and constraint extraction in S300, providing a status label reference for the generation of personalized nursing pathways.

[0093] S230 , performing semantic classification and multi-dimensional label assignment on the rehabilitation status label graph to form a structured label set.

[0094] In this step, the system uses the rehabilitation status label graph generated in S220 as input, performs semantic classification and multi-dimensional label assignment operations, classifies each node into a unified label system according to the defined rules, and provides structured label input for the subsequent path generation process.

[0095] Specifically, the system first semantically classifies each node in the rehabilitation status label graph. Based on the relationship between the label vocabulary and the ontology, the classification rules categorize nodes into four main categories: "Rehabilitation Stage Label," "Risk Behavior Label," "Target Recovery Indicator Label," or "Constraint Label," according to node attributes and contextual meaning. During the classification process, if a node attribute contains multiple semantic tendencies, the system uses a graph-structured context-weighted calculation to determine the primary category.

[0096] Next, the system assigns multi-dimensional labels to the classified nodes. In addition to the basic semantic label, each labeled node also includes attributes such as temporal position (e.g., day, week), anatomical location (e.g., left knee, lumbar spine), and status level (e.g., low risk, medium risk, high risk). Based on the preceding graph structure and the probability and trend information contained in the prediction results, the system sequentially completes each dimensional field to ensure uniform label dimensions and standardized structure.

[0097] Finally, the system will output the set of label nodes after semantic classification and multi-dimensional assignment as a structured label set, which will serve as the input label system for semantic matching with the nursing goal knowledge base in S300; at the same time, some restriction labels will also be extracted in S320 to constrain the nursing path generation process.

[0098] This step builds a graph neural network and time series modeling mechanism based on the data semantic structure to jointly identify rehabilitation status and predict behavioral risk trends. Furthermore, through semantic classification and multidimensional label assignment, a structured label set is generated, transforming the data structure into semantic labels. This provides complete, accurate, and clearly structured state input for subsequent pathway generation and feedback learning. The entire process builds a complete information chain from data structure to label semantics to pathway decision-making, enhancing the system's dynamic responsiveness in multidimensional intervention scenarios and the foundation for personalized regulation of nursing pathways.

[0099] Step S300 at least includes steps S310-S330:

[0100] S310: Obtain the target representation in the nursing target knowledge base, and perform semantic matching with the rehabilitation status label graph.

[0101] In this step, the system first calls upon the built-in nursing goal knowledge base, which records a set of standard nursing goals developed through multidisciplinary collaboration. This knowledge base is hierarchically stored by rehabilitation stage, anatomical region, functional indicators, and behavioral performance, providing a clear structure and facilitating subsequent semantic access. Nursing goal content includes, but is not limited to, semantically explicit representations such as "achieving 90° knee range of motion by the seventh day after surgery," "achieving intermediate gait stability," and "independent standing time greater than 15 minutes."

[0102] Specifically, the system extracts a goal representation from the nursing goal knowledge base. The goal representation consists of a goal semantic unit, a time window tag, a goal decomposition path, and an indicator hierarchy. The system normalizes the nursing goal representation into a structured goal vector, whose semantic structure is consistent with the structured tag set output in S230, thereby providing a unified semantic dimension for semantic matching.

[0103] After target representation standardization is completed, the system uses the rehabilitation state label graph as a reference input for semantic matching and performs target alignment. This semantic matching process calculates similarity based on the label type, position, state label, and stage index of the graph structure nodes, prioritizing matching state nodes and target representations that are in the same rehabilitation stage, in the same anatomical region, and have similar functional semantics.

[0104] The system also incorporates contextual constraints during the matching process. If a goal represents a phased goal, such as "increasing continuous walking time to 10 minutes," the system will retrace the historical nodes associated with "walking" in the rehabilitation state label graph. This ensures that the matching process is not limited to the static attributes of the current state node, but also includes its evolutionary path and trend labels, thereby improving the rationality of semantic matching.

[0105] The system ultimately outputs a target semantic matching result, which includes a target ID-label node ID pair and a corresponding matching confidence score. This result directly serves as the basic input for S320 to extract rehabilitation constraints and construct a context-oriented path between the target and the constraint.

[0106] S320 : Based on the semantic matching result, extract rehabilitation constraints related to the goal from the structured tag set, and construct a goal-constraint mapping relationship.

[0107] This step receives the semantic matching result generated in S310, further filters out the rehabilitation constraints associated with the matched target from the structured tag set output in S230, and constructs a structural mapping relationship between the target and the constraint.

[0108] Specifically, the system first parses the target-label pairs in the semantic matching results and uses the label node ID to retrieve the specific node content in the structured label set, including semantic categories (such as action labels, anatomical region labels, and stage labels), risk labels, and restriction prompt fields. The system then sets restriction extraction rules based on the functional goal type of the matching target (such as muscle strength improvement, stability improvement, and weight training).

[0109] For example, if the goal is to "improve lower limb weight-bearing capacity," the system will prioritize extracting labels such as "recent high pain index," "asymmetric gait," and "inadequate stability level" from the structured label set and consider them as constraints in achieving this goal. If a label node has causal or sequential edges with other state nodes in the rehabilitation state label graph, the system will also extract the upstream nodes pointed to by these edges as indirect constraints.

[0110] The system further constructs a target-constraint mapping. This mapping, centered around the target representation, connects to the extracted constraint nodes and labels each mapping pair with constraint strength levels, behavioral association dimensions, and anatomical dependency labels. The system also supports multi-target aggregate constraint processing. If multiple targets share the same constraint, a "target group-constraint group" association structure is established within the mapping structure, improving the efficiency of constraint merging during subsequent path generation.

[0111] The final output is a structured target-constraint mapping table, which serves as a direct input for S330 to perform mapping screening and encoding.

[0112] S330: Screen and encode the target-constraint mapping relationship to generate a target-constraint pair set.

[0113] In this step, the system receives the goal-constraint mapping table, performs multi-dimensional screening and semantic encoding processing, and generates a standardized set of goal-constraint pairs for subsequent path reasoning and intervention priority sorting.

[0114] First, the system filters the data in the target-constraint mapping relationship. The filtering rules include but are not limited to the following categories:

[0115] Exclusion screening based on constraint conflict degree: When a certain goal corresponds to multiple conflicting constraints, the system selects the constraint with a wider scope of effectiveness and lower state stability based on the constraint level and state evolution trend;

[0116] Priority screening based on behavioral risk level: If the risk label level of a constraint node is higher than the threshold, the system retains the constraint and marks it as a mandatory restriction type;

[0117] Consistent filtering based on target group logic: If multiple targets are mapped to the same constraint, the system retains the mapping relationship between the constraint and all related targets and sets the shared flag.

[0118] The system then performs goal-constraint pair encoding. Based on a unified goal-constraint expression grammar, the encoding process encodes the goal semantic information and its corresponding constraint content into a "goal identifier-constraint identifier" pair. This pair is then assigned a set of structural attributes, including priority, semantic distance, associated goal phase, and impact prediction time window.

[0119] The system removes duplicates, merges, and structures all goal-constraint pairs, ultimately generating a standardized set of goal-constraint pairs. This set serves as direct input to the S400 care pathway generation model, used to construct pathway nodes, assign priorities, and construct behavior restriction logic.

[0120] This step extracts the goal representation from the nursing goal knowledge base, completes semantic matching with the rehabilitation status label graph, further extracts rehabilitation constraints related to the goal, and constructs a standardized goal-constraint mapping structure. This not only ensures the semantic consistency and behavioral feasibility of goal planning, but also provides a set of input pairs with a clear structure and clear constraints for the path generation module, realizing a logical closed loop for path node construction under multidimensional factors, which is conducive to improving the accuracy of path reasoning and the efficiency of personalized adaptation of patient care plans.

[0121] Step S400 at least includes steps S410-S430:

[0122] S410: Obtain the target-constraint pair set and input it into a pathway generation model to perform phased nursing pathway reasoning.

[0123] In this step, the system receives and loads the set of goal-constraint pairs generated in S330 as the input data source for the path reasoning task, entering the staged nursing path generation process. Each goal-constraint pair in this set records information such as the target node ID, constraint node ID, staged goal level, constraint strength, and time window label, which serves as an important basis for constructing path nodes and edges.

[0124] Specifically, the system first performs a structured analysis of the set of goal-constraint pairs, identifying the semantic associations between goals and the logical mutual exclusion relationships between constraints. For example, if two goals are semantically close and the constraint nodes they depend on share a common constraint source, the system clusters them to form a staged care pathway node block.

[0125] The system then feeds this structured node block into a pathway generation model. This model integrates a triple mechanism: graph structure-driven, temporal sequence control, and state transition probability assessment. Its core is to construct a phased care pathway diagram, where each node corresponds to a care goal, and each edge represents the order or dependency between goals.

[0126] During path reasoning, the system relies on the following three types of input:

[0127] The order priority field in the target-constraint pair set is used to control the time sequence between path nodes;

[0128] The risk label field in the rehabilitation status label map is used to evaluate the risk level of the constraint status and affect the direction of path generation;

[0129] The stage labels and anatomical region information in the structured label set are used to plan path coverage and node density.

[0130] The path generation model iteratively constructs multiple candidate paths, gradually adding target nodes. At each step, it determines whether the current target satisfies constraints, conflicts with existing path segments, and evaluates the current path's contribution to overall recovery efficiency. The system then generates multiple candidate pathway graph structures and outputs the phased nursing pathway inference results as input to S420.

[0131] S420: Serialize the nursing pathway reasoning results and set intervention priority labels based on risk factors.

[0132] This step receives the nursing pathway inference result generated in S410, serializes the pathway structure, and sets the intervention priority label for each pathway node based on the risk factors and constraint severity levels.

[0133] Specifically, the system first sequentially expands the nodes of the phased care pathway diagram to generate a pathway node sequence. Each sequence represents a possible path, and each element in the sequence records the pathway node ID, the corresponding target identifier, the bound constraint label, and the phase label.

[0134] After the sequence is expanded, the system enters the priority label setting stage. The evaluation of intervention priority mainly relies on two core indicators:

[0135] Risk factor level: The behavioral risk factor label extracted from S220. If the state label on which a path node depends contains a high-risk behavior tag, the intervention priority of the node is set to high.

[0136] Constraint severity level: Rehabilitation restriction conditions extracted from S320. If a constraint node contains high-level constraint content such as "no weight bearing" or "limited joint movement", the system will prioritize scheduling the corresponding target to avoid delay risks.

[0137] Based on the above evaluation dimensions, the system labels each node in the path sequence with three intervention priority labels: "high", "medium" and "low", and structurally adjusts the order of nodes: high-priority targets are moved forward and low-priority targets are postponed, forming a reasonable time scheduling chain.

[0138] Finally, the system outputs a path sequence with priority labels as the input structure for embedding the path candidate sequence in S430.

[0139] S430: Embed the intervention priority label into the candidate pathway sequence to generate a sorted nursing pathway sequence.

[0140] In this step, the system embeds and integrates the nursing pathway sequence generated in S420 with the priority label to generate the final sorted nursing pathway sequence for subsequent modules to perform feedback modeling and task configuration.

[0141] Specifically, the system uses path nodes as units to construct a path candidate sequence structure. This structure records the following information fields:

[0142] Node target ID;

[0143] the matched rehabilitation stage;

[0144] The bound constraints and risk levels;

[0145] intervention priority labels;

[0146] Node temporal position.

[0147] The system sorts the paths according to the intervention priority labels. The sorting strategy is to prioritize high-risk goals, key phased goals, and severely constrained nodes. On this basis, the time window is logically adjusted to ensure that the path sequence minimizes conflict interference while ensuring rehabilitation efficiency.

[0148] At the same time, the system retains the target aggregation relationship output by the original path generation model during the sorting process. If two target nodes belong to the same stage cluster, their proximity structure will be kept intact to ensure stage continuity.

[0149] After sorting is completed, the system performs a validity check on the path sequence structure, including:

[0150] Whether there is a violation of the original constraint path direction due to priority reordering;

[0151] whether all targets are embedded in the sorting path;

[0152] Whether all constraints have been bound associatively.

[0153] After verification, the system outputs the final sorted nursing pathway sequence as the task input for behavioral feedback modeling and scoring analysis in S500.

[0154] This step is based on the set of goal-constraint pairs output by the previous module. It utilizes graph reasoning and semantic scheduling mechanisms to generate a phased nursing pathway under multiple objectives. Path priorities are set by combining behavioral risk factors and constraint levels. Priority labels are embedded in the path candidate structure to generate a sequence of ranked nursing pathways. This not only achieves a structured arrangement of rehabilitation tasks but also provides a clear and traceable task chain for subsequent feedback scoring and model optimization, facilitating dynamic evaluation of pathway execution and adaptive strategy updates.

[0155] Step S500 at least includes steps S510-S530:

[0156] S510: Obtain rehabilitation feedback data during the execution of the sorted nursing pathway sequence, and perform behavioral trajectory and subjective indicator modeling.

[0157] In this step, the system first indexes execution events according to the node time order from the sorted nursing pathway sequence generated in S430, and simultaneously collects multi-source feedback data, including:

[0158] Movement posture behavior data captured by medical images (such as limb bending angle);

[0159] Behavioral trajectory data (such as gait speed and movement distance) recorded in real time by wearable devices;

[0160] Subjective indicator data recorded by the rehabilitation monitoring platform (such as pain score, heart rate perception, RPE index);

[0161] Align with the historical recovery data of the electronic medical record obtained in S100.

[0162] On this basis, the system establishes a corresponding behavior-subjective joint feedback model for each path node. In order to improve the expression breadth and continuity of the modeling, the following integral model is constructed:

[0163] ① Path feedback functional modeling formula:

[0164]

[0165] in:

[0166] :Indicates the Path nodes in the execution cycle Feedback response functional within;

[0167] : Indicates time The behavior trajectory function at the moment is derived from the multimodal sensor behavior stream (such as movement amplitude and cadence curve);

[0168] : Indicates time The subjective perception index function at the moment is derived from the subjective feedback scale (such as pain value and fatigue);

[0169] : are the weight coefficients of behavioral and subjective signal channels (e.g. , , optimizing learning based on the correlation during historical recovery);

[0170] : No. The total execution time of each path node is derived from the intervention time window set in S430;

[0171] The integral lower limit 0 indicates that the path node starts to execute. The time for collecting feedback.

[0172] The introduction of this formula allows for continuous modeling of different time series feedback quantities, effectively capturing behavioral change trends and subjective evaluation fluctuations, and realizing data response modeling from discrete to continuous, forming a path execution feedback response functional. , laying the foundation for subsequent scoring.

[0173] S520: Perform correlation analysis and score modeling on the modeled behavior trajectory and subjective indicator results to generate a path score sequence.

[0174] In this step, the system generates the feedback response functional results generated in S510. Set the target value with the path node Perform error assessment and build the following scoring calculation model:

[0175] ②Path score distribution function formula:

[0176]

[0177] in:

[0178] :Indicates the The normalized score value of each path node, ranging from (0,1];

[0179] : Ideal response target values ​​set at the pathway nodes (e.g., gait amplitude should be ≥ 0.8 m, pain score should be ≤ 3 points);

[0180] : feedback response functional calculated by S510;

[0181] : The nonlinear adjustment index of the node score curve (for example: Indicates the use of quadratic distance, the larger the value, the more sensitive);

[0182] : Penalty adjustment factor, which controls the tolerance to target deviation (e.g., if the historical tolerance range is 10%, then );

[0183] : Exponential function, the closer the mapping score is to 1, the better the node execution effect is.

[0184] This function converts the "difference between feedback response and target value" into a score value. The exponential function has smoothing and compression characteristics, which makes the score more sensitive when the error is small, while avoiding score distortion caused by extreme feedback, ensuring the stability and interpretability of the scoring system.

[0185] Finally, the system will Construct a path scoring sequence:

[0186]

[0187] This path score sequence will serve as the input basis for weight update in S530.

[0188] S530: Adjust the weights of the path nodes based on the path scoring sequence to form a path structure with updated weights.

[0189] In this step, the system scores the sequence of paths Recalibrate the weight of each node in the sorted care pathway sequence.

[0190] Taking into account the different contributions of different nodes to the overall rehabilitation path, the update function based on the fuzzy weighting model and priority adjustment mechanism is constructed as follows:

[0191] ③ Path node weight update formula (fusion score and priority):

[0192]

[0193] in:

[0194] : Updated Normalized execution weight of path nodes;

[0195] : Generated by S520 Node score;

[0196] : The initial priority label value of the path node, which comes from the S420 embedded information (e.g. representing "high, medium, and low" respectively);

[0197] The denominator is the weighted sum of the scores of all path nodes, which is used for normalization so that the sum of the weights of all nodes is 1.

[0198] The model integrates two types of information: score value and path priority, ensuring that nodes with high scores and high priorities obtain greater execution weights, thus becoming more dominant in subsequent model optimization (S610).

[0199] The resulting new path structure consists of the following:

[0200] sequencing care pathway sequences;

[0201] Path node score value;

[0202] Weight adjustment results .

[0203] This structure will serve as an important input for subsequent path optimization and knowledge transfer model updates.

[0204] This module integrates three high-level mathematical mechanisms: functional integral modeling, exponential scoring mapping, and fuzzy weighted regulation. This completes the entire process from "pathway execution feedback" to "score value generation" and finally to "pathway weight reconstruction," enabling feedback-driven dynamic path evaluation. Compared to traditional static rule-based path planning methods, this solution significantly enhances the personalized adaptability, dynamic adjustment capabilities, and score interpretability of the rehabilitation pathway generation system, helping to optimize the efficiency of nursing intervention execution and the quality of rehabilitation process management.

[0205] Step S600 at least includes steps S610-S630:

[0206] S610: Acquire the path scoring sequence and the path structure after weight update, and update the parameter weights of the path generation model.

[0207] This step builds upon the previous step S530. The system first obtains the pathway score sequence and the updated weighted pathway structure data. The pathway score sequence is a quantitative representation constructed in S520 based on the behavioral trajectory and subjective indicator modeling results and the nursing goal feedback error. The updated weighted pathway structure is the structured pathway graph formed in S530 after reconstructing the pathway node weights based on the pathway score.

[0208] Specifically, the system aligns and calibrates the path score sequence with the updated weighted path structure, matching the differences between the score values ​​of each path node and the weights before and after the weight update to construct a node-level feedback error signal matrix. This error signal matrix serves as a gradient reference for updating the parameters of the path generation model. Through a feedback-guided mechanism, it adjusts the weights of multiple parameters in the model, such as node state transitions, priority sorting, and target constraint associations.

[0209] The path generation model, as understood, includes a function for processing multidimensional input features, a sequence learning module, and a path structure constraint encoding network. In this step, the system jointly optimizes the model's sequence state transition weights, the target constraint embedding layer parameters, and the fusion weights of the score perception module. This optimization utilizes a batch update approach, performing an iterative parameter update after each path generation feedback cycle.

[0210] After the update is completed, the system marks the current model status as "Optimized" and records the update timestamp and data source label for subsequent module calls and migrations.

[0211] S620: Embed the updated pathway generation model into the knowledge transfer framework to perform cross-patient pathway strategy learning.

[0212] After completing the model parameter update in S610, the system further embeds the updated pathway generation model as a learning sample into the preset knowledge transfer framework to carry out pathway strategy learning for cross-patient groups.

[0213] Specifically, the knowledge transfer framework consists of two main modules: a source model knowledge representation library and a set of migration mapping functions. The system first stores the updated path generation model as the source model in the knowledge representation library. It then labels the source patient based on their characteristics (such as orthopedic diagnosis code, age, gender, surgery type, and recovery period), constructing an index structure between patient characteristics and model parameters.

[0214] The system then constructs a target task representation vector based on the characteristics of the patient being evaluated. Using a transfer mapping function, it calculates a similarity index between the target patient and the historical source model. This similarity index, combined with node attributes in the semantic structure (such as the recovery stage label and risk factor level in the rehabilitation status label graph), serves as a reference for weighting the transfer function.

[0215] The system determines the set of source models participating in this pathway generation training based on similarity metrics and, through an aggregation mechanism, weightedly fuses the parameter vectors of multiple source models to generate a cross-patient pathway strategy parameter vector. This parameter vector serves as the initial input or fine-tuning baseline for the target patient's pathway generation task, improving the model's generalization and response efficiency.

[0216] S630: Based on the cross-patient pathway strategy learning results, the data semantic structure and the rehabilitation status label graph are synchronously updated to complete the closed-loop update of the semantic structure.

[0217] After completing the cross-patient strategy learning in S620, the system performs a synchronous update operation of the data semantic structure and the rehabilitation status label map based on the new path strategy output results and the updated model parameter set.

[0218] Specifically, the system first performs structural analysis on the updated path strategy, decomposing and mapping the path node sequence, goal constraint structure, and intervention labels into a semantic structure space. This mapping operation includes label-level semantic alignment (e.g., mapping the stage labels in the new path to the semantic structure node identifiers), structural-level path-node connectivity (e.g., updating the node connectivity relationships in the graph), and attribute-level scoring and feedback fusion (e.g., updating the average score value and achievement label of the node).

[0219] Next, the system generates a new rehabilitation status label graph based on the updated data semantic structure. This label graph structure inherits the label category system constructed in the previous step S230 and integrates the constraint hierarchy, target matching characteristics, and feedback perception factors in the current path generation strategy to enhance and refine the label content.

[0220] Finally, the system uses the updated data semantic structure and rehabilitation status label map as a new map input to replace the previous version, and simultaneously stores it in the semantic structure closed-loop management module, realizing the closed-loop iterative process of "perception-generation-feedback-optimization-migration-update" for the entire path generation and execution chain.

[0221] This implementation process, through the construction of a model parameter update mechanism, a knowledge transfer strategy, and a synchronous graph structure iteration mechanism, continuously optimized and generalized the path generation model tailored to the individual differences of orthopedic patients. This ensured the continuous adaptability, feedback responsiveness, and cross-cycle self-learning capabilities of personalized path planning, providing a structurally stable foundation for subsequent cyclic path execution and rehabilitation intervention optimization. This closed-loop technology path exhibits a high degree of structural integration and semantic adaptability among similar rehabilitation assistance systems, significantly improving the system's intelligent generation level and adaptability to application scenarios.

[0222] Example 2: Figure 2 FIG. 1 is a block diagram showing a system for generating a nursing assistance plan for orthopedic patients according to an embodiment of the present invention. Figure 2 As shown, the structure may include:

[0223] Data processing module 10 is used to obtain patients' electronic medical records, medical imaging data, and rehabilitation monitoring data, and then performs format unification and noise removal on this multi-source data to ensure structural consistency and usability across different data types. Furthermore, this module performs feature alignment and modal nesting modeling on the standardized data, generating a nested structure in a unified format. Based on this structure, a data semantic structure is constructed to form an initial graph node set, providing semantic support for subsequent label modeling and reasoning.

[0224] The state recognition module 20 is used to perform graph neural network modeling and temporal state modeling based on the data semantic structure output by the data processing module. This module extracts rehabilitation characteristics and potential risk factors at different stages, constructs a rehabilitation state label graph, and performs semantic classification and multi-dimensional label assignment. This module further outputs a structured label set, providing a rehabilitation semantic basis for the subsequent target matching and path construction modules.

[0225] The goal matching module 30 receives the rehabilitation state label graph and structured label set generated by the state recognition module and performs semantic matching with the goal representations in the nursing goal knowledge base to identify the set of rehabilitation goals corresponding to the current state. Furthermore, it extracts rehabilitation constraints related to these goals, constructs goal-constraint mapping relationships, and filters and encodes these mapping relationships to generate a set of goal-constraint pairs, providing a complete logical mapping foundation for path reasoning.

[0226] Pathway generation module 40 is used to perform phased nursing pathway reasoning operations based on the set of goal-constraint pairs and input into the pathway generation model, thereby generating a preliminary multi-node nursing pathway structure. This module further serializes the pathway reasoning results and, in conjunction with the risk factor information identified in the previous steps, assigns intervention priority labels. These labels are embedded in the candidate pathway sequence to generate the final ranked nursing pathway sequence, achieving a fusion of targeted intervention and execution control.

[0227] Feedback modeling module 50 is used to acquire multi-source feedback data during the actual execution of the sequenced care pathway, including behavioral trajectories (e.g., gait changes, movement amplitude) and subjective indicators (e.g., pain scores, recovery comfort). This module constructs a feedback modeling structure at the pathway node level and generates a pathway scoring sequence based on a joint behavioral-subjective model. Furthermore, through correlation analysis and a dynamic scoring mechanism, the weights of each pathway node are adjusted to form an updated pathway structure to support closed-loop model optimization.

[0228] Optimization and Update Module 60 is used to update the parameter weights of the pathway generation model in real time based on the pathway scoring sequence and the updated pathway structure. This module also embeds a knowledge transfer framework to enable cross-patient pathway strategy learning. Furthermore, based on the results of pathway strategy transfer learning, the data semantic structure and the rehabilitation status label graph are automatically and synchronously updated, completing the closed-loop correction of the semantic structure and the dynamic adaptive evolution of the system.

[0229] Beneficial effects of this embodiment:

[0230] 1. This system achieves the effective integration of multi-source heterogeneous data by processing electronic medical records, imaging data and rehabilitation monitoring data in a unified format and constructing a data semantic structure, significantly improving the model's adaptability to individual differences.

[0231] 2. By combining graph neural networks with time series modeling, the system can accurately identify the recovery status and behavioral risk factors of orthopedic patients at different stages, providing refined semantic support for subsequent path reasoning and intervention optimization.

[0232] 3. Based on the goal-constraint mapping mechanism and the sequence generation process of the sorted nursing pathway, the system can dynamically output personalized, phased nursing intervention pathways, and perform pathway scoring and weight updates based on real-time feedback, forming a complete closed loop for individual rehabilitation strategy optimization.

[0233] 4. By introducing a knowledge transfer mechanism, the model can achieve strategy migration and synchronous updating of semantic structure between different patients. The system has good versatility, scalability and self-learning capabilities, and is highly practical and innovative in the field of orthopedic intelligent nursing.

[0234] Obviously, the embodiments described above are only some of the embodiments of the present application, rather than all of the embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application has been described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions described in the aforementioned specific embodiments, or to make equivalent replacements for some of the technical features therein. Any equivalent structure made using the contents of the present application specification and the accompanying drawings, directly or indirectly used in other related technical fields, is also within the scope of patent protection of the present application.

Claims

1. A method for generating a nursing assistance plan for orthopedic patients, characterized in that: The following steps are involved: Obtain the patient's electronic medical record data, medical imaging data, and rehabilitation monitoring data, unify the format and remove noise, generate a nested structure, and then construct a data semantic structure to form an initial graph node set; Performing graph neural network modeling and temporal state modeling on the data semantic structure, extracting key recovery stages and behavioral risk factors, constructing a rehabilitation state label graph, and performing semantic classification and multi-dimensional label assignment to form a structured label set; According to the rehabilitation status label graph and the structured label set, semantic matching with the nursing goal knowledge base and extraction of rehabilitation restriction conditions are performed, the goal-constraint mapping relationship is screened and encoded, and a goal-constraint pair set is generated; Obtain the target-constraint pair set, perform phased nursing pathway reasoning, serialization processing, and intervention priority label embedding to construct a sorted nursing pathway sequence; Modeling the rehabilitation feedback data during the execution of the ordered nursing pathway sequence, using the modeling results as input for association analysis and scoring modeling, generating a pathway scoring sequence for weight adjustment of pathway nodes, and forming a pathway structure with updated weights; Based on the pathway scoring sequence and the pathway structure after weight update, the pathway generation model parameter weights are updated, cross-patient pathway strategy learning is performed, and the data semantic structure and rehabilitation status label map are synchronously updated.

2. The method according to claim 1, characterized in that The steps of obtaining the patient's electronic medical record data, medical imaging data, and rehabilitation monitoring data, performing format unification and noise elimination processing, generating a nested structure, and constructing a data semantic structure to form an initial graph node set include: Obtain the patient's electronic medical record data, medical imaging data, and rehabilitation monitoring data, unify the format, and remove noise to obtain multimodal structured data; Performing feature alignment and modality nesting modeling on the multimodal structured data to generate a nested structure in a unified format; A data semantic structure is constructed based on the nested structure, and an initial graph node set is formed.

3. The method according to claim 1, characterized in that The steps of performing graph neural network modeling and temporal state modeling on the data semantic structure, extracting key recovery stages and behavioral risk factors, constructing a rehabilitation state label graph, and performing semantic classification and multi-dimensional label assignment to form a structured label set include: Performing graph neural network modeling and time series state modeling on the data semantic structure to generate a rehabilitation status prediction result; Extracting key recovery stages and behavioral risk factors from the recovery status prediction results to construct a recovery status label map; The rehabilitation status label graph is semantically classified and multi-dimensionally labeled to form a structured label set.

4. The method according to claim 1, wherein The step of performing semantic matching with a nursing goal knowledge base and extracting rehabilitation restriction conditions based on the rehabilitation status label graph and the structured label set includes: Obtaining a goal representation in a nursing goal knowledge base and performing semantic matching with the rehabilitation status label graph; Based on the semantic matching results, rehabilitation constraints related to the goal are extracted from the structured tag set to construct a goal-constraint mapping relationship.

5. The method according to claim 1, wherein The steps of screening and encoding the target-constraint mapping relationship to generate a target-constraint pair set include: The target-constraint mapping relationship is screened and encoded to generate a target-constraint pair set as input for generating the nursing pathway.

6. The method according to claim 1, characterized in that The steps of obtaining the target-constraint pair set, performing staged nursing pathway reasoning, serialization processing, and intervention priority label embedding include: Obtaining the target-constraint pair set and inputting it into a pathway generation model to perform phased nursing pathway reasoning; Serializing the nursing pathway reasoning results and setting intervention priority labels in combination with risk factors; The intervention priority labels are embedded into the pathway candidate sequence to generate a ranked care pathway sequence.

7. The method according to claim 1, characterized in that The steps of modeling the rehabilitation feedback data during the execution of the sequenced nursing pathway and using the modeling results as input for association analysis and scoring modeling include: Obtaining rehabilitation feedback data during the execution of the sequenced nursing pathway, and modeling behavioral trajectories and subjective indicators; Conduct correlation analysis and score modeling on the modeled behavioral trajectory and subjective indicator results to generate a path score sequence; The weights of the path nodes are adjusted based on the path scoring sequence to form a path structure with updated weights.

8. The method according to claim 1, characterized in that The step of updating the pathway generation model parameter weights based on the pathway score sequence and the pathway structure after weight update and performing cross-patient pathway strategy learning includes: Obtaining the path scoring sequence and the path structure after weight update, and updating the parameter weights of the path generation model; The updated pathway generation model is embedded in the knowledge transfer framework for cross-patient pathway strategy learning.

9. The method according to claim 8, characterized in that The step of synchronously updating the data semantic structure and the rehabilitation status label map includes: Based on the cross-patient pathway strategy learning results, the data semantic structure and the rehabilitation status label graph are synchronously updated to complete the closed-loop update of the semantic structure.

10. A system for generating nursing assistance plans for orthopedic patients, characterized in that: include: The data processing module is used to obtain the patient's electronic medical record data, medical imaging data, and rehabilitation monitoring data, unify the format and remove noise, generate a nested structure, and then construct a data semantic structure to form an initial graph node set; A state recognition module is used to perform graph neural network modeling and temporal state modeling on the data semantic structure, extract key recovery stages and behavioral risk factors, construct a rehabilitation state label map, and perform semantic classification and multi-dimensional label assignment; A target matching module is used to perform semantic matching with the nursing target knowledge base and extract rehabilitation constraint conditions based on the rehabilitation status label graph and the structured label set, screen and encode the target-constraint mapping relationship, and generate a target-constraint pair set; A pathway generation module is used to obtain the set of goal-constraint pairs, perform phased nursing pathway reasoning, serialization processing, and intervention priority label embedding to construct a sequence of ranked nursing pathways; A feedback modeling module is used to model the rehabilitation feedback data during the execution of the sequenced nursing pathway, use the modeling results as input to perform association analysis and scoring modeling, generate a pathway scoring sequence, adjust the weights of pathway nodes, and form a pathway structure with updated weights; The optimization update module is used to update the path generation model parameter weights based on the path scoring sequence and the path structure after the updated weights, perform cross-patient path strategy learning, and synchronously update the data semantic structure and rehabilitation status label map.

Images