Medical knowledge graph-driven diagnosis and treatment decision support system

Abstract

The invention discloses a medical knowledge graph-driven diagnosis and treatment decision support system, and relates to the technical field of medical knowledge graph data processing, and the method comprises the steps: obtaining a clinical data flow of a patient, mapping the clinical data flow into an updating instruction for a pre-constructed medical knowledge graph, and dynamically updating the attribute data of entity nodes and relation edges; generating a vectorization representation of the node by taking the target entity node as an original point according to the attribute data of the relation edge between the target entity node and the associated node; screening nodes to be compounded according to a preset rule, and calculating a composite vector between the target core node and the vectorized representation of the nodes to be compounded; and finally, generating a clinical state evaluation result according to the mathematical characteristics of the composite vector. According to the method, the complex medical relationship is quantified into the computable vector space model, dynamic, objective and quantitative evaluation and risk early warning of the clinical state of the patient are achieved, and the accuracy and foresight of diagnosis and treatment decisions are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of medical knowledge graph data processing technology, and more specifically, this application relates to a medical knowledge graph-driven diagnostic and treatment decision support system. Background Technology

[0002] In clinical medicine, particularly in the diagnosis and treatment of endocrine and metabolic diseases, accurately assessing a patient's real-time physiological state and dynamically weighing the risks and benefits of treatment plans are core challenges for clinicians. Early identification and intervention of medical risks are crucial for improving patient prognosis.

[0003] Currently, clinical decision support systems largely rely on two main technological approaches. The first is expert systems based on fixed rule bases, which use predefined if-then logic rules to evaluate patient data. However, the knowledge bases of such systems often exhibit static characteristics, making it difficult to incorporate the latest medical research evidence, and the rules lack organic connections, failing to represent complex interactions between medical concepts. The second is systems based on traditional knowledge graphs. While these systems can structurally express medical knowledge through entity and relational networks, they still have significant limitations in practical applications. Existing knowledge graphs primarily focus on the symbolic representation and storage of knowledge, and their reasoning capabilities are limited to path queries and simple logical judgments, lacking the ability to quantitatively describe relation strength, dynamic changes, and the synergistic effects of multiple factors.

[0004] Specifically, existing technologies face the following pressing technical challenges: First, the system cannot effectively transform real-time, multi-source, heterogeneous clinical data into dynamic adjustments to the internal state of the knowledge graph, resulting in a lag between the knowledge model upon which decision-making relies and the patient's actual physiological state. Second, traditional knowledge graphs struggle to provide a unified, quantitative, and comprehensive assessment of antagonistic clinical factors such as drug efficacy and side effect risks. Doctors often need to make subjective trade-offs based on isolated risk warnings, lacking objective and visualized decision-making support tools. Furthermore, existing early warning mechanisms are mostly based on static threshold triggers of single parameters, failing to capture the potential clinical state evolution trends jointly indicated by subtle changes in multiple physiological parameters, thus missing the valuable window for early intervention.

[0005] Therefore, there is an urgent need in this field for a technical solution that can deeply integrate real-time clinical data, dynamically quantify medical knowledge, and comprehensively assess complex clinical situations and provide early risk warnings. Summary of the Invention

[0006] To address the aforementioned technical problems and provide a medical knowledge graph-driven diagnostic and treatment decision support system, this technical solution resolves the issues raised in the background section.

[0007] In a first aspect, embodiments of this application provide a medical knowledge graph-driven diagnostic and treatment decision support system, characterized by comprising: a data acquisition module: used to acquire the clinical data stream of a target patient and map it into an update instruction for a pre-constructed medical knowledge graph; wherein the medical knowledge graph includes entity nodes and relation edges; a data update module: used to execute the update instruction to update the attribute data of the target entity node and / or the attribute data of the corresponding relation edges in the medical knowledge graph; a data processing first module: used to respond to the update instruction and generate a current vectorized representation for at least one target entity node; wherein generating the vectorized representation includes: taking the target entity node as the origin, calculating a vectorized representation starting from the origin based on the attribute data of the relation edges between the target entity node and multiple associated nodes; a data processing second module: used to determine at least one node to be composited from multiple associated nodes according to a preset node filtering rule, and calculate a composite vector between the vectorized representation of the target entity node and the vectorized representation of the node to be composited; and a data output module: used to generate a clinical status assessment result based on the mathematical characteristics of the composite vector.

[0008] Secondly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned medical knowledge graph-driven diagnostic decision support system.

[0009] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0010] 1. Through data acquisition and update modules, the real-time acquired clinical data stream is continuously mapped and updated to the attribute data of entity nodes and relation edges in the medical knowledge graph. This enables the system to dynamically respond to changes in patient conditions. Furthermore, the first data processing module transforms the complex medical relationships in the graph into a vectorized representation composed of a set of spatial vectors with the target entity node as the origin, converting qualitative medical knowledge into a computable mathematical model. Finally, by calculating the composite vector and analyzing its mathematical characteristics, clinical status assessment results are generated. This integrates the overall information of the target entity node and its associated network, avoiding the limitations of single-indicator judgments, and thus providing a more objective, comprehensive, and quantitative integrated assessment method.

[0011] 2. The system not only assesses at a single point in time but also continuously tracks the vectorized representations of composite vectors and key risk nodes over a time series. By calculating dynamic indicators such as the rate of change of magnitude and the change of direction angle, and further identifying fine-grained time-series patterns such as sudden increases, sustained growth, and abnormal fluctuations, the system can reveal latent patterns in disease progression that are easily overlooked by static analysis, thus providing clinicians with earlier and more predictive risk warnings.

[0012] 3. Through a recursive screening mechanism, the most important nodes to be composited are efficiently selected from a massive number of related nodes based on whether the modulus of the relation edges exceeds a preset modulus threshold. This process simulates the focused thinking of clinicians, avoiding unnecessary complex calculations across the entire knowledge graph and significantly improving the system's computational efficiency. Simultaneously, the entire decision-making process is rooted in a structured medical knowledge graph, and the final evaluation result stems from vectorized calculations of specific nodes and relation edges, making the system's reasoning logic chain clear and traceable. Doctors can retrospectively identify which key risk nodes and what changes triggered the warning, enhancing the transparency and credibility of AI-assisted decision-making and making it more readily accepted in clinical practice. Attached Figure Description

[0013] Figure 1 A schematic diagram of the structure of a medical knowledge graph-driven diagnostic and treatment decision support system provided in an embodiment of this application;

[0014] Figure 2 This is a schematic diagram of the logical flow of the clinical risk trend early warning of the second module of data processing driven by medical knowledge graph provided in the embodiments of this application. Detailed Implementation

[0015] This application provides a medical knowledge graph-driven diagnostic and treatment decision support system, which solves the technical problem in the prior art of insufficient dynamic quantitative characterization and assessment of complex clinical situations and insufficient early risk warning.

[0016] In existing technologies, clinical decision support systems largely rely on static rule bases or traditional knowledge graph technologies. Static rule bases struggle to incorporate the latest medical evidence, lack interrelationships between rules, and fail to represent the interactions between complex medical concepts. Traditional knowledge graphs focus on symbolic storage, limiting reasoning capabilities to path queries and lacking quantitative descriptions of relationship strength, dynamic changes, and multi-factor synergistic effects. For example, in real-time monitoring scenarios for diabetic patients, traditional systems cannot dynamically correlate blood glucose fluctuations, changes in renal function indicators, and pharmacokinetic parameters, leading to delays in treatment plan adjustments.

[0017] To address these challenges, a knowledge graph update mechanism capable of dynamically integrating real-time clinical data is needed, along with the development of quantitative assessment models. The key challenges lie in transforming multi-source heterogeneous data into dynamic adjustment instructions for the knowledge graph, and in mathematically modeling complex clinical factors. By introducing a vector space model, medical entities and their relationships are mapped to computable geometric features, enabling the capture of synergistic effects between multidimensional parameters. Composite analysis based on vector operations allows for a comprehensive assessment of clinical conditions.

[0018] Therefore, this application proposes a diagnostic and treatment decision support system comprising a data acquisition module, a data update module, a first data processing module, a second data processing module, and a data output module. The data acquisition module acquires the patient's clinical data stream and generates knowledge graph update instructions; the data update module executes these instructions to adjust entity and relation attributes. The first data processing module generates vectorized representations for core entities; the second data processing module filters associated nodes and calculates composite vectors; and the data output module generates evaluation results based on the characteristics of the composite vectors.

[0019] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0020] like Figure 1 The diagram shown is a structural schematic of a medical knowledge graph-driven diagnostic decision support system provided in this application embodiment, comprising: a data acquisition module for acquiring clinical data streams of target patients and mapping them to update instructions for a pre-constructed medical knowledge graph; wherein the medical knowledge graph includes entity nodes and relation edges; a data update module for executing update instructions to update the attribute data of target entity nodes and / or the attribute data of corresponding relation edges in the medical knowledge graph; a first data processing module for responding to update instructions and generating a current vectorized representation for at least one target entity node; wherein generating a vectorized representation includes: taking the target entity node as the origin and calculating a vectorized representation starting from the origin based on the attribute data of the relation edges between the target entity node and multiple associated nodes; a second data processing module for determining at least one node to be composited from multiple associated nodes according to preset node filtering rules and calculating a composite vector between the vectorized representation of the target entity node and the vectorized representation of the node to be composited; and a data output module for generating a clinical status assessment result based on the mathematical characteristics of the composite vector.

[0021] The data acquisition module's update instruction mapping involves extracting medical concept entities through semantic analysis and generating attribute update instructions after matching them with knowledge graph nodes. This mechanism ensures that real-time data can dynamically correct the knowledge model, for example, associating newly detected creatinine values ​​with kidney function nodes. The attribute adjustment in the data update module involves basic weights, confidence parameters, and severity parameters, which provide dynamic weighting for subsequent vector calculations. The vectorized representation of the first data processing module uses a polar coordinate system, constructing a spatial vector set through the magnitude and orientation angle of relational edges. For example, the relational vector between drug nodes and side effect nodes can represent the strength of the risk association. The node selection in the second data processing module uses a recursive traversal mechanism, identifying key associated nodes through preset magnitude thresholds, such as automatically capturing secondary nodes related to cardiorenal function in hypertension assessment.

[0022] Specifically, the system first converts real-time collected data such as blood pressure and biochemical indicators into knowledge graph update instructions. For example, when a patient's blood potassium level is abnormal, the confidence parameter of the electrolyte balance node is updated. The first data processing module uses the cardiac function node as the origin and calculates the magnitude and direction angle of its relationship edges with surrounding nodes, generating a set containing multiple spatial vectors. The second data processing module recursively filters associated nodes whose magnitude exceeds a threshold, such as identifying kidney nodes strongly correlated with cardiac function, and performs vector composition operations. The final output composite vector direction angle offset indicates the risk of cardiorenal syndrome, and changes in magnitude reflect the severity of the condition.

[0023] Compared to existing technologies, traditional knowledge graphs only support Boolean relation queries and cannot quantitatively assess changes in association strength. This solution, through a dynamic vector space model, transforms discrete medical concepts into continuous mathematical features, supporting quantitative analysis of multi-factor interaction effects. For example, in drug efficacy evaluation, traditional methods can only provide isolated risk indicators, while composite vectors can simultaneously characterize the dynamic balance between treatment efficacy and side effect risk.

[0024] Through the above technical solutions, this application achieves real-time dynamic updates of the clinical knowledge model, solving the decision-making bias problem caused by data lag. Vectorized representation and compound computation mechanisms can capture subtle but synergistic changes in multiple parameters, such as the early detection of the correlation between abnormal liver function and changes in coagulation indicators. This system provides clinicians with visualized risk assessment results, supporting comprehensive judgment of complex conditions, such as considering the combined effects of hormone levels, cardiovascular parameters, and drug metabolism status in thyroid function assessment.

[0025] Furthermore, the mapping is used as an update instruction for the pre-built medical knowledge graph, specifically including: identifying and extracting medical concept entity identifiers from the clinical data stream using a semantic analysis model; matching the medical concept entity identifiers with entity nodes in the medical knowledge graph; and generating instructions for updating attribute data of corresponding relation edges based on the matching results and preset mapping rules; the attribute data includes basic weight values, confidence correction parameters, and severity parameters.

[0026] In this embodiment, the semantic analysis model refers to an algorithmic model used to identify standardized medical concepts from unstructured text. Specifically, it can be implemented using a deep learning-based named entity recognition model, such as a sequence labeling model constructed by combining a bidirectional long short-term memory network with a conditional random field.

[0027] Medical concept entity identifiers refer to standardized coded clinical medical terms, which can be implemented using the International Classification of Diseases (ICD) or the Unique Medical Language System (UMLS).

[0028] The matching process refers to semantically aligning the extracted entity identifiers with knowledge graph nodes. Specifically, it can be implemented using a retrieval algorithm based on vector similarity or an inference engine based on ontology logic.

[0029] Mapping rules refer to defining the logical relationship between clinical data and knowledge graph update operations. Specifically, they can be implemented using rule engines or decision tree models. For example, when an abnormal laboratory test result is detected, the weight adjustment of the corresponding relation edge is triggered.

[0030] The base weight value is used to characterize the inherent strength of the relationship between medical entities, the confidence correction parameter is used to dynamically adjust the confidence of the relationship based on individual patient data, and the severity parameter is used to quantify the degree of abnormality of clinical indicators.

[0031] For example, the medical concept of "elevated fasting blood glucose" is identified from the patient's chief complaint and mapped to a standardized entity identifier. This entity identifier is then matched against associated nodes under the "diabetes" node in the knowledge graph. If a match is successful, the corresponding relationship edges are updated according to preset rules. For instance, when a patient's fasting blood glucose levels exceed a threshold three times consecutively, the system calculates a confidence correction parameter based on the mapping rules, reducing the association weight between the "dietary control" treatment plan and the "blood glucose control" target, while simultaneously updating the severity parameter based on the deviation of the detected values. This achieves a dynamic transformation of clinical data into a knowledge graph, enabling subsequent decision support modules to perform reasoning based on the latest state.

[0032] By dynamically updating the multi-dimensional attribute parameters of the relationship edges, the knowledge graph can reflect the individualized characteristics of patients in real time, providing an accurate data foundation for subsequent risk assessment. For example, in the scenario of drug side effect monitoring, the system can adjust the confidence parameters of drug metabolism pathway relationships based on the dynamic changes of patients' liver function indicators, thereby identifying potential liver damage risks in advance.

[0033] Furthermore, the specific steps for calculating the vectorized representation starting from the origin are as follows: traverse all associated nodes connected to the target entity node through relational edges; for each relational edge between the target entity node and all associated nodes, calculate the magnitude of the spatial vector corresponding to the relational edge using the first calculation model; based on the type identifier of the relational edge, obtain the direction angle of the spatial vector corresponding to the relational edge by querying a preset mapping table of relational type direction angles; each spatial vector is defined by its magnitude and direction angle; combine the spatial vectors calculated from all associated nodes into a set, and use this set as the vectorized representation of the target entity node.

[0034] In this embodiment, the magnitude of the spatial vector refers to the length of the relation edge in the vector space. Specifically, it can be calculated using the first calculation model combined with the basic weight value, confidence correction parameter and severity parameter to quantify the importance of the relation edge in the current clinical state.

[0035] The direction angle of a relation type refers to the directional identifier of different medical relation types in the vector space. Specifically, a preset mapping table can be used to convert semantic relations into geometric angles, so that relation edges with different medical semantics present a differentiated distribution in the vector space.

[0036] A spatial vector set refers to a whole composed of spatial vectors of multiple related nodes. By combining the vectorized contributions of each related node, the dynamic state characteristics of the target entity node in the knowledge graph can be comprehensively reflected.

[0037] For each connection edge, the first calculation model is invoked to calculate the modulus based on the edge's base weight value, historical confidence correction parameters, and current severity parameters. For example, the base weight value can be set by medical experts, and the confidence correction parameters are dynamically adjusted based on patients' historical data. Then, a preset orientation angle mapping table is queried based on the relationship type identifier; for example, a "drug interaction" relationship can be mapped to 120 degrees, and a "symptom association" relationship to 60 degrees. The spatial vector corresponding to each associated node is uniquely determined by the modulus and orientation angle. Finally, the spatial vectors of all associated nodes are aggregated into a set, forming a multidimensional vectorized representation of the target entity node.

[0038] For example, in the evaluation of insulin treatment plans for diabetic patients, when real-time blood glucose data triggers knowledge graph updates, the system can accurately reflect the strength of the relationship between the current blood glucose level and the risk of complications through modulus calculation, and distinguish the correlation between physiological indicators and drug effects through directional angles, thereby providing quantitative evidence with spatiotemporal dimensions for clinical decision-making.

[0039] Furthermore, the specific calculation formula for the first calculation model is as follows: ;in, This represents the magnitude of the calculated spatial vector. This represents the basic weight value of the relation edge. This represents the confidence adjustment parameter for the relation edge. A parameter indicating the severity of a relation edge.

[0040] In this embodiment, the basic weight value refers to the initial weight of the pre-set relationship edge in the knowledge graph. Specifically, it can be determined through the experience of medical experts or the statistical results of historical data, and is used to reflect the importance of the relationship in the static knowledge system.

[0041] The confidence correction parameter is a correction factor that is dynamically adjusted based on the patient's individual historical data. It can be calculated using a Bayesian probability model or a sliding window statistical method and is used to characterize the change in the confidence of the relationship during the current patient's diagnosis and treatment process.

[0042] Severity parameters are quantitative indicators extracted from clinical data streams. Specifically, they can be obtained by numerically converting diagnostic reports and laboratory test results through a semantic parsing engine, and are used to reflect the degree of abnormality of the current symptoms or physiological indicators.

[0043] When generating a vectorized representation of a target entity node, it is necessary to traverse all its associated nodes and calculate the spatial vector magnitude corresponding to each relation edge. By linearly superimposing the basic weight values ​​with the confidence correction parameter and the severity parameter, the magnitude can simultaneously reflect the static knowledge weight, the confidence of individualized data, and the current severity of the condition.

[0044] For example, when a patient experiences a drug allergic reaction, the severity parameter of the edge connecting the relevant drug node and the allergy symptom node will significantly increase, thereby increasing the magnitude of the vector. This makes the system pay more attention to this risk factor in subsequent composite vector calculations. Thus, the magnitude of the spatial vector no longer depends solely on preset weights, but is dynamically adjusted by fusing real-time data to ensure that the representational capabilities of the knowledge graph are updated in sync with the patient's actual condition.

[0045] A dynamic adjustment mechanism can effectively address the assessment lag problem caused by fixed parameters in traditional systems. For example, in chronic disease management, when a patient's physiological indicator continues to deteriorate but has not reached a preset threshold, the cumulative change in the severity parameter will trigger an increase in the vector magnitude earlier, thus helping doctors identify potential risk trends sooner. Furthermore, through the design of a linear superposition formula, the system achieves a balance between computational complexity and result interpretability, avoiding the difficulty in tracing the clinical decision-making process due to the introduction of too many nonlinear calculations.

[0046] Furthermore, according to the preset node filtering rules, at least one node to be composited is determined from multiple associated nodes. Specifically, the following operations are performed by the collection unit in the second data processing module: calculating the magnitude of the relational edges between the target entity node and all associated nodes; filtering out relational edges whose magnitude exceeds a preset magnitude threshold; taking the associated nodes of the relational edges as nodes to be composited; and repeating the operation of the collection unit with the node to be composited as a new target entity node until the number of repetitions reaches the upper limit or the magnitude of the relational edges between all associated nodes of the new target entity node does not exceed the preset magnitude threshold, thus obtaining at least one node to be composited.

[0047] In this embodiment, the preset modulus threshold is a critical value used to determine whether the relation edge has significant clinical relevance. Specifically, it can be set through medical statistical data analysis. For example, the preset modulus threshold can be set to 0.8.

[0048] The maximum number of repetitions refers to the maximum number of iterations allowed during the recursive search process. For example, it can be set to 3 times to avoid infinite loops.

[0049] The operation of the collection unit refers to expanding the search range through recursive traversal. Its purpose is to discover potential multi-level related nodes, thereby avoiding the omission of indirectly related but clinically significant nodes.

[0050] When the modulus of the relational edges of a target entity node exceeds a preset threshold, its corresponding associated nodes are marked as nodes to be composited. Then, using this node to be composited as a new starting point, the modulus of the relational edges of its associated nodes is recalculated. If edges still exceed the threshold, the search level is expanded further. For example, in a diabetes complication assessment scenario, the initial core node is a blood glucose index node, and its modulus with the kidney function node is 0.85. Since this exceeds the threshold, the kidney function node is included in the set of nodes to be composited. Further recursive searching of the kidney function node reveals that its modulus with the urine protein index node is 0.79, which does not reach the threshold; therefore, the expansion of this path is stopped. By controlling the recursion depth and modulus threshold, effective coverage of key associated nodes is ensured while redundant calculation of invalid nodes is avoided.

[0051] By combining recursive expansion with threshold filtering, clinically significant indirect association nodes can be identified. For example, in cardiovascular risk assessment of hypertensive patients, a second-order association between blood pressure control nodes and retinopathy nodes can be found. Simultaneously, setting an upper limit on the number of recursions balances computational efficiency and search depth. For instance, in emergency scenarios, limiting the number of recursions to two allows for the filtering of key nodes within 10 milliseconds, meeting real-time decision-making requirements.

[0052] Furthermore, calculating the composite vector between the vectorized representation of the target entity node and the vectorized representation of the node to be composited specifically includes: performing vector addition on all spatial vectors contained in the vectorized representation of the target entity node and all spatial vectors contained in the vectorized representation of the node to be composited; and using the result vector as the composite vector.

[0053] In this embodiment, vectorization refers to the process of converting entity nodes and their relationships into mathematical vectors. Specifically, a vector space model can be used to map the attribute parameters of each entity node and its relationships into multi-dimensional vectors. Vector addition refers to the algebraic addition of the values ​​of different vectors in the same dimension. Specifically, this can be achieved using the component superposition method in the Cartesian coordinate system. Vector synthesis reflects the collaborative relationship between multiple nodes. A composite vector is a comprehensive vector formed by vector superposition, and its mathematical characteristics are used to characterize the overall effect between the target entity node and its associated nodes.

[0054] After generating the vectorized representation of the target entity node, the system iterates through all its associated nodes to identify the nodes to be composited that meet the selection criteria. For each node to be composited, the system obtains its corresponding set of vectorized representations. A new composite vector is formed by algebraically adding all the spatial vectors of the target node and the nodes to be composited.

[0055] For example, when the target node vector contains a vector with a magnitude of 3 and a direction angle of 30 degrees, and the composite node vector contains a vector with a magnitude of 2 and a direction angle of 60 degrees, the system calculates the components of each vector on the X and Y axes through trigonometric function decomposition, adds the corresponding components, and resynthesizes a new composite vector. The changes in the magnitude and direction angle of this composite vector can reflect the strength and direction of the synergistic effect between multiple medical concepts.

[0056] By using vector superposition operations, the relationships between medical concepts scattered across different nodes can be transformed into a unified mathematical representation, capturing potential risk trends formed by the combined effects of multiple weak signals.

[0057] For example, in the risk assessment of diabetic complications, traditional methods may only focus on a single indicator of blood glucose, while this approach can integrate the vector changes of multiple factors such as blood glucose fluctuations, renal function indicators, and the degree of vascular lesions, and identify the risk of multi-system synergistic deterioration in advance.

[0058] The rate of disease progression can be determined by the trend of changes in the magnitude of composite vectors, and the main sources of risk can be identified by the directional angle offset, providing data support for the development of personalized intervention plans.

[0059] For example, in the treatment of hypertension, when the composite vector magnitude increases rapidly and the direction angle points to the kidney-related node, the system can indicate the risk of iatrogenic kidney injury and assist doctors in adjusting the medication regimen.

[0060] Furthermore, it also includes a second data processing module: obtaining composite vectors from the data output module on the time series to obtain a set of composite vectors for the time series; calculating the maximum value of the rate of change of the magnitude and the maximum value of the change in the orientation angle of the composite vectors in the set of composite vectors with a preset sequence length of the time series; generating a clinical risk trend warning when the maximum value of the rate of change of the magnitude exceeds a preset rate of change threshold or the change in the orientation angle exceeds a preset change threshold.

[0061] In this embodiment, Figure 2 This is a schematic diagram of the logical flow of the clinical risk trend early warning of the second module of data processing driven by medical knowledge graph provided in this application embodiment. The composite vector set of time series refers to the set formed by arranging composite vectors in chronological order. Specifically, it can be implemented by using a sliding window mechanism to collect composite vector data at fixed time intervals, which is used to reflect the dynamic evolution of the patient's clinical status.

[0062] The maximum rate of change of modulus refers to the maximum relative change in the modulus of the composite vector in the time series. Specifically, it can be determined by calculating the ratio of the absolute value of the difference in modulus between adjacent time points to the initial modulus, and is used to characterize the severity of disease progression.

[0063] The maximum change in direction angle refers to the maximum absolute offset of the direction angle of the composite vector in the time series. Specifically, it can be achieved by using the polar angle difference algorithm to calculate the absolute difference between the current direction angle and the reference direction angle, which is used to indicate abnormal deviations in the direction of disease progression.

[0064] The composite vectors generated by the data output module are continuously recorded and formed into an ordered set according to timestamps. The preset sequence length can be set to a time window of, for example, 24 hours or 72 hours. Within the window, all composite vectors are traversed to calculate the rate of change of magnitude and the change of orientation angle. When the rate of change of magnitude exceeds a preset threshold, such as 30%, or the orientation angle shift exceeds, for example, 45 degrees, an early warning generation mechanism is triggered. Furthermore, local change analysis is performed on key risk nodes. For example, an independent threshold for the rate of change of magnitude is set for nodes related to drug side effects. When the magnitude of a node increases by more than 20% within three consecutive time windows, a specific type of risk warning signal is generated.

[0065] By analyzing composite vector data over time, trends of disease deterioration can be detected early, such as the abnormal increase in composite vector magnitude detected 6-8 hours before the onset of septic shock. Furthermore, combining this with analysis of local changes at key risk points can differentiate between acute and chronic risk types.

[0066] For example, rapid changes in modulus length correspond to acute complications, while continuous deviations in orientation angle may indicate chronic organ dysfunction, thus providing a more precise basis for clinical intervention decisions.

[0067] Furthermore, the formula for calculating the change in direction angle is: ;in, This indicates the change in direction angle. This represents the direction angle of the composite vector at the current time point. This represents the direction angle of the composite vector at the initial reference time point.

[0068] In this embodiment, the change in orientation angle refers to the magnitude of the change in orientation angle of the composite vector over time. Specifically, the orientation angle of the composite vector can be calculated using a polar coordinate transformation algorithm, and the degree of orientation shift can be quantified by absolute value difference calculation. The orientation angle θc of the composite vector at the current time point can be calculated from the Cartesian coordinate components of the composite vector through polar coordinate transformation. The orientation angle θn of the initial reference time point can be set as the orientation angle of the composite vector under the patient's baseline state or the average orientation angle within the historical time window. This approach can identify the qualitative changes in clinical state implied by the directional shift of the composite vector, solving the problem of traditional early warning mechanisms being insensitive to the coordinated changes of multiple parameters. For example, in the early warning scenario of sepsis, when the correlation direction between blood pressure, body temperature, and inflammatory markers undergoes a systematic shift, even if the magnitudes of each parameter are still within the normal range, the abnormality in the change in orientation angle can still trigger an early intervention prompt for septic shock, avoiding delays in treatment due to a single parameter not reaching a threshold.

[0069] Furthermore, generating clinical risk trend warnings also includes: identifying and extracting nodes associated with preset high-risk medical concepts from all nodes to be composited using preset semantic rules, as key risk nodes; calculating the rate of change of magnitude or the change in orientation angle of the vectorized representation of the key risk node itself over time; when the rate of change of magnitude of any key risk node exceeds a preset local rate of change threshold, or its change in orientation angle exceeds a preset local change threshold, then for that key risk node, performing at least one of the following time-series feature recognition operations: calculating the difference between the rate of change of magnitude of the current time series and the rate of change of magnitude of the previous time series. When the difference exceeds a preset surge threshold, a surge characteristic is detected, generating a local clinical risk trend warning indicating a significant increase in acute risk. Within a preset time window, the number of time series with continuously positive modulus change rate is counted. When this number exceeds a preset continuous growth count threshold, a continuous growth characteristic is detected, generating a local clinical risk trend warning indicating the continuous accumulation of chronic risk. The variance of the modulus change rate within the time window is calculated. When the calculation result exceeds a preset fluctuation threshold, an abnormal fluctuation characteristic is detected, generating a local clinical risk trend warning indicating unstable condition or the presence of interfering factors.

[0070] In this embodiment, key risk nodes refer to entity nodes that have a semantic association with preset high-risk medical concepts. Specifically, they can be achieved by matching high-risk disease codes in the medical ontology database using a natural language processing model, in order to focus on pathological elements with urgent clinical intervention.

[0071] After detecting a global risk trend warning, further refined time-series analysis is performed on key risk nodes. For example, in the monitoring scenario of diabetic ketoacidosis patients, the preset high-risk medical concept may include key pathological states such as electrolyte disturbance and acid-base imbalance. When the system identifies a blood potassium concentration node as a key risk node, it continuously tracks the change in its vectorized modulus. If the difference in the modulus change rate exceeds the surge threshold, it can be determined that there is a risk of acute hypokalemia, triggering an emergency potassium supplementation warning; if the modulus continues to increase positively for three consecutive monitoring cycles, it indicates a chronic hyperkalemia trend that requires adjustment of the insulin treatment regimen; when the modulus fluctuation variance exceeds the threshold, it can identify abnormal detection data or unstable treatment response.

[0072] This system enables multi-dimensional dynamic monitoring of pathological processes related to specific high-risk medical concepts, effectively distinguishing between acute risk events and chronic pathological progression, and identifying unstable states during treatment. For example, in the early warning of thyroid storm, the system can simultaneously track vector changes in multiple key nodes such as heart rate, body temperature, and level of consciousness. When a sudden increase in the magnitude of the heart rate node is detected coexisting with a continuous increase in the body temperature node, a risk warning of multi-organ dysfunction can be issued in advance, identifying the trend of disease deterioration on average 2-3 hours earlier than traditional single-indicator alarm mechanisms.

[0073] Furthermore, it also includes a VR visualization module, used to visualize clinical status assessment results and / or clinical risk trend warnings in virtual reality; the VR visualization module specifically includes:

[0074] Data interface unit: used to receive clinical status assessment results and mathematical feature data of composite vectors from the data output module;

[0075] 3D Scene Construction Unit: Used to construct 3D virtual scenes based on pre-built medical knowledge graphs, where entity nodes are mapped to geometric objects in 3D space, and relation edges are mapped to strip-shaped structures connecting geometric objects;

[0076] Dynamic rendering unit: used to dynamically adjust the visual parameters of the 3D scene according to the mathematical characteristics of the composite vector, where the magnitude of the composite vector is mapped to the size parameter of the corresponding geometric object, and the orientation angle of the composite vector is mapped to the orientation parameter of the geometric object in 3D space;

[0077] Risk warning display unit: When a clinical risk trend warning is detected, the corresponding risk node is highlighted in a virtual reality environment by flashing brightly, changing color, or vibrating spatially.

[0078] The workflow of the VR visualization module includes: first, acquiring composite vector data and risk assessment results at the current time point through the data interface unit; then, generating a 3D topology structure of the knowledge graph centered on the target entity node in the 3D scene construction unit; the dynamic rendering unit adjusting the size of each node's geometry in real time according to the magnitude of the composite vector, and determining the node's spatial position in the spherical coordinate system based on the orientation angle; when the system generates a clinical risk trend warning, the risk warning display unit will automatically focus on the corresponding key risk node and enhance risk perception through visual effects.

[0079] By introducing a VR visualization module, doctors can immerse themselves in observing the dynamic relationships between various entity nodes in a medical knowledge graph, intuitively perceiving the evolution trends of clinical states represented by changes in the magnitude of composite vectors and shifts in orientation angles. For example, in a diabetes complication monitoring scenario, doctors can use VR devices to observe a three-dimensional topological network centered on pancreatic function nodes. When the risk of kidney complications increases, the corresponding kidney node will flash red in a pulse and gradually enlarge, helping doctors quickly locate key risk factors.

[0080] This application provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the aforementioned medical knowledge graph-driven diagnostic decision support system. The computer-readable storage medium refers to a physical carrier used for persistently storing program code, specifically a solid-state drive, flash drive, or optical disc. The stored computer program contains a sequence of instructions that can be parsed and executed by a processor. Processor execution refers to the processing of program instructions through a central processing unit or graphics processing unit, specifically implemented using a multi-core processor or distributed computing cluster, to achieve dynamic updating of the medical knowledge graph and clinical status assessment functions.

[0081] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CDRM, optical storage, etc.) containing computer-usable program code.

[0082] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0083] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0084] These computer program instructions can also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0085] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0086] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A medical knowledge graph driven diagnosis and treatment decision support system, characterized in that, The application relates to a clinical state evaluation method and device. The method comprises the following steps: a data acquisition module is used to acquire a clinical data stream of a target patient and map the clinical data stream into an update instruction for a pre-constructed medical knowledge graph; the medical knowledge graph comprises entity nodes and relationship edges; a data update module is used to execute the update instruction to update attribute data of a target entity node and / or attribute data of a corresponding relationship edge in the medical knowledge graph; a data processing first module is used to generate a current vectorized representation of at least one target entity node in response to the update instruction; wherein the vectorized representation is generated by taking the target entity node as a starting point and calculating a vectorized representation starting from the starting point according to attribute data of relationship edges between the target entity node and multiple associated nodes; a data processing second module is used to determine at least one to-be-combined node from the multiple associated nodes according to a preset node screening rule, and calculate a combination vector between the vectorized representation of the target entity node and the vectorized representation of the to-be-combined node; 2.The medical knowledge graph driven diagnosis and treatment decision support system according to claim 1, characterized in that, a data output module is used to generate a clinical state evaluation result according to a mathematical feature of the combination vector. The mapping into the update instruction for the pre-constructed medical knowledge graph specifically comprises the following steps: the clinical data stream is identified and extracted through a semantic analysis model to obtain a medical concept entity identifier; the medical concept entity identifier is matched with entity nodes in the medical knowledge graph; based on a matching result and a preset mapping rule, an instruction for updating attribute data of a corresponding relationship edge is generated; 3.The medical knowledge graph driven diagnosis and treatment decision support system according to claim 1, characterized in that, the attribute data comprises a basic weight value, a confidence correction parameter and a severity parameter. The specific steps of calculating the vectorized representation starting from the starting point are as follows: all associated nodes connected to the target entity node through relationship edges are traversed, and for each relationship edge between the target entity node and all associated nodes, a first calculation model is used to calculate a length of a space vector corresponding to the relationship edge; a direction angle of the space vector corresponding to the relationship edge is obtained by querying a mapping table of preset relationship type direction angles according to a type identifier of the relationship edge; each space vector is defined by the length and the direction angle; 4.The medical knowledge graph driven diagnosis and treatment decision support system according to claim 2, characterized in that, the space vectors calculated from all associated nodes are combined into a set, and the set is taken as the vectorized representation of the target entity node. ; wherein denotes the calculated module of the space vector, denotes the base weight value of the relationship edge, denotes the confidence correction parameter of the relationship edge, denotes the severity parameter of the relationship edge. 5.The medical knowledge graph driven diagnosis and treatment decision support system according to claim 1, wherein, The specific calculation formula of the first calculation model is as follows: The specific steps of determining at least one to-be-combined node from the multiple associated nodes according to the preset node screening rule are as follows: the following operations are performed by a collection unit in the data processing second module: the lengths of relationship edges between the target entity node and all associated nodes are calculated; relationship edges with lengths exceeding a preset length threshold are screened out; associated nodes associated with the relationship edges are taken as the to-be-combined nodes; the operation of the collection unit is repeatedly performed with the to-be-combined nodes as new target entity nodes until the number of repeated executions reaches an upper limit or the lengths of relationship edges between all associated nodes of the new target entity nodes do not exceed the preset length threshold, and at least one to-be-combined node is obtained. 6.The medical knowledge graph driven diagnosis and treatment decision support system according to claim 1, wherein, The calculating the composite vector between the vectorized representation of the target entity node and the vectorized representation of the node to be compounded specifically comprises: performing vector addition operation on all spatial vectors contained in the vectorized representation of the target entity node and all spatial vectors contained in the vectorized representation of the node to be compounded; taking the operation result vector as the composite vector. 7.The medical knowledge graph driven diagnosis and treatment decision support system according to claim 1, wherein, Further comprising a data processing second module: obtaining the composite vector in the data output module in the time sequence to obtain a composite vector set of the time sequence; calculating the maximum value of the modulus change rate and the maximum value of the direction angle change amount of the composite vector in the composite vector set under the preset sequence length of the time sequence; generating a clinical risk trend early warning when the maximum value of the modulus change rate exceeds a preset change rate threshold or the direction angle change amount exceeds a preset change amount threshold. 8.The medical knowledge graph driven diagnosis and treatment decision support system according to claim 7, characterized in that, The calculation formula of the direction angle change amount is: ; wherein denotes the directional angle change amount, denotes the directional angle of the complex vector at the current time point, denotes the directional angle of the complex vector at the initial reference time point. 9.The medical knowledge graph driven diagnosis and treatment decision support system according to claim 7, wherein, The generating the clinical risk trend early warning further comprises: extracting, from all the nodes to be compounded, a node associated with a preset high-risk medical concept as a key risk node through preset semantic rule recognition; calculating the modulus change rate or the direction angle change amount of the vectorized representation of the key risk node itself in the time sequence; when the modulus change rate of any key risk node exceeds a preset local change rate threshold or the direction angle change amount thereof exceeds a preset local change amount threshold, performing at least one of the following time sequence feature recognition operations on the key risk node: calculating the difference between the modulus change rate of the current time sequence and the modulus change rate of the previous time sequence, and when the difference exceeds a preset sudden increase threshold, determining that a sudden increase feature is detected and generating a local clinical risk trend early warning indicating a significant increase in acute risk; counting the number of time sequences in which the modulus change rate is continuously positive within a preset time window, and when the number exceeds a preset continuous growth count threshold, determining that a continuous growth feature is detected and generating a local clinical risk trend early warning indicating a chronic risk that is continuously accumulated; calculating the variance of the modulus change rate within the time window, and when the calculation result exceeds a preset fluctuation threshold, determining that a fluctuation anomaly feature is detected and generating a local clinical risk trend early warning indicating that the disease is unstable or there is an interference factor.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by a processor to realize the system of any one of claims 1-9.

Images