Cerebral hemorrhage intelligent auxiliary decision-making system based on knowledge graph and multi-modal fusion

Abstract

The invention discloses a cerebral hemorrhage intelligent aid decision-making system based on knowledge graph and multi-modal fusion, and particularly relates to the technical field of medical big data, comprising a hematoma texture feature extraction module, a graph edge vector quantization module, an indication fluctuation sequence calculation module, a graph decision joint calibration module and a diagnosis and treatment risk quantitative evaluation module. According to the method, real-time difference fluctuation of intracranial pressure and systolic pressure is utilized, the depth of a decision tree is dynamically expanded, the map path weight is corrected, a self-adaptive reasoning mechanism based on the physical sign time-varying characteristic is constructed, the Euclidean distance between the decision tree and map nodes is calculated, radius judgment is executed, the spatial consistency of logic deduction and knowledge retrieval is locked, and the reliability of the decision tree is improved. And ensuring that the risk assessment model is synchronously calibrated along with the physiological state of the patient and the texture gradient.

Description

Technical Field

[0001] This invention relates to the field of medical big data technology, and in particular to an intelligent auxiliary decision-making system for cerebral hemorrhage based on knowledge graph and multimodal fusion. Background Technology

[0002] Medical big data technology covers the collection, governance, storage and analysis of multi-source medical data. Its core is to use distributed architecture, NLP and deep learning to achieve standardized integration and correlation analysis of data, so as to provide a basis for clinical research and public health decision-making.

[0003] Among them, the intelligent auxiliary decision-making system for cerebral hemorrhage based on knowledge graph and multimodal fusion refers to the multimodal data joint analysis and logical reasoning mechanism for emergency and treatment scenarios of cerebral hemorrhage. For emergency and treatment scenarios, it constructs a domain knowledge ontology through algorithms such as named entity recognition, extracts image features by convolutional neural networks, integrates electronic medical record text features and maps them to the knowledge graph, and outputs diagnostic results or treatment suggestions through graph neural networks and other reasoning.

[0004] Existing technologies rely on static ontology libraries, treating images and physical signs as isolated information and ignoring the real-time driving force of physiological parameter fluctuations in the acute phase. Feature alignment lacks a physical feedback mechanism, making it difficult to capture the coupling relationship between lesions and hemodynamics. Fixed reasoning paths cannot adapt to rapid changes in the condition, resulting in delayed decision-making and making it difficult to provide timely quantitative assessments for high-risk patients. Summary of the Invention

[0005] The main objective of this invention is to provide an intelligent auxiliary decision-making system for cerebral hemorrhage based on knowledge graph and multimodal fusion, which can effectively solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The intelligent auxiliary decision-making system for cerebral hemorrhage based on knowledge graph and multimodal fusion includes the following modules: Hematoma texture feature extraction module: Obtain the gray-level matrix of the hematoma region in CT images of cerebral hemorrhage, perform difference operations and square root operations on the gray levels of adjacent pixels, and generate a density gradient vector set; The graph edge vector quantization module calculates the mean of the density gradient vector group, extracts the semantic relationship edges between disease ontology entities and clinical phenotype entities, performs a multiplication operation between the edge modulus and the mean, and generates the initial physical quantization value of the graph edge. Indication fluctuation sequence calculation module: Collects intracranial pressure and systolic blood pressure monitoring data of patients, performs subtraction and absolute value operation on the current and previous values, and generates surgical indication fluctuation sequence values; The graph decision joint calibration module expands the decision tree depth based on the fluctuation sequence value of surgical indications, corrects the initial physical quantization value of the graph edge according to the fluctuation direction to update the knowledge reasoning path weight, calculates the Euclidean distance between the tree node and the graph topology node and performs radius determination, and establishes the decision tree-graph joint calibration path. The diagnostic and treatment risk quantification module extracts the bleeding volume and brain herniation parameters based on the decision tree graph and the joint calibration path, substitutes them into the formula to perform a weighted summation operation, and generates a diagnostic and treatment risk assessment coefficient for cerebral hemorrhage.

[0007] Preferably, the density gradient vector group includes pixel gradient magnitude, gradient direction vector, and local texture density. The graph edge vector quantization module calculates the mean of the density gradient vector group, extracts semantic relationship edges between disease ontology entities and clinical phenotype entities, multiplies the edge magnitude with the mean, and generates initial physical quantization values ​​for the graph edges. These initial physical quantization values ​​include edge connection weights, entity association strength, and semantic mapping scalars. The indication fluctuation sequence calculation module collects patient intracranial pressure and systolic blood pressure monitoring data, performs subtraction and absolute value operations on the current and previous values, and generates surgical indication fluctuation sequence values. These surgical indication fluctuation sequence values ​​include instantaneous pressure fluctuation values, temporal variation amplitude, and indication oscillation frequency. The graph decision joint calibration module expands the decision tree depth based on the surgical indication fluctuation sequence values, corrects the initial physical quantization values ​​of the graph edges according to the fluctuation direction to update the knowledge reasoning path weights, calculates the Euclidean distance between tree nodes and graph topology nodes, performs radius determination, and establishes a decision tree-graph joint calibration path.

[0008] Preferably, the decision tree graph joint calibration path includes decision logic branches, graph entity links, joint alignment coordinates, and a diagnosis and treatment risk quantification assessment module. Based on the decision tree graph joint calibration path, the module extracts the bleeding volume and brain herniation parameters, substitutes them into the formula to perform a weighted summation operation, and generates a brain hemorrhage diagnosis and treatment risk assessment coefficient. The brain hemorrhage diagnosis and treatment risk assessment coefficient includes the probability of bleeding progression, the brain herniation criticality score, and a comprehensive quantitative index.

[0009] Preferably, the hematoma texture feature extraction module includes a grayscale matrix construction submodule, a multidimensional difference calculation submodule, and a gradient magnitude synthesis submodule. Gray-scale matrix construction submodule: acquire head CT scan images of patients with cerebral hemorrhage, scan the CT values ​​of pixels in the images, filter out the set of pixels whose CT values ​​are within the preset hematoma range, read the spatial coordinates and gray-scale intensity values ​​of each pixel in the set, map the gray-scale intensity values ​​to a two-dimensional grid according to the spatial coordinates, and establish a hematoma pixel gray-scale matrix. Multidimensional difference calculation submodule: Calls the hematoma pixel grayscale matrix, traverses each row of the matrix, performs a subtraction operation between the current pixel grayscale value and the grayscale value of the pixel to the right to obtain the horizontal difference value, traverses each column of the matrix, performs a subtraction operation between the current pixel grayscale value and the grayscale value of the pixel below to obtain the vertical difference value, integrates the horizontal difference value and the vertical difference value to generate a bidirectional pixel difference sequence; The gradient magnitude synthesis submodule: For each coordinate point in the bidirectional pixel difference sequence, the horizontal difference value and the vertical difference value are respectively exponentialized. The two exponential values ​​are summed by addition. The arithmetic square root operation is performed on the summed value to generate a density gradient vector group.

[0010] Preferably, the graph edge vector quantization module includes a gradient mean calculation submodule, a semantic edge extraction submodule, and a physical quantization synthesis submodule: The gradient mean calculation submodule is based on the density gradient vector group. It iterates through the magnitude value of each gradient vector in the vector group, performs a summation operation on all magnitude values ​​to obtain the total magnitude, counts the total number of gradient vectors in the vector group, divides the total magnitude by the total number, and generates the gradient strength average coefficient. Semantic edge extraction submodule: Calls a pre-built knowledge graph database, locates disease ontology entity nodes and clinical phenotype entity nodes, identifies the semantic relationship edge connecting the two, obtains the original length value and spatial direction information of the semantic relationship edge in the pre-built graph space, and generates basic semantic connection vectors. The physical quantization synthesis submodule calls the basic semantic connection vector and the average gradient intensity coefficient, extracts the original length value of the basic semantic connection vector, performs a multiplication operation on the original length value and the average gradient intensity coefficient, assigns the result as the new modulus to the original vector direction, and generates the initial physical quantization value of the map edge.

[0011] Preferably, the indicator fluctuation sequence calculation module includes an indicator data acquisition submodule, a time series difference calculation submodule, and a fluctuation absolute value synthesis submodule: Indication data acquisition submodule: Collects intracranial pressure monitoring data and systolic blood pressure monitoring data of patients at continuous time points, arranges intracranial pressure values ​​in the order of timestamps to form an intracranial pressure time sequence queue, arranges systolic blood pressure values ​​in the order of timestamps to form a systolic blood pressure time sequence queue, and pairs and binds the two time sequence queues as indication pairs at the same time, generating a dual-channel indication time sequence queue; The time-series difference calculation submodule calls the dual-channel indicator time-series queue, traverses each adjacent time point, performs a subtraction operation between the intracranial pressure value at the current time and the intracranial pressure value at the previous time to obtain the intracranial pressure difference, performs a subtraction operation between the systolic pressure value at the current time and the systolic pressure value at the previous time to obtain the systolic pressure difference, merges the two differences into a difference pair at the same time, and generates an indicator instantaneous difference sequence. The absolute value of fluctuation synthesis submodule: For each difference pair in the instantaneous difference sequence of the indication, the absolute value of the intracranial pressure difference and the absolute value of the systolic pressure difference are summed. The summation result is used as the comprehensive fluctuation intensity at that moment. The comprehensive fluctuation intensities of all moments are integrated in chronological order to generate the fluctuation sequence value of the surgical indication.

[0012] Preferably, the map decision joint calibration module includes a depth expansion generation submodule, a weight dynamic adjustment submodule, and a path joint calibration submodule; The depth expansion generation submodule obtains the surgical indication fluctuation sequence value, compares it with the preset stable fluctuation threshold, and if the surgical indication fluctuation sequence value is greater than the stable fluctuation threshold, calculates the ratio of the surgical indication fluctuation sequence value to the stable fluctuation threshold, multiplies the current decision tree preset depth value by the ratio, generates the expansion depth value and generates child nodes, and generates an expansion structure child node set. The weight dynamic adjustment submodule: For the extended structure sub-node set, it calls the initial physical quantization value of the graph edge, detects the positive or negative attribute of the original difference corresponding to the surgical indication fluctuation sequence value. If the original difference is positive, it adds the product of the surgical indication fluctuation sequence value and the weight to the initial physical quantization value of the graph edge. If it is negative, it performs subtraction to obtain the adjusted knowledge reasoning path weight. The joint calibration path module calculates the Euclidean distance between the logical coordinates of the decision tree leaf nodes and the coordinates of the graph topology nodes based on the adjusted knowledge reasoning path weights. It then compares the Euclidean distance with the fault tolerance radius. If the Euclidean distance is less than the fault tolerance radius, it locks the current matching path and establishes a joint calibration path for the decision tree graph.

[0013] Preferably, the diagnostic and treatment risk quantification assessment module includes an end parameter extraction submodule and a weighted summation calculation submodule; End-point parameter extraction submodule: Based on the joint calibration path of the decision tree graph, index along the path topology connection relationship to the end position of the path, locate and lock the logical unit of the end node of the path, obtain the bleeding rate growth parameter stored in the node, obtain the brain herniation probability parameter stored in the node, perform numerical extraction and pairing combination on the bleeding rate growth parameter and the brain herniation probability parameter, and establish the end node risk parameter set; The weighted summation calculation submodule calls the risk parameter set of the terminal node, separates the bleeding volume growth rate parameter and the brain herniation probability parameter in the parameter set, substitutes the growth rate parameter into the growth term of the risk quantification formula, substitutes the probability parameter into the probability term of the risk quantification formula, multiplies the growth rate parameter by the first weight coefficient, multiplies the probability parameter by the second weight coefficient, calculates the algebraic sum of the two products, and generates the brain hemorrhage diagnosis and treatment risk assessment coefficient.

[0014] Compared with the prior art, the present invention has the following beneficial effects: This invention maps the heterogeneity of lesion texture to a physical quantitative attribute of the graph topology by obtaining a density gradient vector group generated from the difference of the hematoma grayscale matrix and multiplying it with the semantic relation edge modulus. Utilizing the real-time fluctuation of the difference between intracranial pressure and systolic blood pressure, the depth of the decision tree is dynamically expanded and the graph path weights are corrected to construct an adaptive reasoning mechanism based on the time-varying characteristics of vital signs. The Euclidean distance between the decision tree and the graph nodes is calculated and a radius determination is performed to ensure spatial consistency between logical deduction and knowledge retrieval, ensuring that the risk assessment model is synchronously calibrated with the patient's physiological state and texture gradient. This achieves a dynamic quantitative representation of the evolution of the acute phase of the disease, improving the responsiveness of auxiliary decision-making to subtle pathological changes in emergency scenarios. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the overall process structure of the present invention. Detailed Implementation

[0016] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0017] like Figure 1 As shown, the intelligent auxiliary decision-making system for cerebral hemorrhage based on knowledge graph and multimodal fusion includes the following modules: Hematoma texture feature extraction module: Obtain the gray-level matrix of the hematoma region in CT images of cerebral hemorrhage, perform difference operations and square root operations on the gray levels of adjacent pixels, and generate a density gradient vector set; The graph edge vector quantization module calculates the mean of the density gradient vector group, extracts the semantic relationship edges between disease ontology entities and clinical phenotype entities, performs a multiplication operation between the edge modulus and the mean, and generates the initial physical quantization value of the graph edge. Indication fluctuation sequence calculation module: Collects intracranial pressure and systolic blood pressure monitoring data of patients, performs subtraction and absolute value operation on the current and previous values, and generates surgical indication fluctuation sequence values; The graph decision joint calibration module compares the surgical indication fluctuation sequence value with a preset stable fluctuation threshold. If the surgical indication fluctuation sequence value is greater than the stable fluctuation threshold, the ratio of the surgical indication fluctuation sequence value to the stable fluctuation threshold is calculated. The current preset depth value of the decision tree is multiplied by the ratio to generate an extended depth value and a child node. At the same time, the initial physical quantization value of the graph edge is called. If the original difference corresponding to the surgical indication fluctuation sequence value is positive, the initial physical quantization value of the graph edge is added to the product of the surgical indication fluctuation sequence value and the weight. If it is negative, a subtraction is performed to obtain the adjusted knowledge reasoning path weight. The Euclidean distance between the logical coordinates of the decision tree leaf node and the coordinates of the graph topology node is calculated. The Euclidean distance is compared with the fault tolerance radius. If the Euclidean distance is less than the fault tolerance radius, the current matching path is locked and a decision tree graph joint calibration path is established. The diagnostic and treatment risk quantification module extracts the bleeding volume and brain herniation parameters based on the decision tree graph and the joint calibration path, substitutes them into the formula to perform a weighted summation operation, and generates a diagnostic and treatment risk assessment coefficient for cerebral hemorrhage.

[0018] Specifically, the hematoma texture feature extraction module includes a grayscale matrix construction submodule, a multidimensional difference calculation submodule, and a gradient magnitude synthesis submodule: Gray-scale matrix construction submodule: acquire head CT scan images of patients with cerebral hemorrhage, scan the CT values ​​of pixels in the images, filter out the set of pixels whose CT values ​​are within the preset hematoma range, read the spatial coordinates and gray-scale intensity values ​​of each pixel in the set, map the gray-scale intensity values ​​to a two-dimensional grid according to the spatial coordinates, and establish a hematoma pixel gray-scale matrix. Multidimensional difference calculation submodule: Calls the hematoma pixel grayscale matrix, traverses each row of the matrix, performs a subtraction operation between the current pixel grayscale value and the grayscale value of the pixel to the right to obtain the horizontal difference value, traverses each column of the matrix, performs a subtraction operation between the current pixel grayscale value and the grayscale value of the pixel below to obtain the vertical difference value, integrates the horizontal difference value and the vertical difference value to generate a bidirectional pixel difference sequence; The gradient magnitude synthesis submodule: For each coordinate point in the bidirectional pixel difference sequence, the horizontal difference value and the vertical difference value are respectively exponentialized. The two exponential values ​​are summed by addition. The arithmetic square root operation is performed on the summed value to generate a density gradient vector group.

[0019] Specifically, the graph edge vector quantization module includes a gradient mean calculation submodule, a semantic edge extraction submodule, and a physical quantization synthesis submodule: The gradient mean calculation submodule is based on the density gradient vector group. It iterates through the magnitude value of each gradient vector in the vector group, performs a summation operation on all magnitude values ​​to obtain the total magnitude, counts the total number of gradient vectors in the vector group, divides the total magnitude by the total number, and generates the gradient strength average coefficient. Semantic edge extraction submodule: Calls a pre-built knowledge graph database, locates disease ontology entity nodes and clinical phenotype entity nodes, identifies the semantic relationship edge connecting the two, obtains the original length value and spatial direction information of the semantic relationship edge in the pre-built graph space, and generates basic semantic connection vectors. The physical quantization synthesis submodule calls the basic semantic connection vector and the average gradient intensity coefficient, extracts the original length value of the basic semantic connection vector, performs a multiplication operation on the original length value and the average gradient intensity coefficient, assigns the result as the new modulus to the original vector direction, and generates the initial physical quantization value of the map edge.

[0020] Furthermore, the indicator fluctuation sequence calculation module includes an indicator data acquisition submodule, a time series difference calculation submodule, and a fluctuation absolute value synthesis submodule: Indication data acquisition submodule: Collects intracranial pressure monitoring data and systolic blood pressure monitoring data of patients at continuous time points, arranges intracranial pressure values ​​in the order of timestamps to form an intracranial pressure time sequence queue, arranges systolic blood pressure values ​​in the order of timestamps to form a systolic blood pressure time sequence queue, and pairs and binds the two time sequence queues as indication pairs at the same time, generating a dual-channel indication time sequence queue; The time-series difference calculation submodule calls the dual-channel indicator time-series queue, traverses each adjacent time point, performs a subtraction operation between the intracranial pressure value at the current time and the intracranial pressure value at the previous time to obtain the intracranial pressure difference, performs a subtraction operation between the systolic pressure value at the current time and the systolic pressure value at the previous time to obtain the systolic pressure difference, merges the two differences into a difference pair at the same time, and generates an indicator instantaneous difference sequence. The absolute value of fluctuation synthesis submodule: For each difference pair in the instantaneous difference sequence of the indication, the absolute value of the intracranial pressure difference and the absolute value of the systolic pressure difference are summed. The summation result is used as the comprehensive fluctuation intensity at that moment. The comprehensive fluctuation intensities of all moments are integrated in chronological order to generate the fluctuation sequence value of the surgical indication.

[0021] Furthermore, the graph decision joint calibration module includes a depth expansion generation submodule, a weight dynamic adjustment submodule, and a path joint calibration submodule; The depth expansion generation submodule obtains the surgical indication fluctuation sequence value, compares it with the preset stable fluctuation threshold, and if the surgical indication fluctuation sequence value is greater than the stable fluctuation threshold, calculates the ratio of the surgical indication fluctuation sequence value to the stable fluctuation threshold, multiplies the current decision tree preset depth value by the ratio, generates the expansion depth value and generates child nodes, and generates an expansion structure child node set. The weight dynamic adjustment submodule: For the extended structure sub-node set, it calls the initial physical quantization value of the graph edge, detects the positive or negative attribute of the original difference corresponding to the surgical indication fluctuation sequence value. If the original difference is positive, it adds the product of the surgical indication fluctuation sequence value and the weight to the initial physical quantization value of the graph edge. If it is negative, it performs subtraction to obtain the adjusted knowledge reasoning path weight. The joint calibration path module calculates the Euclidean distance between the logical coordinates of the decision tree leaf nodes and the coordinates of the graph topology nodes based on the adjusted knowledge reasoning path weights. It then compares the Euclidean distance with the fault tolerance radius. If the Euclidean distance is less than the fault tolerance radius, it locks the current matching path and establishes a joint calibration path for the decision tree graph.

[0022] Furthermore, the diagnostic and treatment risk quantification assessment module includes a terminal parameter extraction submodule and a weighted summation calculation submodule; End-point parameter extraction submodule: Based on the joint calibration path of the decision tree graph, index along the path topology connection relationship to the end position of the path, locate and lock the logical unit of the end node of the path, obtain the bleeding rate growth parameter stored in the node, obtain the brain herniation probability parameter stored in the node, perform numerical extraction and pairing combination on the bleeding rate growth parameter and the brain herniation probability parameter, and establish the end node risk parameter set; The weighted summation calculation submodule calls the risk parameter set of the terminal node, separates the bleeding volume growth rate parameter and the brain herniation probability parameter in the parameter set, substitutes the growth rate parameter into the growth term of the risk quantification formula, substitutes the probability parameter into the probability term of the risk quantification formula, multiplies the growth rate parameter by the first weight coefficient, multiplies the probability parameter by the second weight coefficient, calculates the algebraic sum of the two products, and generates the brain hemorrhage diagnosis and treatment risk assessment coefficient.

[0023] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent auxiliary decision-making system for cerebral hemorrhage based on knowledge graph and multi-modal fusion, characterized in that, The method comprises the following modules: Hematoma texture feature extraction module: obtain the gray matrix of the hematoma area of the CT image of cerebral hemorrhage, perform difference operation and square root processing on the gray scale of adjacent pixels, and generate a density gradient vector group; Atlas edge vector quantization module: calculate the mean value of the density gradient vector group, extract the semantic relationship edges between disease ontology entities and clinical phenotype entities, perform multiplication operation on the edge module length and the mean value, and generate atlas edge initial physical quantization value; Indication fluctuation sequence calculation module: collect intracranial pressure and systolic pressure monitoring data of the patient, perform subtraction and absolute value operation on the current and previous time values, and generate surgery indication fluctuation sequence value; Atlas decision joint calibration module: expand the decision tree depth based on the surgery indication fluctuation sequence value, correct the atlas edge initial physical quantization value according to the fluctuation direction to update the knowledge reasoning path weight, calculate the Euclidean distance between the tree node and the atlas topology node and perform radius determination, and establish the decision tree atlas joint calibration path; Diagnosis and treatment risk quantitative evaluation module: extract the amount of bleeding and brain hernia parameters according to the decision tree atlas joint calibration path, substitute them into the formula to perform weighted summation operation, and generate cerebral hemorrhage diagnosis and treatment risk evaluation coefficient.

2. The intelligent auxiliary decision system for cerebral hemorrhage based on knowledge graph and multi-modal fusion according to claim 1, characterized in that: The density gradient vector group includes pixel gradient amplitude, gradient direction vector, and local texture density. The atlas edge vector quantization module calculates the mean value of the density gradient vector group, extracts the semantic relationship edges between disease ontology entities and clinical phenotype entities, performs multiplication operation on the edge module length and the mean value, and generates atlas edge initial physical quantization value. The atlas edge initial physical quantization value includes edge connection weight, entity correlation strength, and semantic mapping scalar. The indication fluctuation sequence calculation module collects intracranial pressure and systolic pressure monitoring data of the patient, performs subtraction and absolute value operation on the current and previous time values, and generates surgery indication fluctuation sequence value. The surgery indication fluctuation sequence value includes pressure instantaneous fluctuation value, time series variation amplitude, and indication oscillation frequency. The atlas decision joint calibration module expands the decision tree depth based on the surgery indication fluctuation sequence value, corrects the atlas edge initial physical quantization value according to the fluctuation direction to update the knowledge reasoning path weight, calculates the Euclidean distance between the tree node and the atlas topology node and performs radius determination, and establishes the decision tree atlas joint calibration path.

3. The brain hemorrhage intelligent auxiliary decision system based on knowledge graph and multi-modal fusion according to claim 1, characterized in that: The decision tree atlas joint calibration path includes decision logic branch, atlas entity link, and joint alignment coordinates. The diagnosis and treatment risk quantitative evaluation module extracts the amount of bleeding and brain hernia parameters according to the decision tree atlas joint calibration path, substitutes them into the formula to perform weighted summation operation, and generates cerebral hemorrhage diagnosis and treatment risk evaluation coefficient. The cerebral hemorrhage diagnosis and treatment risk evaluation coefficient includes bleeding progression probability, brain hernia critical score, and comprehensive quantitative index.

4. The brain hemorrhage intelligent auxiliary decision system based on knowledge graph and multi-modal fusion according to claim 1, characterized in that: The hematoma texture feature extraction module includes gray matrix construction submodule, multi-dimensional difference calculation submodule, and gradient module length synthesis submodule: Gray matrix construction submodule: obtain the CT scan image of the head of the patient with cerebral hemorrhage, the CT value of the pixel points in the scan image, select the pixel point set whose CT value is within the preset hematoma range, read the spatial position coordinates and gray intensity values of each pixel point in the set, map the gray intensity values to a two-dimensional grid according to the spatial position coordinates, and establish the hematoma pixel gray matrix; Multi-dimensional difference calculation submodule: call hematoma pixel gray matrix, traverse each row of the matrix, perform subtraction operation on the current pixel gray value and the right adjacent pixel gray value to obtain horizontal difference value, traverse each column of the matrix, perform subtraction operation on the current pixel gray value and the lower adjacent pixel gray value to obtain vertical difference value, integrate the horizontal difference value and the vertical difference value, and generate a bidirectional pixel difference sequence; Gradient modulus synthesis submodule: for the horizontal difference value and the vertical difference value corresponding to each coordinate point in the bidirectional pixel difference sequence, respectively, perform power operation on the horizontal difference value and the vertical difference value, perform addition summation on the two power values obtained after operation, and perform arithmetic square root operation on the numerical result obtained by summation, to generate a density gradient vector group.

5. The brain hemorrhage intelligent auxiliary decision system based on knowledge graph and multi-modal fusion according to claim 4, characterized in that: The atlas edge vector quantization module includes a gradient mean calculation submodule, a semantic edge extraction submodule, and a physical quantization synthesis submodule: Gradient mean calculation submodule: based on the density gradient vector group, traverse the modulus numerical value of each gradient vector in the vector group, perform cumulative summation operation on all modulus numerical values to obtain modulus sum, count the total number of gradient vectors in the vector group, divide the modulus sum by the total number, and generate a gradient intensity average coefficient; Semantic edge extraction submodule: call the preset knowledge graph database, locate the disease ontology entity node and the clinical phenotype entity node therein, identify the semantic relationship edge connecting the two, obtain the original length value and spatial direction information of the semantic relationship edge in the preset graph space, and generate a basic semantic connection vector; Physical quantization synthesis submodule: call the basic semantic connection vector and the gradient intensity average coefficient, extract the original length value of the basic semantic connection vector, perform multiplication operation on the original length value and the gradient intensity average coefficient, take the operation result as a new modulus, and assign it to the original vector direction to generate an atlas edge initial physical quantization value.

6. The brain hemorrhage intelligent auxiliary decision system based on knowledge graph and multi-modal fusion according to claim 5, characterized in that: The indicator fluctuation sequence calculation module includes an indicator data acquisition submodule, a time sequence difference value operation submodule, and a fluctuation absolute value synthesis submodule: Indicator data acquisition submodule: collect intracranial pressure monitoring data and systolic pressure monitoring data at consecutive time points of a patient, arrange intracranial pressure values in time stamp order to form an intracranial pressure time sequence queue, arrange systolic pressure values in time stamp order to form a systolic pressure time sequence queue, pair and bind the two time sequence queues as an indicator pair at the same time to generate a double-channel indicator time sequence queue; Time sequence difference value operation submodule: call the double-channel indicator time sequence queue, traverse each adjacent time point, perform subtraction operation on the current time intracranial pressure value and the previous time intracranial pressure value to obtain an intracranial pressure difference, perform subtraction operation on the current time systolic pressure value and the previous time systolic pressure value to obtain a systolic pressure difference, combine the two differences into a difference pair at the same time to generate an indicator instantaneous difference sequence; Fluctuation absolute value synthesis submodule: for each difference pair in the indicator instantaneous difference sequence, take the absolute value of the intracranial pressure difference and the absolute value of the systolic pressure difference to perform addition summation, take the summation result as the comprehensive fluctuation intensity at that time, and integrate the comprehensive fluctuation intensities at all times in time order to generate a surgical indicator fluctuation sequence value.

7. The brain hemorrhage intelligent auxiliary decision system based on knowledge graph and multi-modal fusion according to claim 6, characterized in that: The atlas decision joint calibration module comprises a depth expansion generation submodule, a weight dynamic adjustment submodule, and a path joint calibration submodule. The depth expansion generation submodule: obtains a surgical indication fluctuation sequence value, compares it with a preset stable fluctuation threshold, and if the surgical indication fluctuation sequence value is greater than the stable fluctuation threshold, calculates the ratio of the surgical indication fluctuation sequence value to the stable fluctuation threshold, multiplies the preset depth value of the current decision tree by the ratio, generates an expanded depth value, and generates a subnode, and generates an expanded structure subnode set. The weight dynamic adjustment submodule: for the expanded structure subnode set, calls the atlas edge initial physical quantitative value, detects the original difference value positive and negative attribute corresponding to the surgical indication fluctuation sequence value, if the original difference value is positive, adds the product of the surgical indication fluctuation sequence value and the weight to the atlas edge initial physical quantitative value, if it is negative, performs subtraction, and obtains the adjusted knowledge reasoning path weight. The path joint calibration submodule: based on the adjusted knowledge reasoning path weight, calculates the Euclidean distance between the decision tree leaf node logic coordinate and the atlas topology node coordinate, compares the Euclidean distance with the fault tolerance radius, if the Euclidean distance is less than the fault tolerance radius, locks the current matching path, and establishes the decision tree atlas joint calibration path.

8. The brain hemorrhage intelligent auxiliary decision system based on knowledge graph and multi-modal fusion according to claim 7, characterized in that: The diagnosis and treatment risk quantitative evaluation module comprises a terminal parameter extraction submodule and a weighted summation calculation submodule. The terminal parameter extraction submodule: based on the decision tree atlas joint calibration path, indexes to the path terminal position along the path topology connection relationship, locates and locks the path terminal node logic unit, obtains the bleeding volume growth rate parameter stored in the node, obtains the brain hernia occurrence probability parameter stored in the node, performs numerical extraction and paired combination on the bleeding volume growth rate parameter and the brain hernia occurrence probability parameter, and establishes a terminal node risk parameter set. The weighted summation calculation submodule: calls the terminal node risk parameter set, separates the bleeding volume growth rate parameter and the brain hernia occurrence probability parameter in the parameter set, substitutes the growth rate parameter into the risk quantification formula growth term, substitutes the probability parameter into the risk quantification formula probability term, multiplies the growth rate parameter by the first weight coefficient, multiplies the probability parameter by the second weight coefficient, calculates the algebraic sum of the two product results, and generates a cerebral hemorrhage diagnosis and treatment risk evaluation coefficient.

Images