Fusion Method, System and Medium of Urological Medical Knowledge Graph

Abstract

The present invention relates to the technical field of knowledge graph fusion, and particularly to a method, system and medium for fusing urological medical knowledge graphs. The method includes the following steps: obtaining urological historical medical record data and CT images of lesion sites, and performing image region segmentation to obtain CT segmentation images of lesion sites. Then, based on the medical record data, synchronous intersection recognition of lesion trends is performed on the CT segmentation images to obtain synchronous intersection data of lesion trends, and implicit association deduction is performed to obtain causally implicit association data of lesion trends; the causally implicit association data is processed using a knowledge graph representation learning method to obtain graph association structure learning data, and graph fusion logic adjustment is performed. Finally, based on the adjusted data, a lesion graph fusion logic strategy is designed and sent to the cloud platform for execution. The present invention makes the knowledge graph fusion technology more perfect through optimized processing of the knowledge graph fusion technology.

Description

Technical Field

[0001] The present invention relates to the technical field of knowledge graph fusion, and particularly to a method, system and medium for fusing urological medical knowledge graphs. Background Art

[0002] The method for fusing urological medical knowledge graphs aims to provide comprehensive and accurate knowledge support for clinical diagnosis, treatment planning and scientific research innovation by integrating and correlating multi-source heterogeneous urological medical data. Urology covers a wide range of disease types and complex pathological mechanisms, including urinary tract infections, urological tumors, urinary tract stones, etc. The diagnosis and treatment of various diseases often involve data sources in multiple disciplinary fields, such as patient electronic health records (EHRs), laboratory test results, imaging data, genomic information, clinical guidelines and literature. However, these data are usually scattered in different systems, showing differences in data format, semantic definition and storage structure, making it difficult to comprehensively and comprehensively utilize them, and restricting the potential of intelligent clinical decision-making. As a new type of data management and knowledge expression technology, knowledge graphs can effectively express the complex relationships between entities by modeling data as semantic networks, providing an innovative path for the organization, storage and utilization of medical knowledge. For the construction and fusion of urological medical knowledge graphs, multiple challenges need to be addressed, including semantic inconsistencies in different data sources, differences in data quality, and the requirement for dynamic updates of medical knowledge. However, the traditional method for fusing urological medical knowledge graphs has problems such as large errors in inferring the relationship of lesion principles and low fusion correlation of knowledge graphs. Summary of the Invention

[0003] Based on this, it is necessary to provide a method, system and medium for fusing urological medical knowledge graphs to solve at least one of the above technical problems.

[0004] To achieve the above object, a method for fusing urological medical knowledge graphs, the method includes the following steps:

[0005] Step S1: Obtain urological historical medical record data and CT images of the lesion site; perform image region segmentation on the CT images of the lesion site according to the urological historical medical record data to obtain CT segmented images of the lesion site;

[0006] Step S2: Perform identification of synchronous intersection of lesion trends on the CT segmented images of the lesion site according to the urological historical medical record data to obtain synchronous intersection data of lesion trends; perform implicit association deduction on the synchronous intersection data of lesion trends to obtain causal implicit association data of lesion trends;

[0007] Step S3: Perform knowledge graph representation learning on the causal implicit association data of the lesion trend to obtain the graph association structure learning data; perform graph fusion logic adjustment based on the graph association structure learning data to obtain the lesion graph fusion logic adjustment data;

[0008] Step S4: Design a logical strategy based on the lesion graph fusion logic adjustment data to obtain the lesion graph fusion logic strategy, and send the lesion graph fusion logic strategy to the cloud platform to execute the fusion method of the urology medical knowledge graph.

[0009] The present invention first obtains detailed medical information by collecting the historical medical record data of urology patients and combining the patient's disease background, previous diagnosis and treatment records, and clinical symptoms. Then, by analyzing the CT images of the lesion site, the CT images are processed using image segmentation technology to extract the specific area of the lesion site. This process automatically segments the CT images using medical image processing algorithms (such as convolutional neural networks, CNNs, etc.), enabling the precise localization of the lesion area, thereby providing a high-quality data basis for subsequent disease analysis and prediction. By analyzing the historical medical record data of the patient and the CT segmentation images of the lesion site, synchronous intersection recognition of the lesion trend is performed. This means that by comparing and analyzing the CT images of the lesion site at different time points, the laws and trends of lesion progression are identified. These data can help identify potential associations during the development of the lesion and reveal potential causal relationships between lesions. The implicit association deduction technology plays an important role here. It derives the causal relationship of the lesion through mathematical models and machine learning algorithms and generates causal implicit association data of the lesion trend to support disease prediction and treatment decision-making. The causal implicit association data of the lesion trend is converted into a knowledge graph representation, and a lesion knowledge graph is constructed through graph representation learning technology. This process structurally processes the lesion data and presents the different attributes, types, development trends of the lesion, and the relationships between them in the form of a graph. Then, graph fusion technology is used to fuse and logically adjust the data in the graph to optimize the graph structure and relationships. This process optimizes the inference path and logical derivation of the lesion trend by adjusting the connection methods of each node and edge in the graph, making the prediction model of the lesion more accurate and helping to further discover potential disease patterns and treatment plans. Based on the lesion graph fusion logic adjustment data obtained in step S3, a set of logical strategies for the lesion trend is designed. These strategies aim to provide a systematic guidance to help medical professionals make more scientific and reasonable decisions during the diagnosis and treatment process. Finally, the lesion graph fusion logic strategy will be sent to the cloud platform for execution. Through the computing and processing capabilities of the cloud platform, the fusion method of the urology medical knowledge graph is implemented. This not only enhances the data processing and inference capabilities but also enables knowledge sharing and dynamic update through the cloud platform, thereby providing real-time and accurate medical advice and auxiliary decision-making for clinicians and improving the efficiency and quality of medical services. Therefore, the present invention is an optimization process for a traditional fusion method of a urology medical knowledge graph, solving the problems of large errors in the deduction of the lesion principle relationship and low correlation in the fusion of the knowledge graph in the traditional fusion method of a urology medical knowledge graph, reducing the error in the deduction of the lesion principle relationship, and improving the correlation in the fusion of the knowledge graph.

[0010] Preferably, step S1 includes the following steps:

[0011] Step S11: Obtain urology historical medical record data and CT images of the lesion site;

[0012] Step S12: Perform data desensitization processing on the urology historical medical record data to obtain desensitized historical medical record data;

[0013] Step S13: Perform logical description parsing on the desensitized historical medical record data to obtain logical description data of the medical record;

[0014] Step S14: Perform image region segmentation on the CT images of the lesion site according to the logical description data of the medical record to obtain CT segmentation images of the lesion site.

[0015] The present invention first collects the historical medical record data of urology patients and the CT images of the lesion sites. These data include the patient's basic information, medical records, disease descriptions, treatment processes, and examination results. At the same time, the CT images provide visual information of the lesions for further analysis. By comprehensively obtaining the patient's historical medical records and imaging data, it can ensure that there is sufficient data support for subsequent processing and analysis, and ensure the accurate tracking of disease trends and development changes. The obtained CT images can provide an important basis for subsequent image segmentation and trend prediction. The historical medical record data of urology patients is desensitized to protect the privacy and data security of patients. The desensitization process includes removing the patient's personal identity information (such as name, ID number, etc.) to prevent the leakage of the patient's sensitive information. This step is a key step in medical data processing, meeting the requirements of relevant laws and regulations, and ensuring that the data can be used for scientific research and analysis without infringing on the patient's privacy. The desensitized data ensures that the standards of data protection can be followed during medical data analysis and machine learning, and at the same time enables the data to be widely used for technical research and clinical decision support. By performing logical description parsing on the desensitized historical medical record data, key information is extracted and structured. This process uses natural language processing (NLP) technology to convert the descriptive language in the medical record data into clear logical relationships and treatment processes. For example, analyzing information such as the patient's medical history, symptom descriptions, diagnosis conclusions, treatment plans, and their effects, and converting it into logical description data that can be understood by machines. Through this step, the core treatment information in the medical record can be refined, providing more explicit data support and background knowledge for subsequent analysis of the lesion trend and formulation of treatment plans. According to the logical description data obtained from the historical medical records and combined with the patient's specific condition and lesion site, precise segmentation of the CT image area is performed. This step uses machine learning, image processing algorithms, and deep learning models (such as convolutional neural network CNN) to automatically label and segment the lesion site. By combining the disease descriptions in the medical record, the system can identify the specific location and type of the lesion and accurately mark it in the CT image. This image segmentation technology can greatly improve the positioning accuracy of the lesion area, providing important image data support for subsequent lesion analysis, trend prediction, and treatment plan formulation.

[0016] Preferably, step S2 includes the following steps:

[0017] Step S21: Perform structural and morphological analysis on the CT segmentation image of the lesion site to obtain the structural and morphological data of the lesion site;

[0018] Step S22: Perform synchronous intersection recognition of the lesion trend on the structural and morphological data of the lesion site and the CT segmentation image of the lesion site according to the historical medical record data of urology patients to obtain the synchronous intersection data of the lesion trend;

[0019] Step S23: performing causal correlation analysis on the lesion trend synchronization intersection data to obtain lesion trend synchronization causal data;

[0020] Step S24: perform implicit correlation deduction based on the synchronous causal data of the lesion trend to obtain causal implicit correlation data of the lesion trend.

[0021] The present invention can identify and extract the geometric morphology and spatial distribution characteristics of the lesion area by performing structural morphological analysis on the CT segmentation image of the lesion. This process uses image processing techniques, such as morphological analysis, edge detection, etc., to deeply explore the structural characteristics of the lesion area, including shape, size, boundary clarity and its relationship with surrounding tissues. Through these structural features, the development mode and evolution trend of the lesion can be better understood, providing reliable data support for lesion prediction and treatment plan design. The extraction of structural morphological data can reveal the complexity of the lesion and make more accurate judgments in the diagnosis process. By combining the historical medical record data of the urology department and the structural morphological data of the lesion and the CT segmentation image, the trend of the lesion is synchronously identified. This means that by comparing the time series of the changes in the patient's historical medical records and combining the characteristics of the lesion area in the CT image, the system can identify the progression trend of the lesion and its evolution pattern at different time nodes. This step realizes the vertical integration of lesion data, helps to reveal the potential development path of the lesion, and helps to predict the future trend of the disease, thereby providing a more personalized and accurate basis for clinical treatment decisions. Causal correlation analysis is performed on the synchronous intersection data of the lesion trend to explore the potential causal relationship of the development of the lesion. By establishing a causal reasoning model, we can analyze the relationship between changes in the lesion area and the patient's medical history, symptoms, and treatment measures, identify the factors that lead to lesion progression, and explore the causal chain between these factors. Causal association analysis can help doctors understand the root causes of lesion development and reveal the evolution path of the disease under different conditions. This can not only provide a basis for individualized treatment, but also provide scientific guidance for the prediction and preventive measures of similar cases in the future. Implicit association deduction based on lesion trend synchronization causal data means exploring deeper implicit associations between lesions through advanced data mining and machine learning techniques. This process uses algorithms to automatically discover unmanifested lesion factors and their potential causal relationships, and deduce implicit patterns in lesion progression. Implicit association deduction can identify potential risk factors for lesion development, even early, imperceptible change trends. The resulting lesion trend causal implicit association data.

[0022] Preferably, step S22 comprises the following steps:

[0023] Step S221: Perform spatial domain analysis on the structural morphological data of the lesion site to obtain lesion location distribution data;

[0024] Step S222: Perform a descriptive trend comparison on the lesion location distribution data based on the historical urology medical record data to obtain the descriptive trend comparison data of the lesion distribution;

[0025] Step S223: Simulate the similarity difference in tissue density at the lesion boundary for the lesion location distribution data and the CT segmentation images of the lesion sites based on the descriptive trend comparison data of the lesion distribution to obtain the similarity difference in tissue density at the lesion boundary;

[0026] Step S224: Analyze the density difference in boundary transition for the lesion location distribution data based on the similarity difference in tissue density at the boundary tissue to obtain the density difference in boundary transition of the lesion tissue;

[0027] Step S225: Identify the synchronous intersection of lesion trends for the lesion site structural morphology data based on the similarity difference in tissue density at the lesion boundary and the density difference in boundary transition of the lesion tissue to obtain the synchronous intersection data of the lesion trends.

[0028] Through spatial domain analysis of the structural and morphological data of the lesion site, the present invention can further refine the spatial distribution and morphological characteristics of the lesion to obtain the lesion location distribution data. This analysis method uses spatial analysis techniques, such as region growing, heat map generation, etc., to spatially locate the lesion area in the CT image, identify the specific location of the lesion site in the body and its relative distribution. By making a descriptive trend comparison between the urology historical medical record data and the lesion location distribution data, the evolution law of the lesion is mined. The historical medical record data includes the lesion information, treatment records, examination results, etc. of the patient at different time periods, while the lesion location distribution data provides the spatial changes of the lesion site. By comparing and analyzing these two types of data, the spatio-temporal evolution trend of the lesion can be revealed, and the change pattern of the lesion area over time can be identified. Based on the descriptive trend comparison data of the lesion distribution, combined with the lesion location distribution data and the CT segmentation image of the lesion site, a similarity difference simulation of the tissue density at the lesion boundary is performed. This process uses density analysis techniques to compare the density differences between the lesion boundary area and the surrounding normal tissues, and then simulates the boundary characteristics of the lesion tissue. In this way, the density differences between the lesion area and the healthy tissues can be identified, and the morphological characteristics of the lesion and its transition area with the normal tissues can be revealed. According to the aforementioned tissue density similarity difference, a boundary transition density difference analysis is performed on the lesion location distribution data. This step focuses on the transition area between the lesion area and the normal tissues, analyzes the density differences in these areas, and reveals the fuzzy zone of the lesion boundary. Through this analysis, the specific scope of the lesion can be delimited more clearly, helping doctors determine the expansion trend of the lesion, evaluate the relationship between the lesion and the surrounding tissues, especially at the boundary of the lesion, how to distinguish it from the healthy tissues. This density difference analysis helps to improve the accuracy of CT image segmentation and reduce misjudgment and missed judgment. Combining the data of the tissue density similarity difference at the lesion boundary and the boundary transition density difference of the lesion tissue, a synchronous intersection recognition of the lesion trend is performed on the structural and morphological data of the lesion site. Through the intersection analysis of these two types of density difference data, the system can identify the synchronous trend in the development of the lesion, that is, the morphological changes of the lesion area at different stages and the transition law between it and the normal tissues. This process helps to identify the synchronous pattern of lesion progression from multiple dimensions, providing precise support for the early detection, location, and treatment intervention of the lesion. At the same time, the synchronous intersection data of the lesion trend can be used for subsequent lesion prediction.

[0029] Preferably, step S223 includes the following steps:

[0030] Perform multi-scale spatial coordinate transformation on the lesion location distribution data to obtain multi-scale lesion area calibration data;

[0031] Perform multi - point adjacent density distribution analysis on the multi - scale lesion area calibration data and the CT segmentation images of the lesion sites according to the descriptive trend comparison data of the lesion distribution, and obtain the multi - point adjacent density distribution data;

[0032] Perform hierarchical density fitting on the multi - point adjacent density distribution data to obtain the hierarchical lesion boundary density data;

[0033] Perform position - density trend deviation analysis based on the hierarchical lesion boundary density data to obtain the position - density trend deviation data;

[0034] Perform a simulation of the tissue density similarity difference at the lesion boundary on the lesion location distribution data and the CT segmentation images of the lesion sites according to the hierarchical lesion boundary density data and the position - density trend deviation data, and obtain the tissue density similarity difference at the lesion boundary.

[0035] Through multi-scale spatial coordinate transformation of the lesion location distribution data, the present invention can calibrate the lesion area at different scale levels. During this process, the spatial coordinate system of the lesion is converted into representations at multiple scales, which can help reveal the manifestations and distribution characteristics of the lesion at different scales (such as microscopic and macroscopic scales). This multi-scale analysis method is particularly effective for complex lesion areas because it can take into account both local details and overall trends, ensuring more accurate and comprehensive calibration of the lesion area, thereby providing more detailed data support for subsequent lesion analysis. By combining the descriptive trend comparison data of lesion distribution and performing multi-point adjacent density distribution analysis, the density change situation between different points in the lesion area can be revealed. This analysis not only focuses on the characteristics of a single lesion location, but by considering multiple points around the lesion site, analyzes the density distribution rules of these points. In this way, the density differences and change trends between different positions of the lesion can be identified, further improving the accuracy of lesion detection and segmentation. The purpose of this process is to clarify the spatial tissue structure of the lesion. By performing hierarchical density fitting on the multi-point adjacent density distribution data, the density characteristics of the lesion boundary can be modeled at different levels. The hierarchical density fitting method divides the lesion area into several sub-regions according to different density levels to more accurately describe the changes in the lesion boundary. The fitting result of each layer reflects the density change trend of the lesion area, which can help doctors identify the boundary of the lesion, tissue density differences, and its expansion pattern. Through this hierarchical analysis method, the complexity of the lesion boundary can be more clearly revealed. By performing position density trend deviation analysis on the hierarchical lesion boundary density data, the abnormal density change trend of the lesion area at different spatial positions can be revealed. This analysis method can identify the deviation between the density changes of each layer inside the lesion area and normal tissues, thereby revealing the nature and evolution direction of the lesion. The position density trend deviation data can reflect the spatial inhomogeneity of the lesion area and the abnormal pattern of density change. By combining the hierarchical lesion boundary density data and the position density trend deviation data and performing lesion boundary tissue density similarity difference simulation, the density difference between the lesion boundary and the surrounding normal tissues can be quantified. This simulation reveals the characteristics and expansion trend of the lesion boundary by comparing the density similarity between the lesion area and its adjacent normal tissues. The density similarity difference simulation helps identify the specific boundary of the lesion and the relationship with the surrounding tissues.

[0036] Preferably, step S24 includes the following steps:

[0037] Step S241: Structurally classify the lesion trend synchronous causal data to obtain the lesion development stage division data;

[0038] Step S242: Deduce the stage trigger factors based on the lesion development stage division data to obtain the lesion stage trigger factor deduction data;

[0039] Step S243: Perform layer-by-layer factor event simulation and analysis on the deduced data of the triggering factors at the lesion stage to obtain the layer-by-layer factor event analysis data;

[0040] Step S244: Conduct implicit association deduction based on the layer-by-layer factor event analysis data and the deduced data of the triggering factors at the lesion stage to obtain the implicit association data of the causal relationship of the lesion trend.

[0041] The present invention first needs to perform detailed structured collation on the lesion trend and related causal data. By classifying, labeling, and categorizing the data of the lesion trend, different types of lesions and their respective development stages can be effectively identified. For example, according to factors such as the symptoms, manifestations, progression speed, and treatment response of the lesion, it can be divided into different stages such as early stage, middle stage, and late stage. Through structured categorization processing, it helps to more accurately analyze the lesion stage and identify the triggering factors in the subsequent steps, providing clear basic data for subsequent causal deduction and analysis. Based on the data of different development stages of the lesion, the key triggering factors affecting the lesion process are deduced. At each stage, the development of the lesion is affected not only by biological factors but also by factors such as the environment, living habits, and external interference. By deeply analyzing the lesion data at each stage, it is possible to identify which factors play a key role in a specific stage. These factors include the physiological changes of the disease, the fluctuations of the immune response, the effects of drug treatment, etc. The deduced stage triggering factors provide data support for in-depth analysis of the development of the lesion and an important basis for formulating personalized treatment plans. By simulating and analyzing the interaction between each triggering factor and the lesion event layer by layer. Each layer of factors will have its unique mechanism of action and will interact with the previous factors. In this process, by gradually analyzing the specific role of each layer of factors, simulating its impact on the lesion process, and revealing the internal connections of different factors. This process can help clarify which factors play a promoting or inhibitory role at different stages of the lesion, providing an important reference for further precisely adjusting the treatment strategy and disease management. By comprehensively analyzing the layer-by-layer factor event analysis data and the deduced data of the triggering factors at the lesion stage, further explore and deduce the implicit correlation between various factors of the lesion. Different factors have complex interactions at different lesion stages, and some seemingly irrelevant factors are also associated due to changes in specific conditions. Through implicit association deduction, potential causal chains can be discovered, which are difficult to directly reveal through traditional analysis methods. The results of implicit association deduction provide a deeper understanding of the lesion development trend and also provide a more scientific and accurate basis for formulating prevention and treatment strategies.

[0042] Preferably, step S3 includes the following steps:

[0043] Step S31: Construct a graph relationship for the implicit causal association data of the lesion trend to obtain a causal trend relationship graph of the lesion;

[0044] Step S32: Perform knowledge graph representation learning on the causal trend relationship graph of the lesion to obtain graph association structure learning data;

[0045] Step S33: Adjust the graph fusion logic of the causal trend relationship graph of the lesion according to the graph association structure learning data to obtain graph fusion logic adjustment data of the lesion.

[0046] In the present invention, by converting the implicit causal association data of the lesion trend into a graph relationship, a causal trend relationship graph of the lesion is constructed. This graph is based on the interaction and influence between different lesion factors, establishing nodes (lesion factors) and edges (the causal relationships between them). The construction of the graph relationship can not only display the dynamic changes at each stage of the lesion, but also reveal the relationships and dependencies of different lesion factors at different development stages. This process helps to clarify the causal chain of the lesion, identify potential lesion triggering factors and their mutual influences, and provide a clear graph structure for subsequent analysis. By applying knowledge graph representation learning, deep learning analysis is performed on the nodes and edges in the causal trend relationship graph of the lesion, so as to extract more complex association patterns and structures. This learning process helps to automatically identify the potential relationships, similarities and influence paths between various lesion factors in the graph. Through the learning of the knowledge graph, the correlation and potential causal connections between certain factors can be discovered. Especially in the case of large amounts of data and complex relationships, some deep-level information that is difficult to identify through manual analysis can be revealed. This process improves the intelligence and adaptability of the graph. By combining the data obtained from the graph association structure learning, the graph fusion logic is adjusted to further optimize the causal trend relationship graph of the lesion. Through logical adjustment, on the basis of the original graph structure, the relationships and association strengths between lesion factors can be dynamically corrected and updated. For example, it is found that the causal effect of certain factors is more significant than expected, or the influence paths of certain factors change at different stages. Through these logical adjustments, the graph more accurately reflects the real development process and internal mechanism of the lesion.

[0047] Preferably, step S33 includes the following steps:

[0048] Step S331: Perform topological association density analysis on the causal trend relationship graph of the lesion according to the graph association structure learning data to obtain topological association density data;

[0049] Step S332: Evaluate the relative centrality of the graph nodes of the causal trend relationship graph of the lesion based on the topological association density data to obtain the relative centrality of the graph nodes;

[0050] Step S333: Perform multi-stage weight adjustment on the lesion causal trend relationship graph according to the topological association density data and the relative centrality of the graph nodes to obtain multi-stage weight adjustment data of the nodes;

[0051] Step S334: Perform graph fusion logic adjustment on the lesion causal trend relationship graph according to the topological association density data, the relative centrality of the graph nodes, and the multi-stage weight adjustment data of the nodes to obtain lesion graph fusion logic adjustment data.

[0052] The present invention first performs topological association density analysis on the lesion causal trend relationship graph. The topological association density reflects the distribution characteristics of nodes and edges in the graph and the degree of their mutual connection. By analyzing the topological structure of the graph, it is possible to identify which lesion factors or nodes have closer relationships, and these factors play more important roles in the spread and progression of the lesion. The topological association density data reveals the tightness of the association of each lesion factor, providing a basis for the analysis and adjustment of subsequent steps. For example, some nodes show a high degree of aggregation, indicating that these factors play key roles in the lesion process. Based on the topological association density data, the relative centrality of the nodes in the graph is evaluated. The centrality of a node refers to the importance and influence of the node in the entire graph structure, usually measured by its connection strength with other nodes. A higher centrality value means that the node has an important position in the lesion causal chain and is a key lesion triggering factor or a key path node. By evaluating the relative centrality of the graph nodes, it is possible to help identify the most influential factors in the lesion process. According to the topological association density data and the relative centrality of the nodes, multi-stage weight adjustment of the nodes is performed. This process can dynamically adjust the weights of each node in different lesion stages. For example, in some stages, certain lesion factors become the main influencing factors, while in other stages, the roles of these factors weaken and other factors become more important. Through multi-stage weight adjustment, the intensity of the roles of each node in the lesion causal graph at different times can be more accurately reflected, thus helping to identify and track the key stages of lesion development. By combining the topological association density data, the relative centrality of the nodes, and the multi-stage weight adjustment data of the nodes, comprehensive graph fusion logic adjustment is performed on the lesion causal trend relationship graph. This adjustment process involves re-optimizing the entire graph structure, aiming to more accurately reflect the causal relationships and influence paths between different nodes and edges. The fusion logic adjustment will not only further strengthen the roles of important nodes but also balance the factor weights in each stage.

[0053] Preferably, the present invention also provides a fusion system for a urological medical knowledge graph, which is used to execute the fusion method of the urological medical knowledge graph as described above. The fusion system for the urological medical knowledge graph includes:

[0054] An image region segmentation module, configured to obtain urology historical medical record data and CT images of the lesion site; perform image region segmentation on the CT images of the lesion site according to the urology historical medical record data to obtain CT segmentation images of the lesion site;

[0055] An association analysis module, configured to perform synchronous intersection recognition of lesion trends on the CT segmentation images of the lesion site according to the urology historical medical record data to obtain synchronous intersection data of lesion trends; perform implicit association deduction on the synchronous intersection data of lesion trends to obtain causal implicit association data of lesion trends;

[0056] A graph fusion logic adjustment module, configured to perform knowledge graph representation learning on the causal implicit association data of lesion trends to obtain graph association structure learning data; perform graph fusion logic adjustment according to the graph association structure learning data to obtain lesion graph fusion logic adjustment data;

[0057] A logic strategy design module, configured to perform logic strategy design based on the lesion graph fusion logic adjustment data to obtain a lesion graph fusion logic strategy, and send the lesion graph fusion logic strategy to the cloud platform to execute the fusion method of the urology medical knowledge graph.

[0058] Preferably, the present invention further provides a computer-readable storage medium, on which a computer program is stored, characterized in that when the computer program is executed, the fusion method of the urology medical knowledge graph described in any one of the above is implemented.

[0059] The beneficial effects of the present invention are as follows. First, by collecting the historical medical record data of urology patients and combining the patient's disease background, previous diagnosis and treatment records, and clinical symptoms, detailed medical information is obtained. Then, by analyzing the CT images of the lesion site and using image segmentation technology to process the CT images, the specific area of the lesion site is extracted. This process automatically segments the CT images by using medical image processing algorithms (such as convolutional neural networks, CNNs, etc.), enabling the precise localization of the lesion area, thereby providing a high-quality data basis for subsequent disease analysis and prediction. By analyzing the historical medical record data of the patient and the CT segmentation images of the lesion site, synchronous intersection recognition of the lesion trend is performed. This means that by comparing and analyzing the CT images of the lesion site at different time points, the laws and trends of lesion progression are identified. These data can help identify potential associations during the development of the lesion and reveal the potential causal relationships between lesions. The implicit association deduction technology plays an important role here. It derives the causal relationship of the lesion through mathematical models and machine learning algorithms and generates causal implicit association data of the lesion trend, providing support for disease prediction and treatment decisions. The causal implicit association data of the lesion trend is transformed into a knowledge graph representation, and a lesion knowledge graph is constructed through graph representation learning technology. This process structurally processes the lesion data and presents the different attributes, types, development trends of the lesion, and the relationships between them in the form of a graph. Then, graph fusion technology is used to fuse and logically adjust the data in the graph, optimizing the graph structure and relationships. This process optimizes the inference path and logical derivation of the lesion trend by adjusting the connection methods of each node and edge in the graph, making the prediction model of the lesion more accurate and helping to further discover potential disease patterns and treatment plans. Based on the lesion graph fusion logic adjustment data obtained in step S3, a set of logical strategies for the lesion trend is designed. These strategies aim to provide a systematic guidance to help medical professionals make more scientific and reasonable decisions during the diagnosis and treatment process. Finally, the lesion graph fusion logic strategy will be sent to the cloud platform for execution. Through the computing and processing capabilities of the cloud platform, the fusion method of the urology medical knowledge graph is implemented. This not only enhances the data processing and reasoning capabilities but also enables knowledge sharing and dynamic update through the cloud platform, thereby providing real-time and accurate medical advice and auxiliary decisions for clinicians and improving the efficiency and quality of medical services. Therefore, the present invention makes an optimization process for a traditional fusion method of a urology medical knowledge graph, solves the problems of large errors in the deduction of the lesion principle relationship and low correlation in the fusion of the knowledge graph in the traditional fusion method of a urology medical knowledge graph, reduces the error in the deduction of the lesion principle relationship, and improves the correlation in the fusion of the knowledge graph. Brief Description of the Drawings

[0060] Figure 1Schematic diagram of the step process of a fusion method for a urological medical knowledge graph;

[0061] Figure 2 is Figure 1 Schematic diagram of the detailed implementation steps of step S2 in

[0062] The realization, functional features and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. Specific embodiments

[0063] The technical method of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those skilled in the art within the scope of the present invention without creative work belong to the scope of protection of the present invention.

[0064] In addition, the accompanying drawings are only schematic diagrams of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor methods and / or microcontroller methods.

[0065] It should be understood that although the terms "first", "second", etc. may be used here to describe each unit, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit can be called the second unit, and similarly the second unit can be called the first unit. The term "and / or" used here includes any and all combinations of one or more of the listed associated items.

[0066] To achieve the above object, please refer to Figures 1 to 2 , a fusion method for a urological medical knowledge graph, the method comprising the following steps:

[0067] Step S1: Obtain urological historical medical record data and CT images of the lesion site; perform image region segmentation on the CT images of the lesion site according to the urological historical medical record data to obtain CT segmentation images of the lesion site;

[0068] Step S2: Identify the synchronous intersection of lesion trends for the CT segmentation images of the lesion sites based on the historical urology medical record data, and obtain the synchronous intersection data of lesion trends; perform implicit association deduction on the synchronous intersection data of lesion trends to obtain the causal implicit association data of lesion trends;

[0069] Step S3: Perform knowledge graph representation learning on the causal implicit association data of lesion trends to obtain the learning data of graph association structure; perform graph fusion logic adjustment according to the learning data of graph association structure to obtain the adjusted data of lesion graph fusion logic;

[0070] Step S4: Design a logical strategy based on the adjusted data of lesion graph fusion logic to obtain the fusion logic strategy of lesion graph, and send the fusion logic strategy of lesion graph to the cloud platform to execute the fusion method of the urology medical knowledge graph.

[0071] In the embodiment of the present invention, with reference to Figure 1 As shown, it is a schematic diagram of the step flow of a fusion method for a urology medical knowledge graph of the present invention. In this example, the fusion method for the urology medical knowledge graph includes the following steps:

[0072] Step S1: Obtain the historical urology medical record data and the CT images of the lesion sites; perform image region segmentation on the CT images of the lesion sites according to the historical urology medical record data to obtain the CT segmentation images of the lesion sites;

[0073] In the embodiment of the present invention, during the process of obtaining the historical urology medical record data and the CT images of the lesion sites, the medical record data is first preprocessed to ensure that the personal information and sensitive data therein are desensitized. The historical urology medical record data includes the patient's basic information, symptom description, diagnosis record, treatment process, and examination results. The data is extracted from the hospital electronic medical record system through a standardized interface and subjected to structured conversion to ensure a unified data format. The CT images of the lesion sites are obtained from the hospital's medical image management system. The image data usually adopts the DICOM format and must be subjected to format conversion and size unification to ensure that the resolution and gray value of all images are consistent. Next, an image processing algorithm based on deep learning is used to perform image region segmentation on the CT images of the lesion sites. Specifically, an image segmentation algorithm based on a convolutional neural network (CNN) is adopted. First, the input image is preprocessed, and the image quality is improved through image enhancement techniques (such as contrast adjustment and edge sharpening). Then, the trained model is used for region segmentation. The goal of image segmentation is to accurately delimit the lesion sites, exclude irrelevant regions, and generate the CT segmentation images of the lesion sites. This process can accurately extract the target regions for subsequent analysis.

[0074] Step S2: Identify the synchronous intersection of lesion trends for the CT segmentation images of the lesion sites based on the historical urology clinic visit record data to obtain synchronous intersection data of lesion trends; perform implicit association deduction on the synchronous intersection data of lesion trends to obtain causal implicit association data of lesion trends;

[0075] In the embodiments of the present invention, the synchronous intersection of lesion trends is identified based on the historical urology clinic visit record data and the CT segmentation images of the lesion sites. First, the information such as disease history, symptom changes, treatment process, etc. in the medical record data is deeply analyzed to extract the disease progression trend of the patient. On the CT images, image processing techniques are used to extract different morphological features of the lesion area, such as the edge, shape, size, and position of the lesion area. These image features are combined with the clinical information in the historical medical record data, and the evolution trend of the lesion is analyzed through specific algorithms (such as decision trees or support vector machines SVM), and the synchronous intersection of lesion trends is identified. By comparing the CT segmentation images at different time points with the records in the medical record data, the development trend of the lesion can be identified, and synchronous intersection data of lesion trends is formed. Further, perform implicit association deduction on the synchronous intersection data of lesion trends, and use association rule mining methods, such as the Apriori algorithm or the FP-growth algorithm, to find potential causal relationships. These causal relationships help to infer the implicit factors of the lesion trend and provide a basis for the subsequent construction of the knowledge graph. During the deduction process, the algorithm generates causal implicit association data of lesion trends by calculating the dependence relationship between events in the lesion progression, revealing the potential connections between different lesion factors.

[0076] Step S3: Perform knowledge graph representation learning on the causal implicit association data of lesion trends to obtain graph association structure learning data; perform graph fusion logic adjustment according to the graph association structure learning data to obtain lesion graph fusion logic adjustment data;

[0077] In the embodiments of the present invention, the causal implicit association data of the lesion trend will be used for knowledge graph representation learning. First, the causal implicit association data of the lesion trend is converted into graph-structured data, where nodes represent different lesion factors or pathological states, and edges represent the causal relationships between these factors or states. Then, graph embedding techniques are used to perform representation learning on the nodes and edges in the graph. Specifically, a graph embedding algorithm based on random walk is adopted, such as DeepWalk or Node2Vec. These methods learn the low-dimensional representations of nodes by simulating the random walk paths in the graph, so that similar nodes have similar vector representations in the embedding space. Through the graph representation learning of the causal implicit association data of the lesion trend, a structured representation of the lesion causal relationship can be obtained, that is, the graph association structure learning data. These learned node representations are further used for graph fusion. Specifically, the graph fusion logic is adjusted according to the graph association structure learning data. During the adjustment process, a graph alignment method is adopted. First, the node correspondence relationships in different lesion graphs are determined through graph isomorphism matching, and then the multiple manifestation forms of the same lesion node are unified, so as to obtain the graph fusion logic adjustment data for the lesion graph.

[0078] Step S4: Design a logic strategy based on the graph fusion logic adjustment data for the lesion graph to obtain a graph fusion logic strategy for the lesion graph, and send the graph fusion logic strategy for the lesion graph to the cloud platform to execute the fusion method of the urology medical knowledge graph.

[0079] In the embodiments of the present invention, based on the graph fusion logic adjustment data for the lesion graph, a graph fusion logic strategy for the lesion graph is designed. First, by analyzing the graph fusion logic adjustment data for the lesion graph, important causal nodes and their mutual relationships during the lesion development process are identified and converted into executable logic strategies. When designing the logic strategy, a decision tree algorithm is used to divide different lesion development paths, and different intervention measures are selected according to different lesion stages. The design of the logic strategy not only considers the temporal progression of the lesion, but also combines the individual differences of the patient, such as factors like age, gender, and genetic background. After multiple simulations and adjustments, a graph fusion logic strategy suitable for clinical application is finally designed. After the design is completed, this strategy is sent to the cloud platform through the system interface. The cloud platform is responsible for executing this strategy and continuously optimizing the strategy execution effect through a real-time data feedback mechanism to ensure the high efficiency and accuracy of the fusion method of the urology medical knowledge graph in practical applications.

[0080] Preferably, step S1 includes the following steps:

[0081] Step S11: Obtain urology historical medical record data and CT images of the lesion site;

[0082] Step S12: Perform data desensitization processing on the urology historical medical record data to obtain desensitized data of the historical medical record.

[0083] Step S13: Perform logical description parsing on the desensitized data of historical medical records to obtain medical record logical description data;

[0084] Step S14: Perform image region segmentation on the CT images of the lesion sites according to the medical record logical description data to obtain CT segmentation images of the lesion sites.

[0085] In the embodiments of the present invention, it starts from obtaining urological historical medical record data and CT images of the lesion site. First, the urological historical medical record data is extracted from the hospital's electronic medical record management system. The medical record data includes the patient's medical history, symptom description, diagnosis result, treatment process, examination records, etc. The data extraction uses a standardized interface to ensure the accurate transmission of data. For the CT images of the lesion site, the image data is usually saved in DICOM format and extracted from the medical image management system. These CT image data contain the lesion scan results of the patient at different time points and usually record the image information of the lesion site. Through the image data management system, all CT images are identified and stored according to the patient ID, and the correctness and integrity of the image data are ensured. The data sources of the medical record data and CT images are the medical information system and the image storage management system respectively. During the extraction process, a preliminary quality check is performed on the data to ensure that there is no data loss or format error. The urological historical medical record data is subjected to data desensitization processing. The purpose of desensitization processing is to protect patient privacy while maintaining the usability of the data. First, all personal identity information is removed from the historical medical record data, including the patient's name, ID number, contact information, etc. These personal sensitive information is replaced with irrelevant identifiers or random codes. In addition, data obfuscation processing is also required for specific areas involved in the medical history description. For example, standardize the disease names, treatment plans, and drug names, and adopt a unified coding system (such as the ICD-10 coding system). For non-public information in other medical record contents, encryption processing is adopted to ensure the security during data transmission. During the data desensitization process, it is necessary to ensure that the desensitized data will not affect the accuracy of subsequent analysis. Therefore, the desensitization process uses substitution or masking of patient identity information and sensitive content instead of deletion, so as to ensure the integrity of the data and the effectiveness of subsequent analysis. Perform logical description parsing on the desensitized data of the historical medical record. First, perform text preprocessing on the desensitized medical record data, including removing stop words, standardizing disease terms, etc. Then, use natural language processing (NLP) technology to parse the medical record text, identify the key information in the medical record, such as symptoms, diagnoses, treatment plans, etc., and convert them into structured data. Through lexical analysis, entity recognition, and relationship extraction techniques, the logical description information in the medical record data is obtained. For example, identify "the patient has symptoms of frequent urination" as a symptom description, or "diagnosed with cystitis" as a disease diagnosis record. For the time information in the medical record data, it is also necessary to use time series analysis methods to time-process the patient's visit date and medical history record for subsequent analysis. The main tools used in this process are rule-based dictionary matching methods and deep learning models, especially using models based on BERT (Bidirectional Encoder Representations from Transformers) for semantic understanding of medical texts.The output of this step is structured logical description data, which contains various symptoms, diagnoses, and treatment information in the medical record, and standardizes and times these information to support the subsequent CT image region segmentation. Perform image region segmentation on the CT image of the lesion site according to the logical description data of the medical record for the visit. First, use the medical record information obtained in step S13 to determine the potential location and type of the lesion site. This information mainly includes the type of disease, the region of the lesion, and relevant hints in the symptom description. For example, if it is marked as "bladder tumor" in the patient's medical record, then the lesion manifestations in the bladder region need to be concerned in the CT image. Based on this medical record information, perform region calibration of the CT image through an image segmentation algorithm. In this process, use an edge detection-based image segmentation method, such as the Canny edge detection algorithm, to perform preliminary region division by extracting the contour of the lesion region. Next, use the region growing algorithm to further refine the lesion region to ensure the accuracy of the segmentation result. The region growing algorithm starts from the lesion site and expands outward by setting seed points until the entire lesion region is accurately segmented. Finally, use a convolutional neural network (CNN) based on deep learning to optimize the segmentation result. The CNN model can accurately identify the shape, size, and edge of the lesion region by learning a large amount of labeled data. Through these steps, finally obtain the CT segmentation image of the lesion site, that is, the accurate division result of the lesion region in the CT image, providing a basis for the subsequent analysis of the lesion trend.

[0086] Preferably, step S2 includes the following steps:

[0087] Step S21: Perform structural and morphological analysis on the CT segmentation image of the lesion site to obtain the structural and morphological data of the lesion site;

[0088] Step S22: Perform synchronous intersection recognition of the lesion trend on the structural and morphological data of the lesion site and the CT segmentation image of the lesion site according to the historical urology visit medical record data to obtain the synchronous intersection data of the lesion trend;

[0089] Step S23: Perform causal association analysis on the synchronous intersection data of the lesion trend to obtain the synchronous causal data of the lesion trend;

[0090] Step S24: Perform implicit association deduction according to the synchronous causal data of the lesion trend to obtain the causal implicit association data of the lesion trend.

[0091] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes:

[0092] Step S21: Perform structural and morphological analysis on the CT segmentation image of the lesion site to obtain the structural and morphological data of the lesion site;

[0093] In the embodiments of the present invention, structural and morphological analysis is performed on the CT segmentation images of the lesion site. First, the CT segmentation images of the lesion site are preprocessed through image processing algorithms to ensure that the image quality is suitable for subsequent analysis. This includes operations such as image denoising, contrast enhancement, and edge sharpening. The median filtering method is used to remove the noise in the image and enhance the important details in the image. Then, morphological analysis methods are applied to analyze the structure of the lesion site. Through techniques such as dilation, erosion, opening operation, and closing operation in morphological analysis, geometric features of the lesion area are extracted, such as the shape, size, and edge smoothness of the lesion area. This process uses morphological operators to extract the key structural information in the image. After obtaining the structural features, geometric feature extraction methods (such as rectangle fitting, circle fitting, etc.) are used to further analyze the shape of the lesion area to obtain the structural and morphological data of the lesion site. The structural and morphological data includes the geometric shape, size, edge features, etc. of the lesion, providing structured input data for subsequent lesion trend analysis and causal deduction.

[0094] Step S22: Identify the synchronous intersection of lesion trends for the structural and morphological data of the lesion site and the CT segmentation images of the lesion site based on the historical urology medical record data, and obtain the synchronous intersection data of lesion trends;

[0095] In the embodiments of the present invention, the synchronous intersection of lesion trends for the structural and morphological data of the lesion site and the CT segmentation images of the lesion site is identified based on the historical urology medical record data. First, disease information, symptom descriptions, examination records, etc. in the medical record data are extracted to identify features related to the lesion, such as the location, type of the lesion, and the patient's symptoms. Based on this information, combined with the CT segmentation images of the lesion site, the location and morphological features of the lesion area are identified. Then, multi-dimensional data fusion technology is adopted to compare and match the structural and morphological data of the CT images of the lesion site with the lesion descriptions in the medical record data to identify the intersection of lesion trends. This process uses data synchronization identification algorithms, such as the method based on dynamic time warping (DTW). By comparing the image segmentation data with the lesion time series in the medical record, the trend of lesion changes is detected. For example, if the lesion area shows a gradual increase in the CT images at different time points of the patient, and the medical record data describes symptoms such as "disease progression" or "lesion spread", the intersection between these data can be determined, thus obtaining the synchronous intersection data of lesion trends. The synchronous intersection data of lesion trends records the change trend of the lesion area, the consistency with the medical record description, and its potential evolution pattern.

[0096] Step S23: Perform causal association analysis on the synchronous intersection data of lesion trends to obtain the synchronous causal data of lesion trends;

[0097] In the embodiments of the present invention, causal association analysis is performed on the synchronous intersection data of the lesion trends. The causal association analysis aims to find out the causal relationship of the lesion changes by analyzing the synchronous intersection data of the lesion trends. First, preparations for causal inference are made on the synchronous intersection data of the lesion trends, and the characteristics of the changes in the lesion area are extracted, including the morphological changes, size changes of the lesions, and the associations with the symptom manifestations. Statistical methods, such as correlation coefficient analysis and hypothesis testing, are used to analyze the correlations between the changes in the lesion area and the patient's symptoms, diagnoses, and treatment plans. In addition, the Bayesian network method is used for causal reasoning to establish a causal relationship model between the lesion trends and the symptom manifestations. The Bayesian network can obtain the causal relationship between the lesion development and the medical history data by calculating the conditional probability. For example, through speculation by the Bayesian network algorithm, whether the enlargement of certain lesion sites is related to an increased risk of bladder cancer, or whether the appearance of a certain symptom (such as hematuria) directly leads to the exacerbation of the lesion. The synchronous causal data of the lesion trends output by this step includes the change patterns of the lesions, the inducing factors of the symptoms, and the causal paths during the lesion evolution process.

[0098] Step S24: Perform implicit association deduction based on the synchronous causal data of the lesion trends to obtain the causal implicit association data of the lesion trends.

[0099] In the embodiments of the present invention, implicit association deduction is performed based on the synchronous causal data of the lesion trends. The goal of the implicit association deduction is to discover potential and unmanifested association patterns by analyzing the causal association data. First, statistical inference methods are used to deeply mine the synchronous causal data of the lesion trends. Through techniques such as cluster analysis and principal component analysis (PCA), the implicit laws of the lesion development are discovered. Cluster analysis can group cases with similar characteristics of the changes in the lesion area, revealing the associations between different types of lesion areas. Principal component analysis, on the other hand, extracts the main change patterns in the lesion data by reducing the dimensionality, thereby revealing the potential factors that are not manifested during the lesion development process. Then, association rule mining algorithms (such as the Apriori algorithm) are used to mine the hidden laws and associations from the causal data. For example, in the CT images of certain patients, a lesion of a certain morphology is shown, and there is no direct description of the relationship between this morphology and the symptoms in the medical records. Through association rule mining, the potential relationship between this lesion morphology and other symptoms can be deduced. The results of the implicit association deduction provide more insights into the lesion evolution process, help speculate on the future development trends of the lesions, and provide a more accurate basis for medical decision-making.

[0100] Preferably, step S22 includes the following steps:

[0101] Step S221: Perform spatial domain analysis on the structural morphology data of the lesion site to obtain the lesion position distribution data;

[0102] Step S222: Conduct a descriptive trend comparison of the lesion location distribution data based on the historical urology medical record data to obtain the descriptive trend comparison data of the lesion distribution;

[0103] Step S223: Simulate the similarity difference in tissue density at the lesion boundary for the lesion location distribution data and the CT segmentation images of the lesion sites based on the descriptive trend comparison data of the lesion distribution to obtain the similarity difference in tissue density at the lesion boundary;

[0104] Step S224: Analyze the density difference in the boundary transition of the lesion location distribution data based on the similarity difference in tissue density at the boundary tissue to obtain the density difference in the boundary transition of the lesion tissue;

[0105] Step S225: Identify the synchronous intersection of the lesion trends for the structural morphology data of the lesion sites based on the similarity difference in tissue density at the lesion boundary and the density difference in the boundary transition of the lesion tissue to obtain the synchronous intersection data of the lesion trends.

[0106] In the embodiments of the present invention, the structural and morphological data of the lesion site will be processed through spatial domain analysis. First, the structural and morphological data of the lesion site, including the geometric shape, size, distribution, etc. of the lesion area, will be converted into spatial data in a three-dimensional coordinate system. By analyzing the distribution of the lesion area in three-dimensional space, a spatial clustering algorithm (such as the K-means clustering algorithm) is used to group the lesion areas, and the spatial positions of each type of lesion are determined. Next, spatial interpolation techniques, such as the inverse distance weighted (IDW) method, are used to estimate the density of the lesion area, and then the position distribution data of the lesion area in space is obtained. The goal of this step is to accurately represent the spatial distribution of the lesion area, facilitating subsequent lesion trend analysis and comparison with historical medical record data. The lesion position distribution data will include the spatial position coordinates, distribution density, and the probability of the presence of lesions at different spatial positions, providing the necessary spatial information for subsequent synchronous intersection recognition of lesion trends. Based on the historical urology clinic record data, a trend comparison analysis is performed between the lesion position distribution data and the relevant descriptions in the medical records. First, the medical history information of the patients in the medical records, such as past medical history, symptom records, clinical diagnoses, and treatment results, is extracted, with particular attention paid to the time series data related to the lesion site. By comparing the medical record data of different patients, the time evolution trend of the lesion position is identified. The lesion position distribution data is mapped to the lesion description part in the medical record data for quantitative analysis. Descriptive statistical methods, such as mean, variance, and trend tests (such as Kendall rank correlation test), are used to perform trend comparison analysis on the data of different lesion positions, identifying the laws and changing trends of the lesion distribution. The descriptive trend comparison data of the lesion distribution will show the association between the spatial distribution changes of the lesion site and the time series of symptoms, diagnoses, and treatments described in the medical record. For example, whether the gradual expansion of a certain lesion position is consistent with the deterioration of symptoms or the change in treatment effect. This data will help predict the future trend of the lesion position distribution and provide data support for subsequent lesion boundary analysis. Based on the descriptive trend comparison data of the lesion distribution, a simulation of the tissue density similarity difference of the lesion boundary between the lesion position distribution data and the CT segmentation image of the lesion site is performed. First, the density values of each pixel area in the CT segmentation image of the lesion site are extracted to represent the density differences of different tissue types (such as normal tissue and lesion tissue). By comparing the differences in tissue density, the boundary between the lesion area and the surrounding tissue is identified. Then, similarity measurement methods, such as cosine similarity or Euclidean distance, are used to compare the density of the lesion boundary with that of the surrounding tissue, and the tissue density similarity difference of the lesion area is calculated. This process can determine the clarity of the lesion area boundary by comparing the density differences between adjacent pixel points. By simulating the tissue density differences of the lesion boundary, the density similarity difference value of the lesion boundary is obtained, which can quantify the difference between the lesion area boundary and the normal tissue and reflect the transitional nature between the lesion area and the surrounding tissue.This data will help to reveal the clarity of the lesion boundary and its histological characteristics compared with the surrounding tissues. Based on the density similarity difference of the lesion boundary tissues, a boundary transition density difference analysis of the lesion location distribution data is carried out. First, the boundary information of the lesion area is extracted, and the tissue density near the boundary is carefully analyzed to determine the transition area at the boundary. Within this area, there is a gradual change in the density between normal and lesion tissues. Therefore, by calculating the density difference within the boundary area, the density change in the transition area can be quantified. For further analysis, a gradient algorithm is used to analyze the density change at the lesion boundary, calculating the density gradient at each pixel point and statistically analyzing the density differences in different regions. This process uses an image gradient operator (such as the Sobel operator) to extract the transition characteristics of the lesion boundary, reflecting the tissue transition density difference in the lesion area. The goal of the boundary transition density difference analysis is to reveal the transition characteristics between the lesion area and normal tissues and evaluate the boundary transition characteristics of lesion development. The finally obtained lesion tissue boundary transition density difference data can describe in detail the transition process between the lesion area and the surrounding tissues and provide necessary information for further identification of the synchronous intersection of lesion trends. Based on the density similarity difference of the lesion boundary tissues and the lesion tissue boundary transition density difference, a synchronous intersection identification of the lesion trend for the lesion site structural morphology data is carried out. First, the density similarity difference and boundary transition density difference data of the lesion boundary are combined with the structural morphology data of the lesion site to form a multi-dimensional data set. This data set includes information such as the morphological characteristics of the lesion (such as size, shape, location), the clarity of the boundary, and the density transition of the tissues. By applying data fusion techniques, this information is weighted and integrated to more comprehensively describe the change trend of the lesion area. Then, trend analysis methods such as dynamic time warping (DTW) and collaborative filtering algorithms are used to synchronously analyze the lesion data at different time points to find the intersection part between the lesion trends. This analysis process can reveal the evolution law of the lesion area in time and space and the potential direction of lesion development. The synchronous intersection data of the lesion trend will ultimately provide the evolution pattern of the lesion area, helping to identify the characteristic changes of the lesion at different stages, thus providing key clinical evidence for the construction of the urology medical knowledge graph.

[0107] Preferably, step S223 includes the following steps:

[0108] Perform a multi-scale spatial coordinate transformation on the lesion location distribution data to obtain multi-scale lesion area calibration data;

[0109] Based on the lesion distribution descriptive trend comparison data, perform a multi-point adjacent density distribution analysis on the multi-scale lesion area calibration data and the CT segmentation image of the lesion site to obtain multi-point adjacent density distribution data;

[0110] Perform a hierarchical density fitting on the multi-point adjacent density distribution data to obtain hierarchical lesion boundary density data;

[0111] Based on the hierarchical lesion boundary density data, perform position density trend deviation analysis to obtain position density trend deviation data;

[0112] According to the hierarchical lesion boundary density data and the position density trend deviation data, perform a simulation of the tissue density similarity difference at the lesion boundary for the lesion location distribution data and the CT segmentation image of the lesion site to obtain the tissue density similarity difference at the lesion boundary.

[0113] In the embodiments of the present invention, multi-scale spatial coordinate transformation is performed on the spatial coordinate information in the lesion location distribution data, so as to accurately calibrate the lesion area through the spatial information at different scales. The Scale-Invariant Feature Transform (SIFT) method is adopted to identify the characteristic positions of the lesion area at different resolutions through multi-level spatial analysis. Specifically, each spatial coordinate point in the lesion location distribution data is converted into local features at different scales, and multiple scale layers are generated through a series of image transformation functions (such as Gaussian pyramid and Laplacian pyramid). For each scale layer, the adjacent-level image difference analysis is used to obtain the fine positioning of the lesion location. Then, combining the spatial transformation matrices of each scale layer, the transformed data is fused to obtain the final multi-scale lesion area calibration data. This process can effectively improve the positioning accuracy of the lesion area at different resolutions and provide efficient and accurate basic data for subsequent density analysis. By analyzing the distribution of the lesion area at multiple scales, multi-point adjacent density distribution analysis is carried out. First, the descriptive trend comparison data of the lesion distribution is combined with the multi-scale lesion area calibration data to determine the positions of the center point and boundary points of the lesion area. The proximity measurement method based on point sets (such as k-nearest neighbor algorithm) is used to analyze the adjacency relationship within the lesion area and between it and the surrounding tissues. By calculating the density difference between each lesion point and its adjacent area, the adjacent density distribution of each point is obtained. In this process, first, the CT segmentation image of the lesion area is extracted, and the density characteristics of the lesion boundary and its adjacent tissues are analyzed. Then, the weighted density distribution function is used to perform weighted calculation on each adjacent position to obtain more accurate density distribution data. This analysis can not only accurately reveal the spatial adjacency relationship of the lesion area, but also effectively reflect the density difference between the lesion area and the surrounding tissues, providing detailed spatial data support for further lesion boundary analysis. Based on the multi-point adjacent density distribution data, the density of the lesion area is fitted in layers. First, according to the density distribution characteristics of the lesion area, the lesion area is divided into multiple layers, usually based on the hierarchical characteristics of the lesion boundary. Through the Gaussian Mixture Model (GMM) or other hierarchical density models, the density of the lesion area is fitted to determine the density change trend of the lesion area at different layers. Specifically, each layer represents the density distribution of different depths or different adjacent relationships in the lesion area, and the density change of the lesion area is fitted at different layers. By fitting the density distribution of each layer, a fine-layered lesion boundary density data is obtained. This process can identify the density differences at different layers within the lesion area, help accurately describe the boundary characteristics and change rules of the lesion area, and provide data support for subsequent lesion boundary recognition. Based on the layered lesion boundary density data, position density trend deviation analysis is carried out on the density change trend of the lesion area. First, according to the hierarchical density characteristics of the lesion boundary, the density change trend of different positions in the lesion area is calculated, and the density deviation between layers is quantified.Use a difference equation or a gradient algorithm (such as weighted least squares method) to calculate the density change trend at different positions. By comparing the density changes at each position in the lesion area, analyze the deviation degree of the density trend at different positions in the lesion area. This process can effectively evaluate whether the density changes at the lesion boundary are consistent, identify the areas with large density deviations, and thus infer the expansion direction or the lesion trend of the lesion. Finally, the obtained position information and density deviation data will provide a quantitative basis for the subsequent precise simulation of the lesion boundary and the analysis of the lesion trend. Using the hierarchical lesion boundary density data and the position density trend deviation data, simulate the similarity difference of the tissue density at the lesion boundary for the lesion position distribution data and the CT segmentation image of the lesion site. First, combine the previously obtained hierarchical lesion boundary density data with the position density trend deviation data to generate a comprehensive density difference data model. Then, through a tissue density similarity calculation method (such as mean square error or structural similarity index SSIM), simulate the density difference between the lesion area and the surrounding tissues. Specifically, the tissue density difference at the lesion boundary will be represented by the difference in pixel values in the CT image, and the similarity difference is calculated by comparing the density changes between the lesion area and the surrounding tissues. The result of the simulation will provide a quantitative basis for the boundary clarity of the lesion area, help identify the transition area between the lesion and the surrounding tissues, and the expansion trend of the lesion. The finally obtained lesion boundary tissue density similarity difference data will provide basic support for the subsequent refined identification and trend analysis of the lesion boundary, and assist in constructing the lesion trend part in the urology medical knowledge graph.

[0114] Preferably, step S24 includes the following steps:

[0115] Step S241: Structurally classify the synchronous causal data of the lesion trend to obtain the data for dividing the lesion development stage;

[0116] Step S242: Deduce the stage trigger factors based on the data for dividing the lesion development stage to obtain the data for deducing the lesion stage trigger factors;

[0117] Step S243: Layer-by-layer factor event simulation and analysis of the data for deducing the lesion stage trigger factors to obtain the data for layer-by-layer factor event analysis;

[0118] Step S244: Implicit association deduction based on the data for layer-by-layer factor event analysis and the data for deducing the lesion stage trigger factors to obtain the implicit association data of the lesion trend causality.

[0119] In the embodiments of the present invention, first, based on the causal data synchronized with the lesion trend, through in-depth analysis of the causal data, the different development stages of the lesion are classified and processed in a structured manner. In specific operations, first, various causal relationships in the lesion data are identified and marked, including the starting point, development process, and ending point of the lesion. The clustering analysis algorithm (such as the K-means clustering method) is used to group these causal data, and similar causal data are classified into the same stage. Each development stage is detailedly divided according to the key time nodes, clinical manifestations, and relevant pathological data of the lesion. Through the time series analysis of the lesion data, combined with the clinical manifestations of the lesion and the changes in the lesion area of the CT image, several representative development stages are divided, such as the initial stage, development stage, and late stage of the lesion. Finally, the divided data of the lesion development stage provides a clear phased basis for subsequent stage deduction and factor analysis, which can help accurately identify the characteristics of different stages of the lesion and provide data support for the reasoning and deduction of the knowledge graph. Based on the divided data of the lesion development stage, the deduction of the triggering factors of the lesion stage is carried out. By deeply analyzing the characteristics of the lesion and clinical data at each stage, the triggering factors related to the lesion process are extracted. For example, in the initial stage of the lesion, it is affected by certain specific external factors (such as infection, trauma, etc.), while in the late stage of the lesion, it is closely related to genetic factors or long-term chronic medical history. By constructing a multivariate regression model, the triggering factors at each stage of the lesion and their interactions are analyzed, and the deduction is carried out stage by stage. In the specific deduction process, combined with the triggering factors, lesion types, and clinical treatment data in the historical medical records, the internal and external triggering factors in the lesion development process are identified. Using the method of inductive reasoning, by comparing the clinical data of different lesion stages, the causes and triggering events at each stage are obtained, and finally, the deduction data of the triggering factors of the lesion stage are formed. This data will be used as the basis for the next layer-by-layer factor analysis to support the causal deduction of the lesion trend. Based on the deduction data of the triggering factors of the lesion stage, the layer-by-layer factor event simulation and analysis are carried out. First, according to the deduced lesion triggering factors, at different development stages of the lesion, the triggering factors at each stage are analyzed and simulated layer by layer. By simulating the events and changes triggered by different triggering factors at each stage of the lesion, an event chain is gradually constructed. The event correlation relationship between each factor is represented by a causal graph model, and the Bayesian network algorithm is used to calculate the occurrence probability of the event. This model can capture the dependence relationship between different factors and predict future events based on the time series relationship of the lesion triggering factors. Through layer-by-layer simulation, the key events at different stages of the lesion and their transformation paths are analyzed. In specific operations, first, the triggering factors at each stage are summarized, and then through the probability model of the event occurrence, the influence of different triggering factors on the lesion progression is simulated, and finally, the layer-by-layer factor event analysis data are obtained. This data reflects the hierarchical changes in the lesion process and provides a detailed event chain for the implicit association deduction.Combine the layer-by-layer factor event analysis data and the lesion stage trigger factor deduction data to perform implicit association deduction. Through the causal deduction method, the implicit associations that are not directly observed in the lesion trend are identified. First, the events in the layer-by-layer factor event analysis data and the deduced trigger factor data are combined, and the potential causal associations between different stages are analyzed through statistical analysis methods (such as principal component analysis or causal relationship inference). Specifically, the time series analysis method is used to explore the implicit patterns in the development of the lesion, such as some potential risk factors gradually accumulate at a certain stage and eventually trigger the aggravation of the lesion. Through the complex network analysis model (such as the convolutional network, GCN), the causal relationship of the lesion trend is deduced to identify the implicit connection between the various factors. Through this process, the potential causal relationship that is not directly identified in the lesion development process can be found, and then the causal implicit association data of the lesion trend can be obtained. This data will provide important support for the in-depth analysis of lesion trends and the prediction of future lesion risks, and provide more comprehensive causal deduction information for the construction of the urology medical knowledge graph.

[0120] Preferably, step S3 comprises the following steps:

[0121] Step S31: constructing a graph relationship for the lesion trend causal implicit correlation data to obtain a lesion causal trend relationship graph;

[0122] Step S32: Perform knowledge graph representation learning on the lesion causal trend relationship graph to obtain graph association structure learning data;

[0123] Step S33: Performing graph fusion logic adjustment on the lesion causal trend relationship graph according to the graph association structure learning data to obtain lesion graph fusion logic adjustment data.

[0124] In the embodiments of the present invention, first, a graph relationship construction is performed on the causal implicit association data of the lesion trend. The goal of the graph relationship construction is to form structured graph data by analyzing the causal implicit association data of the lesion, where each node represents a lesion causal event or a lesion stage, and each edge represents the association relationship between different lesion causal events or stages. Specifically, in implementation, using the concepts of nodes and edges in graph theory, the causal relationships among lesion stages, triggering factors, and lesions are represented as a graph structure. The attributes of each node include the specific stage of the lesion, the occurrence time, the triggering factor, clinical data, etc., and the attributes of the edge describe the strength and type of the association relationship (such as causal relationship, influence relationship, etc.). In this process, the graph convolutional network (GCN) algorithm in deep learning is used to learn the potential relationships between nodes. This method can gradually construct the causal trend relationship graph of the lesion from local information according to the implicit association data. During the graph construction process, by hierarchically analyzing the progression path of the lesion, the temporal relationship, influence, and dependency between different lesion stages and events are determined to form a complete causal trend relationship graph of the lesion. Based on the causal trend relationship graph of the lesion obtained in step S31, knowledge graph representation learning is performed to obtain graph association structure learning data. Specifically, when operating, graph embedding technology is used to perform representation learning on the causal trend relationship graph of the lesion, and each node and edge in the graph are transformed into a low-dimensional vector representation. This step uses graph embedding algorithms such as TransE or Node2Vec. By mapping nodes and edges to a low-dimensional space, similar nodes and edges have similar representations in the low-dimensional space. Specifically, the graph embedding algorithm first calculates the embedding vector of each node according to the structural relationship of the nodes and edges in the graph, and then optimizes the embedding vector by optimizing the objective function (such as minimizing the distance between the embedding vector and the adjacency relationship) until the relationship between nodes is effectively represented in the embedding space. Through this process, the association patterns of nodes and edges in the graph can be effectively captured to form graph association structure learning data, which provides a basis for further graph fusion and knowledge reasoning. During the knowledge graph representation learning process, the optimization method used is an optimization technique based on negative sampling to improve learning efficiency and reduce computational complexity. Based on the graph association structure learning data, logical adjustment of graph fusion is performed on the causal trend relationship graph of the lesion to obtain lesion graph fusion logical adjustment data. The purpose of graph fusion is to merge and optimize lesion data from different sources to form a unified and complete lesion graph. First, the causal trend relationship graph of the lesion is analyzed to identify redundant and conflicting information between different data sources. By comparing similar or duplicate nodes and edges in the graph, a graph aggregation algorithm, such as the information transfer mechanism in the graph convolutional network (GCN), is used to integrate the information in multiple data sources, eliminate redundancy, and optimize the representation of nodes and edges.During the atlas fusion process, logical adjustments also need to be made according to the actual laws of lesion development to ensure that the causal relationships and influence paths between different stages are reasonable. For example, if contradictions or inconsistencies occur in the data of a certain lesion stage during the fusion process, constraint conditions can be used for correction and optimization, and these conflicts can be eliminated by adjusting the weights between nodes, the directions or types of edges in the atlas. Finally, through atlas fusion, the logical adjustment data of the lesion atlas fusion is obtained, and these data reflect the comprehensive trends and causal relationships of the lesions, providing accurate structural data support for the further application and reasoning of the urology medical knowledge atlas.

[0125] Preferably, step S33 includes the following steps:

[0126] Step S331: Perform a topological association density analysis on the lesion causal trend relationship atlas according to the atlas association structure learning data to obtain topological association density data;

[0127] Step S332: Evaluate the relative centrality of the atlas nodes of the lesion causal trend relationship atlas based on the topological association density data to obtain the relative centrality of the atlas nodes;

[0128] Step S333: Adjust the multi-stage weights of the nodes of the lesion causal trend relationship atlas according to the topological association density data and the relative centrality of the atlas nodes to obtain the multi-stage weight adjustment data of the nodes;

[0129] Step S334: Perform logical adjustment of the atlas fusion on the lesion causal trend relationship atlas according to the topological association density data, the relative centrality of the atlas nodes, and the multi-stage weight adjustment data of the nodes to obtain the logical adjustment data of the lesion atlas fusion.

[0130] In the embodiments of the present invention, first, a topological association density analysis is performed on the lesion causal trend relationship graph. The goal is to reveal the closeness and its distribution characteristics between nodes and edges in the graph, and obtain topological association density data. Topological association density is an index to measure the correlation strength and compactness between nodes in the graph, which is usually completed by calculating the degree of each node (i.e., the number of edges directly connected to the node) and the connection pattern of the edges. Specifically, the adjacency matrix in graph theory is used to represent the graph, and the weighted calculation of the matrix is used to quantify the correlation strength between nodes. By analyzing the mutual connections between nodes, it is determined which nodes are in the core position and which nodes are in the peripheral position in the graph. This analysis is based on the adjacency relationship of nodes and the weights of edges, and uses standard topological density calculation methods, such as degree centrality of nodes and edge density, etc., and combines with the density formula of the graph to calculate the local and global densities of the entire graph. The topological association density data provides the connection compactness of different regions in the graph, providing a basis for subsequent node evaluation and graph optimization. Based on the topological association density data, in this step, a relative centrality evaluation is performed on the nodes in the lesion causal trend relationship graph to obtain the relative centrality of the graph nodes. Node centrality is an index to describe the importance of a node in the graph, and its evaluation basis is the connection situation of the node with other nodes. During the evaluation, first, the degree centrality of each node in the graph is calculated (i.e., the number of edges directly connected to the node), and at the same time, the position of the node in the topological structure is considered. The weighted centrality method is used to weight the influence of each node through the topological association density data, and the direct and indirect relationships between the node and other nodes are evaluated to ensure that the centrality evaluation not only depends on the degree of the node, but also considers the network position where the node is located. Specific evaluation methods include traditional graph centrality indexes such as Betweenness (betweenness centrality) and Closeness (closeness centrality), combined with the topological density information of the graph, to evaluate the influence of the node in the entire lesion causal relationship graph. The relative centrality is to calculate the centrality ranking of each node by comparison among all nodes, and sort according to the importance of the nodes, and finally obtain the relative centrality data of each node. Based on the topological association density data and the relative centrality of the graph nodes, a multi-stage weight adjustment is performed on the nodes in the lesion causal trend relationship graph. First, according to the topological association density data and node centrality obtained in steps S331 and S332, the importance of each node in different lesion stages is determined, and corresponding weights are assigned. The logic followed by the weight adjustment is that nodes in the core area and with a high centrality have a strong influence in the early stage of lesion development, while nodes located in the periphery or with a lower importance gradually weaken their influence in the later stage. By calculating the weighted values of each node in each stage, the role and relationship of the nodes in the graph are adjusted to ensure that the node weights in each stage are more in line with the development trend of the lesion.Specifically, the weighted graph method is adopted to adjust the weight values of each node. The weight value of a node varies with the stage it is in during the progression of the lesion. By combining the topological density analysis and the evaluation data of node centrality, it is ensured that the weight adjustment of the node is carried out according to the relative importance of the node and the stage of the lesion process it is in, thereby obtaining the multi-stage weight adjustment data of the node. Based on the topological correlation density data, the relative centrality of the atlas nodes, and the multi-stage weight adjustment data of the nodes, the atlas fusion logic adjustment of the lesion causal trend relationship atlas is carried out, and finally the lesion atlas fusion logic adjustment data is obtained. The atlas fusion logic adjustment is a process of optimizing the data in the lesion causal trend relationship atlas, aiming to integrate the causal relationships and influence paths of each node. First, using the topological correlation density data and the atlas structure evaluated by node centrality, combined with the multi-stage weight adjustment data of the nodes, analyze the interaction and dependence relationships between the nodes in the atlas. Then, by adjusting the nodes and their connection relationships, optimize the causal connections between the nodes in the atlas. Especially for those nodes with a higher centrality, ensure that their causal relationships are highlighted, while the relationships of the marginal nodes can be appropriately weakened. The fusion logic adjustment also includes adjusting the connection strength and direction between the nodes in the atlas, eliminating redundant information, and enhancing the structural consistency and accuracy of the lesion relationship atlas. This process can be completed through an atlas optimization algorithm, such as the graph filtering method in the graph convolutional network, which dynamically adjusts the associations between nodes according to their relative importance and multi-stage weights. Finally, the obtained lesion atlas fusion logic adjustment data provides precise structural support for subsequent knowledge reasoning and disease prediction.

[0131] Preferably, the present invention also provides a fusion system for a urological medical knowledge atlas, which is used to execute the fusion method of the urological medical knowledge atlas as described above. The fusion system for the urological medical knowledge atlas includes:

[0132] An image region segmentation module, which is used to obtain urological historical medical record data and CT images of the lesion site; perform image region segmentation on the CT images of the lesion site according to the urological historical medical record data to obtain CT segmentation images of the lesion site;

[0133] An association analysis module, which is used to identify the synchronous intersection of lesion trends in the CT segmentation images of the lesion site according to the urological historical medical record data to obtain synchronous intersection data of lesion trends; perform implicit association deduction on the synchronous intersection data of lesion trends to obtain causal implicit association data of lesion trends;

[0134] An atlas fusion logic adjustment module, which is used to perform knowledge graph representation learning on the causal implicit association data of lesion trends to obtain atlas association structure learning data; perform atlas fusion logic adjustment according to the atlas association structure learning data to obtain lesion atlas fusion logic adjustment data;

[0135] A logical strategy design module, which is used to design a logical strategy based on the data adjusted by the lesion atlas fusion logic, obtain a lesion atlas fusion logical strategy, and send the lesion atlas fusion logical strategy to the cloud platform to execute the fusion method of the urology medical knowledge atlas.

[0136] Preferably, the present invention also provides a computer-readable storage medium, on which a computer program is stored, and characterized in that when the computer program is executed, it implements the fusion method of the urology medical knowledge atlas described in any one of the above.

[0137] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, it is intended to cover all changes falling within the meaning and scope of the equivalent elements of the application document within the present invention.

[0138] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A fusion method for urology medical knowledge graph, characterized in that: The following steps are involved: Step S1: Acquire the historical medical records of the urology department and the CT image of the lesion; perform image region segmentation on the CT image of the lesion according to the historical medical records of the urology department to obtain a CT segmented image of the lesion; Step S2: Perform synchronous intersection recognition of lesion trends on the CT segmented images of the lesion site according to the historical medical records of the urology department to obtain synchronous intersection data of lesion trends; perform implicit association deduction on the synchronous intersection data of lesion trends to obtain causal implicit association data of lesion trends; wherein step S2 is specifically as follows: Step S21: performing structural morphological analysis on the CT segmented image of the lesion to obtain structural morphological data of the lesion; Step S22: Performing synchronous intersection recognition of lesion trend on the lesion site structure morphology data and the lesion site CT segmentation image according to the historical medical record data of the urology department to obtain synchronous intersection data of lesion trend; wherein step S22 is specifically as follows: Step S221: Perform spatial domain analysis on the structural morphological data of the lesion site to obtain lesion location distribution data; Step S222: performing a descriptive trend comparison on the lesion location distribution data according to the historical medical records of the urology department to obtain descriptive trend comparison data of the lesion distribution; Step S223: simulating the lesion boundary tissue density similarity difference on the lesion location distribution data and the lesion site CT segmentation image based on the lesion distribution descriptive trend comparison data to obtain the lesion boundary tissue density similarity difference; Step S224: performing boundary transition density difference analysis on the lesion position distribution data according to the lesion boundary tissue density similarity difference to obtain the lesion tissue boundary transition density difference; Step S225: performing lesion trend synchronization intersection identification on the lesion site structure morphology data according to the lesion boundary tissue density similarity difference and the lesion tissue boundary transition density difference to obtain lesion trend synchronization intersection data; Step S23: performing causal correlation analysis on the lesion trend synchronization intersection data to obtain lesion trend synchronization causal data; Step S24: perform implicit correlation deduction based on the synchronous causal data of the lesion trend to obtain causal implicit correlation data of the lesion trend; wherein, step S24 is specifically as follows: Step S241: performing structured classification processing on the lesion trend synchronization causal data to obtain lesion development stage division data; Step S242: deducing the stage triggering factors according to the lesion development stage division data to obtain lesion stage triggering factor deduction data; Step S243: performing layer-by-layer factor event simulation analysis on the lesion stage trigger factor deduction data to obtain layer-by-layer factor event analysis data; Step S244: perform implicit correlation deduction based on the layer-by-layer factor event analysis data and the lesion stage trigger factor deduction data to obtain lesion trend causal implicit correlation data; Step S3: Perform knowledge graph representation learning on the causal implicit association data of the lesion trend to obtain graph association structure learning data; perform graph fusion logic adjustment based on the graph association structure learning data to obtain lesion graph fusion logic adjustment data; Step S4: Design a logic strategy based on the lesion map fusion logic adjustment data to obtain the lesion map fusion logic strategy, and send the lesion map fusion logic strategy to the cloud platform to execute the fusion method of the urology medical knowledge map.

2. The fusion method of urology medical knowledge graph according to claim 1 is characterized in that: Step S1 includes the following steps: Step S11: Acquire the historical medical records of the urology department and the CT images of the lesion site; Step S12: performing data desensitization processing on the historical medical record data of the urology department to obtain desensitized data of the historical medical record data; Step S13: Performing logical description analysis on the desensitized data of historical medical records to obtain logical description data of the medical records; Step S14: performing image region segmentation on the CT image of the lesion site according to the logical description data of the medical history to obtain a CT segmented image of the lesion site.

3. The fusion method of urology medical knowledge graph according to claim 1 is characterized in that: Step S223 includes the following steps: Perform multi-scale spatial coordinate transformation on the lesion location distribution data to obtain multi-scale lesion area calibration data; According to the descriptive trend comparison data of lesion distribution, multi-point neighboring density distribution analysis is performed on the multi-scale lesion area calibration data and the CT segmentation image of the lesion part to obtain multi-point neighboring density distribution data; Perform layered density fitting on the multi-point adjacent density distribution data to obtain layered lesion boundary density data; Based on the layered lesion boundary density data, position density trend deviation analysis is performed to obtain position density trend deviation data; According to the layered lesion boundary density data and the position density trend deviation data, the lesion position distribution data and the CT segmentation image of the lesion site are simulated to obtain the lesion boundary tissue density similarity difference.

4. The fusion method of urology medical knowledge graph according to claim 1 is characterized in that: Step S3 includes the following steps: Step S31: constructing a graph relationship for the lesion trend causal implicit correlation data to obtain a lesion causal trend relationship graph; Step S32: Perform knowledge graph representation learning on the lesion causal trend relationship graph to obtain graph association structure learning data; Step S33: Performing graph fusion logic adjustment on the lesion causal trend relationship graph according to the graph association structure learning data to obtain lesion graph fusion logic adjustment data.

5. The fusion method of urology medical knowledge graph according to claim 4 is characterized in that: Step S33 includes the following steps: Step S331: performing a topological correlation density analysis on the lesion causal trend relationship graph according to the graph correlation structure learning data to obtain topological correlation density data; Step S332: Based on the topological association density data, the relative centrality of the nodes in the lesion causal trend relationship map is evaluated to obtain the relative centrality of the nodes in the map; Step S333: performing multi-stage weight adjustment on the nodes of the lesion causal trend relationship map according to the topological association density data and the relative centrality of the map nodes to obtain multi-stage weight adjustment data for the nodes; Step S334: Performing graph fusion logic adjustment on the lesion causal trend relationship graph according to the topological association density data, the graph node relative centrality and the node multi-stage weight adjustment data to obtain the lesion graph fusion logic adjustment data.

6. A fusion system of urology medical knowledge graph, characterized in that: The fusion method for the urology medical knowledge graph according to claim 1 is used to implement the fusion system of the urology medical knowledge graph, comprising: The image region segmentation module is used to obtain the historical medical records of the urology department and the CT image of the lesion site; the image region of the CT image of the lesion site is segmented according to the historical medical records of the urology department to obtain the CT segmentation image of the lesion site; The association analysis module is used to identify the synchronous intersection of lesion trends on the CT segmented images of the lesion parts according to the historical medical records of the urology department, and obtain the synchronous intersection data of lesion trends; perform implicit association deduction on the synchronous intersection data of lesion trends, and obtain the causal implicit association data of lesion trends; The graph fusion logic adjustment module is used to perform knowledge graph representation learning on the causal implicit association data of the lesion trend to obtain graph association structure learning data; perform graph fusion logic adjustment based on the graph association structure learning data to obtain lesion graph fusion logic adjustment data; The logic strategy design module is used to design the logic strategy based on the lesion map fusion logic adjustment data, obtain the lesion map fusion logic strategy, and send the lesion map fusion logic strategy to the cloud platform to execute the fusion method of the urology medical knowledge map.

7. A computer-readable storage medium, characterized in that: A computer program is stored thereon, characterized in that when the computer program is executed, the method for fusing the urology medical knowledge graph as described in any one of claims 1-5 is implemented.

Images