A personalized clinical pathway generation method and system based on dynamic graph reconstruction

Abstract

The application relates to a personalized clinical path generation method based on dynamic graph reconstruction, and belongs to the technical field of medical data processing. The method comprises the following steps: S1, constructing a clinical path meta-graph, wherein the clinical path meta-graph is a directed graph, and comprises a variation compensation node pool and a conflict resolution weight matrix W; S2, collecting multi-source heterogeneous medical data of a patient, and performing data cleaning, feature extraction and standardization processing to obtain a patient state feature vector; S3, performing dynamic graph reconstruction on the clinical path meta-graph based on the patient state feature vector; and S4, performing candidate path planning on the reconstructed graph and outputting a candidate diagnosis and treatment path. The application improves the computability, adaptability, continuity and consistency of the clinical path generation under a complex condition and a multi-combined syndrome scenario. Meanwhile, the application also provides a system capable of realizing the above method.

Description

Technical Field

[0001] This invention belongs to the field of medical data processing technology and relates to a method and system for generating personalized clinical pathways based on dynamic map reconstruction. Background Technology

[0002] Clinical pathways are important tools for standardizing diagnosis and treatment processes and improving the quality and efficiency of medical care. However, existing technologies suffer from the following problems in complex clinical scenarios: they are primarily based on single diseases and fixed processes, resulting in static pathway structures that are difficult to adapt to individual patient differences and dynamic changes in their conditions; nodes are mostly text-based descriptions, lacking calculable constraints and standardized data structures, making automatic matching and real-time updates difficult; and when multiple diseases coexist, the lack of a unified pathway fusion and quantitative conflict resolution mechanism can easily lead to mutually exclusive suggestions and increase the burden of manual decision-making.

[0003] Therefore, there is a need for a computationally comprehensible, fusion-enabled, and dynamically reconfigurable clinical pathway generation method that can provide stable auxiliary references without replacing physicians' decision-making. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method and system for generating personalized clinical pathways based on dynamic atlas reconstruction.

[0005] To achieve the above objectives, the present invention provides the following technical solution: On the one hand, this invention provides a method for generating personalized clinical pathways based on dynamic atlas reconstruction, comprising the following steps: S1: Construct a clinical pathway meta-graph, wherein the clinical pathway meta-graph is a directed graph. Including the mutation compensation node pool With the conflict resolution weight matrix W; the mutation compensation node pool Used to store unconventional but potentially useful alternative nodes for scenarios involving complications, conflicts, or special conditions; the conflict resolution weight matrix W is composed of state feature vectors. Dynamically generated node representation vectors from self-supervised pre-training; S2: Collect multi-source heterogeneous medical data of patients and perform data cleaning, feature extraction and standardization to obtain patient state feature vectors. ; S3: Based on patient state feature vector Perform dynamic atlas reconstruction on the clinical pathway metamap; S4: Perform candidate path planning on the reconstructed graph and output candidate treatment paths.

[0006] Furthermore, the node set in step S1 V Each node includes a node identifier, a set of pre-triggered conditions, a set of mutual exclusions, resource requirements, and a recommendation probability; the edge set...E Each edge includes a starting node, an ending node, an edge weight, and a standard time constraint; the edge weight is used to characterize the recommendation probability or adverse reaction risk of node flow.

[0007] Further, step S3 includes the following steps: dividing the node matching results into blocked state, standard active state, and adaptively adjusted active state; performing elastic matching of hard constraint matching and soft constraint adaptation on the nodes, deleting blocked state nodes and retaining active state nodes; then calculating the path connectivity coefficient C and the coverage of key diagnostic and treatment targets R, triggering the automatic triggering process of the variation compensation node pool when C falls into a preset low connectivity interval or R<1, and triggering the manual intervention process when the preset conditions are still not met after automatic triggering; constructing a joint graph and identifying mutually exclusive node groups in multi-disease scenarios, calculating the priority score of conflict nodes based on the conflict resolution weight matrix W and retaining high priority nodes; finally, on the graph to be optimized that has passed dual detection, dynamically correcting the edge time constraint and edge risk weight based on the severity score of the disease.

[0008] Furthermore, the path connectivity coefficient is:

[0009] in Let be the number of valid core links reachable from the starting node to the ending node. The total number of core procedures; the coverage of key diagnostic and treatment objectives is:

[0010] in For a predefined set of key diagnostic and treatment objectives, This is the subset of targets that the currently active node can cover; if the preset conditions of C and R are still not met after automatic compensation, the broken link and missing target information are output to trigger manual intervention.

[0011] Furthermore, the priority score of the conflicting nodes is calculated using the following formula:

[0012] Where W is the conflict resolution weight matrix, which is dynamically calculated and generated from the patient state feature vector x and the self-supervised pre-trained node representation vector; The node feature vector includes indicators of severity, safety, and evidence-based level; the node with the highest score is retained and the remaining conflicting nodes are removed, and then double detection is performed again to avoid introducing new structural defects after conflict resolution.

[0013] Furthermore, the generation of the dynamic conflict resolution weight vector W includes the following steps: Construct a historical best practice library: extract successful diagnosis and treatment sequences that achieve preset positive prognostic indicators from historical electronic medical records as a positive sample set; Node representation pre-training: Clinical path modeling with masking is performed on the positive sample set, randomly masking some diagnosis and treatment nodes. The masked nodes are then predicted using a Transformer network to obtain node representation vectors containing deep clinical semantics. ; Calculate mutual information prior weights: Calculate each candidate diagnosis / treatment node Mutual information value between the positive prognostic index G and the positive prognostic index G , as the global prior weight of the historical best practice; Dynamic weight calculation: When faced with mutual exclusion node conflicts, the dynamic conflict resolution weight vector is calculated using the following formula:

[0014] in, This is the current patient state attention vector extracted based on contrastive learning. For vector dimensions, This is the mutual information adjustment coefficient.

[0015] Furthermore, the severity score of the condition is as follows:

[0016] As a disease-specific severity index, For abnormal vital signs, Number of complications , , For the weighting coefficients, satisfying It is formulated by clinical experts based on the specific disease; The correlation coefficient of the disease based on the severity score is calculated as follows:

[0017] in, The correlation coefficient between the disease and the patient's condition. This is the clinical adjustment coefficient, which is then used to correct the edge standard time constraint.

[0018] Furthermore, the conventional edge risk weights in step S3 are revised as follows:

[0019] in, As the benchmark risk weight, This is the risk gain coefficient; For associated edges introduced by mutation compensation nodes, a compensation gain term is superimposed on the correction:

[0020] in, An additional risk coefficient is added to compensate for the edge.

[0021] Furthermore, in step S4, the multi-objective path cost function is constructed as follows:

[0022] in, , , , These are the weighting coefficients. , , , These represent time cost, risk cost, resource cost, and priority benefit, respectively. Under the conditions of satisfying prerequisite dependencies, time window, resource capacity, and target coverage constraints, the top K candidate paths and their ranking results are output, and compensation operation identifiers and special quality control suggestions are added to the compensation nodes.

[0023] On the other hand, the present invention provides a personalized clinical pathway generation system based on dynamic atlas reconstruction, including a memory, a processor, and a communication interface; the memory is used to store a computer program; the processor is used to implement the personalized clinical pathway generation method based on dynamic atlas reconstruction as described above when the computer program is executed; the communication interface is used to collect medical data and output candidate paths with a hospital information system.

[0024] The beneficial effects of this invention are as follows: it improves the computability, adaptability, continuity, and consistency of clinical pathway generation in complex disease and multi-comorbidity scenarios. Furthermore, this invention also provides a system capable of implementing the above methods.

[0025] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0026] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 A flowchart of a personalized clinical pathway generation method based on dynamic atlas reconstruction; Figure 2 This is a schematic diagram of path reconstruction under complex conditions of a single disease as described in Example 2; Figure 3 This is a schematic diagram of the multi-disease quantitative conflict resolution described in Example 3; Figure 4 A diagram illustrating the implementation of a personalized clinical pathway generation system. Detailed Implementation

[0027] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0028] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0029] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0030] Example 1: like Figure 1 As shown, this invention provides a method for generating personalized clinical pathways based on dynamic atlas reconstruction, comprising the following steps: First, in step S1, a clinical pathway meta-graph is constructed. Clinical guidelines and expert consensus rules are parameterized and mapped into a computable directed graph. Configure the mutation compensation node pool. The conflict resolution weight matrix W is used. Its core components are the node set and the edge set, and it includes a built-in mutation compensation node pool and conflict resolution weight matrix, specifically defined as follows: Node set V : Represents specific diagnostic and treatment procedures; each node is defined as a 5-tuple data structure. in: A unique identifier for the node; It is a set of pre-triggered condition logic, consisting of hard constraints and soft constraints, all of which are quantifiable clinical test indicator thresholds, allergy history, or protocol adjustment rules. This is a mutually exclusive set, containing diagnostic and treatment operations that cannot be executed simultaneously with this node; Resource requirements for node execution; The probability of recommendation execution for each node. Node construction adopts a dimensional decomposition method of "operation type + medical parameter constraints", combining parameters such as liver and kidney function, age, allergy history, and drug resistance to generate subdivided nodes, supporting subsequent incremental updates.

[0031] Edge set E: Each edge is defined as a quadruple data structure ,in: As the starting node, This is the terminating node; is the edge weight, representing the recommendation probability or adverse reaction risk of node transition; is the standard time constraint of the edge. This represents the time interval for node transition.

[0032] Mutation Compensation Node Pool This is used to store alternative nodes that are unconventional but can be used for scenarios involving complications, conflicts, or special conditions. Each compensation node has a predefined adaptation threshold and correlation with the treatment goal. When the automatic triggering conditions are met, the system will automatically integrate the data; when the manual triggering conditions are met, integration will be confirmed by a doctor.

[0033] Conflict resolution weight matrix W: This invention breaks through the limitations of traditional static manual weighting and adopts a dynamic weight generation mechanism driven by self-supervised learning.

[0034] Subsequently, in step S2, multi-source heterogeneous patient data is acquired and a patient state feature vector is constructed. This step includes three sub-processes: data acquisition, data cleaning, and feature extraction and standardization. Ultimately, the multi-source heterogeneous patient data is transformed into a standardized patient state feature vector, making patient states computable and providing a quantitative basis for node matching. Specifically, patient data is collected from EHR, LIS, PACS, and allergy history management systems, including at least: demographic information, diagnostic information, medical orders, laboratory indicators, imaging conclusions, allergy history, complications, and past medical history. Considering the possibility of missing, abnormal, and inconsistent coding in the original medical data, this invention performs a unified data preprocessing procedure before entering the graph matching process, including missing value imputation, outlier rule verification, and invalid code mapping, finally constructing the original feature set. After standardization, we get: Continuous features are processed using normalization or standard scores, while discrete features are processed using one-hot encoding or target encoding.

[0035] After patient vectorization, the system proceeds to step S3, the dynamic atlas reconstruction stage based on elastic matching. This stage is the core of the invention, comprising four sub-processes: node suitability determination and pruning, dual-detection compensation mechanism, multi-atlas fusion conflict resolution, and dynamic parameter adjustment. In step S3, the system performs joint hard and soft constraint determination on each node in the meta-atlas. To balance clinical safety and pathway flexibility, the invention divides the node matching results into three states: blocking state, standard activation state, and adaptive adjustment activation state. The matching state function can be expressed as:

[0036] in, For a set with hard constraints, [0, 1] represents the soft constraint fit. The first preset threshold; =0, the node is determined to be in a blocked state and is pruned from the graph; =1 indicates the standard active state, and the node is retained according to the standard parameters; =2 indicates adaptive adjustment of the activation state; the node is retained, but the trigger parameter adaptive strategy is implemented. This hierarchical judgment method avoids clinical distortions caused by coarse-grained "keep all or delete all" strategies.

[0037] In step S3, the system performs a dual-detection mechanism on the pruned map. It's important to note that dual-detection involves more than just calculating two indicators; the key is forming a closed-loop process of "detection—judgment—treatment—re-detection." First, the system calculates the path connectivity coefficient:

[0038] in Let be the number of valid core links reachable from the starting node to the ending node. The total number of core components. Then calculate the coverage of key diagnostic and treatment objectives:

[0039] in For a predefined set of key diagnostic and treatment objectives, This is the target subset that the currently active node can cover. If... and =1, indicating the reconstructed graph is qualified, proceed to path planning step S4; if or <1, triggering the automatic triggering process of the mutation compensation node pool, from After screening and incorporating candidate nodes, the system is re-examined; if the automatic compensation still fails to meet the standards, a manual intervention process is triggered, and the broken links, missing targets, and candidate compensation lists are output for doctors to confirm and revise.

[0040] When a patient has multiple diagnoses, the system proceeds to step S3 to perform multi-map fusion and conflict resolution. The system first loads the corresponding meta-maps for each disease and constructs a joint map, then identifies mutually exclusive node groups. For each conflict group, the system uses a conflict resolution weight matrix... W Calculate node priority score:

[0041] in, For conflict resolution weight matrix, The node feature vector can include indicators such as severity, safety, and evidence-based level. Finally, the system retains the node with the highest score and removes the remaining conflicting nodes, and then performs double detection again to avoid introducing new structural defects after conflict resolution.

[0042] In step S3, the system dynamically adjusts the parameters of the atlas that has passed dual detection. It's important to emphasize that S3 does not involve directly adjusting parameters on any arbitrary atlas; rather, it presupposes that the structure is connected and the target is complete. For single-disease scenarios, the atlas to be optimized is a reconstructed atlas that meets the target after compensation; for multi-disease scenarios, the atlas to be optimized is a joint atlas that meets the target after conflict resolution and compensation. Subsequently, the system calculates the severity score based on the patient's feature vector.

[0043] in: The disease-specific severity index is extracted from the diagnosis-related subvectors and mapped to [0,5] after standardization. The vital signs abnormality index is extracted from the vital signs subvector and mapped to [0,3] after standardization. The number of complications is extracted from the complication diagnosis subvector and normalized to be mapped to [0,2]. , , For the weighting coefficients, satisfying It is formulated by clinical experts based on the specific disease.

[0044] Then, the disease severity correlation coefficient is calculated to characterize the relationship between the severity of the patient's condition and the time constraints of the treatment process. The coefficient ranges from 0 to 1 and is calculated using the disease severity score and the clinical adjustment coefficient. The clinical adjustment coefficient, set by specialists based on evidence-based medicine, characterizes the degree of influence of disease severity on time constraints. The calculation formula is as follows:

[0045] in, The correlation coefficient between the disease and the patient's condition. This is a clinical adjustment coefficient, with a value range of [0.2, 0.8], customized according to department or disease. The severity of the condition is scored; when the condition is mild, Approaching 1; when the condition worsens, This reduces the time window for critical processes.

[0046] Then, the system uses Modify the standard time constraints on the edges:

[0047] in, Represents a node arrive Standard time constraints (units can be set to hours or days). This indicates the personalized time constraint. To perform floor operations, the time constraint must be a positive integer; Used to set a minimum time limit to prevent zero or negative time windows and ensure scheduling feasibility.

[0048] Simultaneously, the system dynamically adjusts the edge risk weights. For regular edges, the adjusted risk weight can be expressed as:

[0049] in, As the benchmark risk weight, This is the risk gain coefficient. With... Increase The corresponding increase is made to enhance risk constraints in high-risk scenarios. For associated edges introduced by mutation compensation nodes, a compensation gain term is superimposed on the above basic correction value, expressed as:

[0050] in, A risk coefficient is added to the compensation edge. This can increase the strength of the risk constraint when the condition worsens or a compensatory operation is introduced. In the modified... , and Based on this, the system performs topology sorting and constructs a multi-objective path cost function:

[0051] in, , , , These are the weighting coefficients. , , , These represent the time cost, risk cost, resource cost, and priority benefit, respectively. Under the constraints of prerequisite dependencies, time window, resource capacity, and target coverage, the system outputs the top K candidate paths and their ranking.

[0052] Finally, the candidate paths are instantiated into digital treatment sequences, outputting the node execution order, suggested execution time windows, resource requirements, and quality control prompts; compensation nodes are appended with "compensation operation identifiers" and specific monitoring suggestions. If real-time updates to the patient's feature vector cause changes in the node state, the system automatically triggers reconstruction and reordering of the process; after manual adjustments by the doctor, the system automatically reviews connectivity and target coverage and outputs updated results, with the final treatment decision led by the clinician.

[0053] Example 2: like Figure 2 As shown, this embodiment provides path reconstruction under complex conditions of a single disease, using a patient with "severe pneumonia complicated by renal insufficiency and quinolone allergy" as an example to illustrate the process of the present invention.

[0054] S1: Constructing a meta-map of clinical pathways for pneumonia Configure the mutation compensation node pool With conflict resolution weight matrix W The anti-infection nodes include quinolone solutions and non-quinolone alternatives, and the edge attributes include standard time constraints. and risk weight .

[0055] S2: Collect EHR, LIS, PACS, and allergy history data, and obtain patient status feature vectors after cleaning, coding, unification, and standardization. For example, the patient meets the criteria of "quinolone allergy = yes, impaired kidney function = yes, and high severity of infection". S3: Node applicability determination and pruning: Quinolone-related nodes are pruned because hard constraints are not met, while check and monitoring nodes remain active.

[0056] Dual detection and compensation triggering: after pruning if or <1, trigger The system automatically selects compensation nodes and integrates them into the graph, then re-examines them until the threshold is met or manual intervention is required.

[0057] This embodiment is for a single disease and does not perform multi-map conflict resolution.

[0058] Dynamic parameter correction: based on severity scoring Adjust the time constraints and risk weights, and preferably adopt: , , , For example, when When the price rises, the critical time window narrows and the risk cost increases.

[0059] S4: In reconstructing the map Topology sorting and candidate path planning are performed using a multi-objective cost function: Output the top K candidate paths and their order, and instantiate them as a treatment sequence (execution order, suggested time window, resource requirements, and quality control recommendations). Doctors can manually adjust nodes and time windows, and the system automatically checks connectivity and target coverage before updating the output.

[0060] The path reconstruction output under complex conditions of a single disease is shown in Table 1.

[0061] Table 1

[0062] This embodiment demonstrates how to maintain path executability through a closed loop of "pruning-compensation-re-rearrangement" when forbidden nodes are inactive.

[0063] Example 3: like Figure 3 As shown, this embodiment provides path reconstruction under the quantitative conflict resolution of multiple diseases. Taking a patient with "pneumonia complicated with heart failure and mild dehydration" as an example, it illustrates the map fusion, conflict adjudication and compensation repair process under multiple disease conditions.

[0064] S1: Construct the pneumonia meta-map separately With Heart Failure Metamap And set the conflict resolution weight matrix W and the mutation compensation node pool. Where W is used for calculating the priority of conflicting nodes, Pre-configured alternative nodes can be used for target completion.

[0065] S2: Collect patient EHR, LIS and medical order data, clean and standardize them to obtain patient state feature vector x, which is used to simultaneously represent infection status, cardiac function status and volume status.

[0066] S3: Node Suitability Determination and Pruning: These are respectively in... and The process involves matching hard and soft constraints to obtain a set of activated nodes. Calculate the connectivity coefficient C and target coverage R for the preliminary fused map; Identify mutually exclusive node groups (e.g., fluid replenishment strategy conflicts) and calculate the priority of conflicting nodes using the following formula:

[0067] Retain high-scoring nodes and remove the remaining conflicting nodes; if C is below the target or R < 1 after conflict resolution, trigger... Automatic compensation and re-inspection; if the conditions are still not met after automatic compensation, manual intervention is triggered.

[0068] On the joint map of compliance, the time constraints and risk weights are dynamically adjusted according to the severity of the disease.

[0069] S4: Perform path planning on the revised joint graph, output the top K candidate paths and their ranking, and provide the execution order of key nodes, suggested time windows, and resource requirements for doctors' reference. This embodiment achieves a closed loop of "fusion—conflict adjudication—compensation and repair—re-examination" when multiple diseases coexist and medical orders are mutually exclusive, ensuring that candidate paths are continuous and key objectives are covered.

[0070] Example 4: like Figure 4 As shown, this embodiment utilizes a personalized clinical pathway generation system, including a processor, memory, communication interface, and doctor workstation. The memory stores program instructions for executing the aforementioned method steps. The system may further include a meta-map construction module, a patient data processing module, a dynamic map reconstruction module, a candidate path generation module, and a rule configuration module, used to realize patient data collection, map reconstruction, candidate path generation, and result display. The system can interact with hospital information systems such as HIS, LIS, and PACS, and supports incremental triggering after patient status updates and automatic review after manual adjustments by doctors. The final treatment decision is still made by the clinician.

[0071] Example 5: An electronic device, comprising a memory and a processor; The memory is used to store computer programs; The processor is configured to implement the method described in Embodiment 1 when executing the computer program.

[0072] Example 6: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in Embodiment 1.

[0073] Example 7: A computer program product includes a computer program that, when executed by a processor, implements the method described in Example 1.

[0074] In the above embodiments, the reference to "this embodiment" in the specification indicates that a specific feature, structure, or characteristic described in connection with the embodiment is included in at least some embodiments, but not necessarily all embodiments. Multiple appearances of "this embodiment" do not necessarily refer to the same embodiment.

[0075] In the above embodiments, although the invention has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory structures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed. The embodiments of the invention are intended to cover all such substitutions, modifications, and variations falling within the broad scope of the appended claims.

[0076] As will be understood by those skilled in the art, the computer-readable storage medium described in this embodiment allows for the implementation of all or part of the steps in the above method embodiments by computer program-related hardware. The aforementioned computer program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0077] The electronic terminal provided in this embodiment includes a processor, a memory, a transceiver, and a communication interface. The memory and the communication interface are connected to the processor and the transceiver and complete communication between them. The memory is used to store computer programs, the communication interface is used to perform communication, and the processor and the transceiver are used to run the computer programs, so that the electronic terminal performs the steps of the above method.

[0078] In this embodiment, the memory may include random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device.

[0079] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0080] This invention can be used in a wide range of general-purpose or special-purpose computing system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices, etc.

[0081] This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for generating personalized clinical pathways based on dynamic atlas reconstruction, characterized in that: Includes the following steps: S1: Construct a clinical pathway meta-graph, wherein the clinical pathway meta-graph is a directed graph. Including the mutation compensation node pool With the conflict resolution weight matrix W; the mutation compensation node pool Used to store unconventional but potentially useful alternative nodes for scenarios involving complications, conflicts, or special conditions; the conflict resolution weight matrix W is composed of state feature vectors. Dynamically generated node representation vectors from self-supervised pre-training; S2: Collect multi-source heterogeneous medical data of patients and perform data cleaning, feature extraction and standardization to obtain patient state feature vectors. ; S3: Based on patient state feature vector Perform dynamic atlas reconstruction on the clinical pathway metamap; S4: Perform candidate path planning on the reconstructed graph and output candidate treatment paths.

2. The personalized clinical pathway generation method based on dynamic atlas reconstruction according to claim 1, characterized in that: Node set in step S1 V Each node includes a node identifier, a set of pre-triggered conditions, a set of mutual exclusions, resource requirements, and a recommendation probability; the edge set... E Each edge includes a starting node, an ending node, an edge weight, and a standard time constraint; the edge weight is used to characterize the recommendation probability or adverse reaction risk of node flow.

3. The personalized clinical pathway generation method based on dynamic atlas reconstruction according to claim 1, characterized in that: The steps include: Step S3 includes the following steps: dividing the node matching results into blocked state, standard active state, and adaptively adjusted active state; performing elastic matching of hard constraint matching and soft constraint adaptation on the nodes, deleting blocked state nodes and retaining active state nodes; then calculating the path connectivity coefficient C and the coverage of key diagnostic and treatment targets R, triggering the automatic triggering process of the variation compensation node pool when C falls into the preset low connectivity interval or R<1, and triggering the manual intervention process when the preset conditions are still not met after automatic triggering; constructing a joint graph and identifying mutually exclusive node groups in multi-disease scenarios, calculating the priority score of conflict nodes based on the conflict resolution weight matrix W and retaining high priority nodes; finally, on the graph to be optimized that has passed dual detection, dynamically correcting the edge time constraint and edge risk weight based on the severity score of the disease.

4. The personalized clinical pathway generation method based on dynamic atlas reconstruction according to claim 3, characterized in that: The path connectivity coefficient is: in Let be the number of valid core links reachable from the starting node to the ending node. The total number of core procedures; the coverage of key diagnostic and treatment objectives is: in For a predefined set of key diagnostic and treatment objectives, The target subset that can be covered by the currently active node; If the preset conditions for C and R are still not met after automatic compensation, the information on the broken link and missing target will be output to trigger manual intervention.

5. The personalized clinical pathway generation method based on dynamic atlas reconstruction according to claim 3, characterized in that: The priority score of the conflicting node is calculated using the following formula: Where W is the conflict resolution weight matrix, which is dynamically calculated and generated from the patient state feature vector x and the self-supervised pre-trained node representation vector; The node feature vector includes indicators of severity, safety, and evidence-based level; the node with the highest score is retained and the remaining conflicting nodes are removed, and then double detection is performed again to avoid introducing new structural defects after conflict resolution.

6. The personalized clinical pathway generation method based on dynamic atlas reconstruction according to claim 3, characterized in that: The generation of the dynamic conflict resolution weight vector W includes the following steps: Construct a historical best practice library: extract successful diagnosis and treatment sequences that achieve preset positive prognostic indicators from historical electronic medical records as a positive sample set; Node representation pre-training: Clinical path modeling with masking is performed on the positive sample set, randomly masking some diagnosis and treatment nodes. The masked nodes are then predicted using a Transformer network to obtain node representation vectors containing deep clinical semantics. ; Calculate mutual information prior weights: Calculate each candidate diagnosis / treatment node Mutual information value between the positive prognostic index G and the positive prognostic index G , as the global prior weight of the historical best practice; Dynamic weight calculation: When faced with mutual exclusion node conflicts, the dynamic conflict resolution weight vector is calculated using the following formula: in, This is the current patient state attention vector extracted based on contrastive learning. For vector dimensions, This is the mutual information adjustment coefficient.

7. The personalized clinical pathway generation method based on dynamic atlas reconstruction according to claim 3, characterized in that: The severity score of the condition is as follows: As a disease-specific severity index, For abnormal vital signs, Number of complications , , For the weighting coefficients, satisfying It is formulated by clinical experts based on the specific disease; The correlation coefficient of the disease based on the severity score is calculated as follows: in, The correlation coefficient between the disease and the patient's condition. This is the clinical adjustment coefficient, which is then used to correct the edge standard time constraint.

8. The personalized clinical pathway generation method based on dynamic atlas reconstruction according to claim 3, characterized in that: In step S3, the regular edge risk weights are corrected as follows: in, As the benchmark risk weight, This is the risk gain coefficient; For associated edges introduced by mutation compensation nodes, a compensation gain term is superimposed on the correction: in, An additional risk coefficient is added to compensate for the edge.

9. The personalized clinical pathway generation method based on dynamic atlas reconstruction according to claim 1, characterized in that: In step S4, the multi-objective path cost function is constructed as follows: in, , , , These are the weighting coefficients. , , , These represent time cost, risk cost, resource cost, and priority benefit, respectively. Under the conditions of satisfying prerequisite dependencies, time window, resource capacity, and target coverage constraints, the top K candidate paths and their ranking results are output, and compensation operation identifiers and special quality control suggestions are added to the compensation nodes.

10. A personalized clinical pathway generation system based on dynamic atlas reconstruction, characterized in that: The system includes a memory, a processor, and a communication interface; the memory is used to store a computer program; the processor is used to implement the personalized clinical pathway generation method based on dynamic atlas reconstruction as described above when the computer program is executed; the communication interface is used to collect medical data and output candidate paths with a hospital information system.

Images