CN-121983221-A - Medical image feature conduction identification method and system

Abstract

A medical image feature conduction identification method and a system relate to the technical field of medical image identification. The method comprises the steps of extracting a built-in defect knowledge graph of a large model, analyzing the built-in defect knowledge graph to generate an image feature conduction reference, extracting target medical image data, establishing a hierarchical corresponding relation with the image feature conduction reference, outputting an adaptive feature set, calling a bidirectional knowledge conduction link to perform association operation to generate enhanced association features, further excavating potential relation among the features, enhancing expressive force and distinction degree of the features, dividing key regions based on the enhanced association features, constructing a global covered defect feature conduction link, accurately identifying defect features of different regions and conduction relation thereof, and finally integrating according to medical image report specifications to form a structured image identification report, so that accuracy, consistency and efficiency of image identification are improved.

Inventors

• DONG RUI
• Song Zai
• ZHAN YONG
• TAO QILIN
• LI YI
• LI MENGTING
• LU YIFEI
• ZHOU SIYI

Assignees

• 复旦大学附属儿科医院

Dates

Publication Date

20260505

Application Date

20260408

Claims (10)

1. 1. A medical image feature conduction identification method, comprising: retrieving a built-in defect knowledge graph of the large model, analyzing the associated levels and the conduction logic of defect feature nodes and anatomical structure nodes in the defect knowledge graph, and generating an image feature conduction reference; Extracting anatomical structure distribution data and a pixel feature matrix of a target medical image, executing feature conduction adaptation operation based on an anatomical structure mapping rule, establishing a hierarchical correspondence between target medical image features and image feature conduction references, and outputting an adaptation feature set; invoking a bidirectional knowledge conduction link of the large model, connecting each feature node in the adaptive feature set to the starting end of the bidirectional knowledge conduction link, and performing association operation on the starting end, the upstream defect feature, the downstream defect feature and the parallel defect feature in the implementation defect knowledge graph to generate an enhanced association feature; Dividing a key region of a target medical image based on anatomic positioning information of the reinforced association features, constructing feature conduction sub-links in the key region, and connecting all feature conduction sub-links in series by fusing feature nodes across regions to form a defect feature conduction link covered in a whole domain; and extracting node association data, conduction path information and defect matching results in the defect feature conduction link, and integrating according to medical image report specifications to form a structural image identification report, wherein the structural image identification report comprises a feature conduction map, defect association description and anatomical positioning labels.
2. 2. The method for identifying medical image feature conduction according to claim 1, wherein the retrieving a defect knowledge graph built in a large model, resolving association levels and conduction logics of defect feature nodes and anatomical structure nodes in the defect knowledge graph, and generating an image feature conduction reference comprises: A defect knowledge graph built in the large model is called, the defect knowledge graph stores anatomical structure features, image representation features and defect association features corresponding to birth defects of various children, all the features exist in an independent node form, and the nodes are connected through marked association relations; Extracting core image feature nodes corresponding to all defect types in the defect knowledge graph, wherein each core image feature node comprises a plurality of information dimensions, the plurality of information dimensions comprise feature dimensions, expression forms, anatomical positioning and pathology association basis, and each information dimension corresponds to a specific clinical case data support; Analyzing the conduction logic among the core image feature nodes, marking the sequence of occurrence or the association of occurrence of the different core image feature nodes in defect evolution, wherein the conduction logic is formed based on the pathological mechanism of defect occurrence and the time sequence of development progress; Based on the occurrence frequency of the conduction logic recorded in the clinical case database, the conduction logic of each core image feature node is prioritized, and the ordering result is determined by combining the support conclusion in the defect pathological mechanism research literature; Constructing a conduction path template of the core image feature node based on the ordered conduction logic, wherein each conduction path template corresponds to the feature conduction rule of a type of defect, and the conduction path template comprises a starting node identifier, an intermediate node sequence, a terminating node identifier and conduction direction parameters among nodes; extracting anatomical structure nodes directly associated with the core image feature nodes in the defect knowledge graph, establishing a unique mapping relation between the core image feature nodes and the anatomical structure nodes, and marking specific anatomical position coordinates and anatomical range boundaries corresponding to each core image feature node; embedding three-dimensional space position information of anatomical structure nodes into conduction path templates, adding space dimension attributes for each conduction path template, and marking adjacent relation, overlapping range and distance parameters of anatomical positions corresponding to the core image feature nodes in a three-dimensional space by the space dimension attributes; carrying out standardization processing on the conduction path templates embedded with the spatial attributes, integrating all the standardized conduction path templates, and constructing an initial image feature conduction reference frame, wherein the initial image feature conduction reference frame comprises feature conduction paths corresponding to various defects, spatial association data and pathological mechanism description; The conducting paths in the initial image feature conducting reference frame are prioritized and screened, and the core conducting paths and the reference conducting paths are distinguished based on the data completeness and pathological mechanism definition of clinical cases related to the conducting paths, so that the image feature conducting reference comprising the hierarchical paths is generated.
3. 3. The method for identifying feature conduction of medical image according to claim 1, wherein the extracting the anatomical structure distribution data and the pixel feature matrix of the target medical image, performing feature conduction adaptation operation based on the anatomical structure mapping rule, establishing a hierarchical correspondence between the feature of the target medical image and the image feature conduction reference, and outputting an adaptation feature set, includes: Extracting overall anatomical structure distribution data of a target medical image through an image segmentation algorithm, wherein the overall anatomical structure distribution data covers all identifiable anatomical structure names, three-dimensional space positions, morphological parameters and spatial association relations among the identifiable anatomical structure names, the three-dimensional space positions and the morphological parameters in the image, and is obtained by scanning the image area by area and matching an anatomical structure feature library; Generating a pixel characteristic matrix of the target medical image by adopting a pixel matrix analysis technology, and decomposing the pixel characteristic matrix into a plurality of region characteristic submatrices according to an anatomical region division rule, wherein the pixel characteristic matrix comprises characteristic data of each pixel point, and the characteristic data comprises a gray value, a texture parameter and an edge gradient value; Carrying out level comparison on the overall anatomical structure distribution data and anatomical structure nodes in the image feature conduction reference, marking specific anatomical regions which are completely matched with the reference anatomical structure nodes in the target medical image, and recording the three-dimensional coordinate range of each specific anatomical region in the image; extracting features of the regional feature submatrices corresponding to each specific anatomical region, wherein the extracted image features comprise gray level distribution ranges, gray level concentration intervals, texture trend parameters, edge morphology curves and internal structure density distribution of the specific anatomical region; performing dimension comparison on the extracted image features of each specific anatomical region and core image feature nodes associated with corresponding anatomical structure nodes in the image feature transmission reference, aligning feature dimensions one by one, and marking matching parameters and fit points of each feature dimension; screening image features matched with the dimensions of the reference core image feature nodes, and marking the image features as candidate adaptation features, wherein each candidate adaptation feature comprises a corresponding anatomical region name, a three-dimensional coordinate range, feature dimension data and matching parameter information of the reference node; labeling spatial and logical correlations between candidate adaptation features based on three-dimensional spatial relationships of anatomical regions corresponding to the candidate adaptation features in combination with conduction logic of core image feature nodes in an image feature conduction reference; According to the conduction path template in the image feature conduction reference, adjusting the arrangement sequence of the candidate adaptation features, so that the sequence of the candidate adaptation features is consistent with the node sequence in the corresponding conduction path template, and a feature sequence corresponding to the reference conduction path is formed; Supplementary detail data are supplemented for each candidate adaptation feature in the feature sequence, wherein the supplementary content comprises specific performance parameters of the feature, three-dimensional coordinates in the image and association parameters of adjacent candidate adaptation features, and the supplementary data are generated based on an original pixel matrix of the target medical image and feature description in an image feature conduction reference; And performing secondary level alignment on the supplemented feature sequence and a conduction path template in the image feature conduction reference, integrating all the matched feature sequences, and outputting an adaptive feature set.
4. 4. The method for identifying feature conduction of medical image according to claim 1, wherein the step of calling a bidirectional knowledge conduction link of a large model, and performing association operation between each feature node in the adaptive feature set and an upstream defect feature, a downstream defect feature and a parallel defect feature in the implementation defect knowledge graph to generate an enhanced association feature, comprises: invoking a bidirectional knowledge transfer link of the large model, wherein the bidirectional knowledge transfer link is a characteristic association transfer channel constructed based on a defect knowledge graph and comprises a longitudinal transfer path related to a pathological process and a transverse transfer path related to accompanying association; Extracting each feature node in the adaptive feature set, taking each feature node as an independent input unit to be connected into a corresponding initial position of the bidirectional knowledge transmission link, wherein each feature node comprises feature dimension data, anatomical positioning information and initial transmission relation parameters; Based on the characteristic type of the input node, the anatomical positioning information and the matching parameters of the bidirectional knowledge transmission link starting node, each input node is connected into a corresponding transmission path, so that the characteristic dimensions of the input node and the characteristic dimensions of the link starting node are completely aligned; Tracing an upstream defect characteristic node corresponding to an input node through a longitudinal conduction path of a bidirectional knowledge conduction link, wherein the upstream defect characteristic node is a characteristic node which appears before the input node in a pathological process, and the tracing basis is a pathological logic association parameter preset in the bidirectional knowledge conduction link; expanding parallel defect characteristic nodes corresponding to the input nodes through transverse conduction paths of the bidirectional knowledge conduction links, wherein the parallel defect characteristic nodes are characteristic nodes which are simultaneously appeared with the input nodes in clinical cases, and the expansion basis is preset accompanying association parameters in the bidirectional knowledge conduction links; extracting complete information of upstream defect feature nodes obtained by tracing and parallel defect feature nodes obtained by expanding, wherein the complete information comprises feature description, anatomical positioning coordinates, association strength parameters and conduction priority parameters, and the complete information is all from stored data of a defect knowledge graph; Performing fusion operation on upstream defect characteristic node information, parallel defect characteristic node information and corresponding input node information, reserving core characteristic data of the input nodes, and adding associated parameters of the upstream nodes and the parallel nodes to form fusion characteristic nodes; And comparing the conduction relation between the fusion feature nodes with preset conduction logic of the bidirectional knowledge conduction link, strengthening association parameters of the fusion feature nodes based on rules summarized by pathological mechanisms and clinical cases, arranging all the strengthened fusion feature nodes according to the conduction sequence and association relation parameters of the bidirectional knowledge conduction link, and integrating to form strengthening association features comprising input nodes, upstream defect feature nodes, downstream defect feature nodes, parallel defect feature nodes and all the association parameters.
5. 5. The method for identifying feature conduction of medical image according to claim 1, wherein the step of dividing the key region of the target medical image based on the anatomical positioning information of the enhanced association feature, constructing feature conduction sub-links in the key region, and forming a defect feature conduction link covered in a global manner by fusing feature nodes across regions in series with all feature conduction sub-links comprises: the method comprises the steps of extracting the enhanced association feature, dividing a key region of a target medical image based on anatomical positioning information of each feature node in the enhanced association feature, wherein the key region is a set of anatomical regions corresponding to the feature nodes and covers all anatomical parts associated with defect features; Dividing a key region of the target medical image based on the anatomical positioning information of each fusion feature node, wherein the key region is a set of anatomical regions corresponding to the fusion feature nodes and covers all anatomical parts associated with the existence of the defect feature; Calculating the coordinates of the anatomical region corresponding to each fusion feature node in the key region, and analyzing the spatial association parameters among the fusion feature nodes, wherein the spatial association parameters comprise the distance value, the overlapping range proportion and the adjacent relation identification of the anatomical position; determining the conduction sequence of the fusion feature nodes in the key region by combining the space association parameters and the conduction direction parameters in the reinforcement association features, wherein the conduction sequence simultaneously meets the requirements of continuity parameters in space and the requirements of pathological rationality in logic; Sequentially connecting the fusion feature nodes in the key areas according to the determined conduction sequence to construct initial defect feature conduction sub-links, wherein each initial defect feature conduction sub-link corresponds to defect feature conduction logic in one key area; Screening fusion feature nodes belonging to a plurality of key areas in the reinforced association features, marking the fusion feature nodes belonging to the plurality of key areas as cross-key area fusion feature nodes, and establishing feature conduction channels among different key areas by the cross-key area fusion feature nodes; The cross-critical area fusion feature nodes are used as connection hinges, initial defect feature conduction sub-links corresponding to different critical areas are connected in series to form a preliminary cross-regional conduction link, and the preliminary cross-regional conduction link comprises a plurality of initial defect feature conduction sub-links and connection nodes among the initial defect feature conduction sub-links; supplementing conduction relation data in the preliminary trans-regional conduction links, and adding corresponding support information in each conduction link, wherein the support information comprises a pathological basis literature mark and a clinical case number, and all support information is from a defect knowledge graph of the large model; The method comprises the steps of binding original pixel data of a target medical image with a preliminary trans-regional conduction link, enabling each fusion characteristic node in the preliminary trans-regional conduction link to correspond to a specific pixel region in the image, checking whether the fusion characteristic nodes in the reinforced association characteristic are all incorporated into the bound preliminary trans-regional conduction link, adding the non-incorporated fusion characteristic nodes to corresponding positions of the link according to anatomical positioning information and conduction logic of the non-incorporated fusion characteristic nodes, and forming a global covered defect characteristic conduction link covering all critical regions related to defects in the target medical image.
6. 6. The method for identifying conduction of medical image features according to claim 2, wherein analyzing the conduction logic between the core image feature nodes, labeling the sequence of occurrence or the association of the occurrence of the different core image feature nodes in defect evolution, comprises: the method comprises the steps of calling a clinical case database stored in a large model, wherein the clinical case database comprises diagnosed child birth defect image cases and corresponding diagnosis reports, and each child birth defect image case records image representation data and development progress time nodes of defects in detail; Screening cases related to the core image feature nodes from the clinical case database based on the matching parameters of the feature description of the core image feature nodes and the image representation data of the clinical cases, wherein each core image feature node corresponds to a plurality of clinical cases; Analyzing the appearance sequence of the core image feature nodes case by case, recording the time stamp or the image layer mark of the first appearance of different core image feature nodes in each clinical case, forming a node appearance sequence list, and counting the frequency values of the appearance sequence of the core image feature nodes in all relevant clinical cases according to each possible sequence combination as a statistical unit to form a sequence combination frequency table; the method comprises the steps of calling pathology mechanism research data in a built-in knowledge base of a large model, wherein the pathology mechanism research data comprise molecular biology mechanisms and anatomic evolution rules of defect occurrence and development; Verifying rationality of the occurrence sequence of the core image feature nodes by combining a sequence combination frequency table and pathology mechanism research data, and determining main conduction logic among the core image feature nodes, wherein the main conduction logic is node sequence combination which has high frequency occurrence records in the clinical case database and accords with the pathology mechanism; identifying secondary conduction logic among the core image feature nodes, wherein the secondary conduction logic is node sequence combination with frequency value ratio which does not reach clinical statistical standard but has pathological mechanism support or appears in specific labeling cases; labeling triggering condition parameters of each conduction logic, wherein the triggering condition parameters comprise factors influencing conduction path selection, and specifically comprise defect type codes, child age stage range and anatomical structure state parameters; The method comprises the steps of marking a primary conduction logic as a priority conduction path and marking a secondary conduction logic as an alternative conduction path, so that all marked conduction logics are arranged to form a conduction logic list of core image feature nodes, wherein the conduction logic list comprises node combinations, occurrence frequency values, pathology basis abstracts, triggering condition parameters and path type identifiers of the conduction logics.
7. 7. The method for recognizing the conduction of the features of the medical image according to claim 3, wherein the feature extraction is performed on the regional feature submatrix corresponding to each specific anatomical region, the extracted image features include a gray distribution range, a gray concentration interval, a texture trend parameter, an edge morphology curve, and an internal structure density distribution of the specific anatomical region, and the method comprises the following steps: identifying edge lines with abrupt gray value changes in the images through an edge detection algorithm, and determining boundary coordinates of each specific anatomical region based on the coordinates of the edge lines, wherein the boundary coordinates are used for delineating the range of the specific anatomical region; dividing the specific anatomical region into a plurality of non-overlapping subareas according to the detail distribution condition of the anatomical structure in the specific anatomical region, wherein each subarea comprises a relatively independent anatomical detail or characteristic unit; Counting the gray value data of all pixels in each sub-area, and determining the distribution range interval of gray values, the gray interval which appears in a concentrated way and the morphological characteristics of gray distribution as the gray distribution characteristics of the sub-areas; analyzing the change trend of the gray value of the pixel in the subarea by adopting a multidirectional scanning algorithm, and recording the main extension direction angle, the consistent degree parameter of the trend and the density change curve of the texture as the texture trend characteristics of the subarea; Extracting line data of gray value mutation in a subarea, and analyzing the continuous degree, bending radian parameter and thickness variation value of the line as the edge morphological characteristics of the subarea; Analyzing aggregation state parameters, density difference values and structural hierarchy distribution of pixels in a subarea as internal structural features of the subarea; Integrating gray distribution characteristics, texture trend characteristics, edge morphological characteristics and internal structure characteristics of each sub-area to form comprehensive characteristic description of each sub-area, wherein the comprehensive characteristic description of each sub-area comprises all image characteristic data of the sub-area; Based on the spatial position relation of all subareas in a specific anatomical region, arranging comprehensive feature descriptions of all subareas according to a spatial coordinate sequence to form a feature sequence of the specific anatomical region, wherein the feature sequence of the specific anatomical region reflects the association of the features of all subareas in space; And carrying out fusion operation on the feature sequences of the specific anatomical region based on the spatial association relation of each sub-region, and extracting core feature information after the fusion operation, wherein the core feature information is a feature set representing the main image representation of the specific anatomical region and comprises features of multiple dimensions, and the multiple dimensions comprise gray scale, texture, edge and internal structure.
8. 8. The method for identifying feature conduction of medical image according to claim 4, wherein tracing back upstream defect feature nodes corresponding to input nodes through longitudinal conduction paths of the bidirectional knowledge conduction link, wherein the upstream defect feature nodes are feature nodes appearing before the input nodes in a pathological process, and tracing back is based on pathological logic association parameters preset in the bidirectional knowledge conduction link, and the method comprises the following steps: determining a specific coordinate position of the input node in the bidirectional knowledge conduction link based on the feature description of the input node, the anatomical positioning information and the matching parameters of the input node in the bidirectional knowledge conduction link; calling a longitudinal conduction path rule corresponding to the coordinate position in the bidirectional knowledge conduction link, wherein the longitudinal conduction path rule is constructed based on the pathological development process of the defect and comprises a possible upstream node identifier and conduction condition parameters corresponding to each node; Based on the longitudinal conduction path rule, screening potential upstream nodes directly related to input nodes, wherein the potential upstream nodes are preamble characteristic nodes which directly cause the input nodes to appear in a pathological process, and each input node corresponds to a plurality of potential upstream nodes; extracting feature description, anatomical positioning coordinates and conduction condition parameters of each potential upstream node from the defect knowledge graph; comparing the conduction condition parameters of each potential upstream node with the actual condition parameters of the input nodes, wherein the actual condition parameters of the input nodes comprise corresponding anatomical region state parameters and feature expression values; Screening potential upstream nodes with the conduction condition parameters conforming to the actual condition parameters of the input nodes to eliminate potential upstream nodes with the conduction condition parameters not meeting the actual condition parameters of the input nodes, wherein the potential upstream nodes with the conduction condition parameters conforming to the actual condition parameters of the input nodes are actual upstream nodes of the input nodes in a pathological process; based on the conduction logic between the upstream node and the input node after pathological mechanism data analysis and screening, marking a causal relationship chain which is caused by the upstream node or causes the input node to appear; tracing the upstream nodes of the more preamble corresponding to the upstream nodes after screening, repeating the operations of potential node screening, conduction condition parameter comparison and logic verification until tracing to the initial node of the longitudinal conduction path, and recording all the upstream nodes in the whole process according to the tracing sequence to form an upstream node sequence, wherein the upstream node sequence comprises the information of each upstream node, the conduction logic of the adjacent nodes and the conduction condition parameters; And integrating the associated data of the upstream node sequence and the input nodes, marking the conduction relation type and pathological basis identification of the input nodes and each upstream node, and forming a longitudinal upstream node tracing result of the input nodes.
9. 9. The method for identifying feature conduction of medical images according to claim 5, wherein the step of connecting the initial defect feature conduction sub-links corresponding to different key regions in series with the cross-key region fusion feature node as a connection hub to form a preliminary cross-region conduction link comprises: Extracting all fusion feature nodes with a plurality of key region anatomical positioning information in the reinforced association features, marking the fusion feature nodes with the plurality of key region anatomical positioning information as cross-key region fusion feature nodes, and establishing feature conduction channels among different key regions through the cross-key region fusion feature nodes; labeling all key area names corresponding to each cross-key area fusion characteristic node, recording characteristic expression parameters, associated node identifications and conduction direction parameters of the cross-key area fusion characteristic nodes in each key area, and grasping complete information of the nodes in different areas; Invoking an initial defect feature conduction sub-link corresponding to each cross-key region fusion feature node, wherein each initial defect feature conduction sub-link corresponds to one key region, and the initial defect feature conduction sub-link comprises conduction logic and associated node data of the cross-key region fusion feature node in the corresponding region; Analyzing conduction roles of cross-key-area fusion feature nodes in different initial defect feature conduction sub-links, wherein the conduction roles are divided into an initial node, an intermediate node and a termination node, and the positioning of the conduction roles is determined based on the position of the node in the initial defect feature conduction sub-link and conduction direction parameters; Determining a mode of connecting different initial defect feature conduction sub-links by cross-key region fusion feature nodes based on the conduction roles, and if the cross-key region fusion feature nodes are all intermediate nodes in a plurality of initial defect feature conduction sub-links, adopting a direct series connection mode for connection; taking the cross-key region fusion characteristic nodes as connection hinges, and sequentially connecting all corresponding initial defect characteristic conduction sub-links in series according to a determined connection mode to form a preliminary cross-region conduction link; Analyzing the consistency of the conduction logic in the preliminary trans-regional conduction links based on the pathological mechanism and the clinical rule of the defects, detecting whether the association relation conflict exists in the conduction direction after the conduction sub-links with different initial defect characteristics are connected, and if so, adjusting the conduction direction parameters according to the pathological mechanism and the clinical rule; Performing parameter adjustment on the transmission links with logic conflict, referring to the defect knowledge graph, selecting transmission logic conforming to a pathological mechanism to replace the conflict links, and adding cross-region transmission associated description data, wherein the associated description data are used for describing pathological basis and clinical case support conditions of different initial defect characteristic transmission sub-links connected by cross-key region fusion characteristic nodes; and integrating all connected cross-region conductive links to form a cross-region defect characteristic conductive link frame covering all key regions.
10. 10. A medical image feature conduction identification system, comprising: A processor; a machine-readable storage medium storing machine-executable instructions for the processor; Wherein the processor is configured to perform the medical image feature conduction identification method of any one of claims 1-9 via execution of the machine executable instructions.

Description

Medical image feature conduction identification method and system Technical Field The invention relates to the technical field of medical image recognition, in particular to a medical image feature conduction recognition method and system. Background In the medical field, early and accurate identification of birth defects of children is important to guaranteeing health, formulation and timely effective intervention measures of children. Medical image identification is one of the important means for diagnosing birth defects of children, however, the conventional medical image identification method has a plurality of limitations. On the one hand, the existing image recognition mainly depends on professional experience and subjective judgment of doctors, and understanding and judgment of image features may be different among different doctors, so that consistency and accuracy of diagnosis results are difficult to guarantee. Especially, when facing the complicated and changeable birth defect image of children, doctors may misdiagnose or miss diagnosis due to the factors such as insufficient experience or fatigue. On the other hand, the conventional image recognition method lacks of system integration and deep mining of child birth defect knowledge. The birth defects of children relate to various types and complex anatomical structure changes, and the existing method is often only used for analyzing the characteristics in the images in isolation, but the characteristics are not combined with the whole knowledge system of the birth defects of the children, so that the types and the degrees of the defects are difficult to comprehensively and accurately identify. Disclosure of Invention In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a medical image feature conduction identification method, which includes: retrieving a built-in defect knowledge graph of the large model, analyzing the associated levels and the conduction logic of defect feature nodes and anatomical structure nodes in the defect knowledge graph, and generating an image feature conduction reference; Extracting anatomical structure distribution data and a pixel feature matrix of a target medical image, executing feature conduction adaptation operation based on an anatomical structure mapping rule, establishing a hierarchical correspondence between target medical image features and image feature conduction references, and outputting an adaptation feature set; invoking a bidirectional knowledge conduction link of the large model, connecting each feature node in the adaptive feature set to the starting end of the bidirectional knowledge conduction link, and performing association operation on the starting end, the upstream defect feature, the downstream defect feature and the parallel defect feature in the implementation defect knowledge graph to generate an enhanced association feature; Dividing a key region of a target medical image based on anatomic positioning information of the reinforced association features, constructing feature conduction sub-links in the key region, and connecting all feature conduction sub-links in series by fusing feature nodes across regions to form a defect feature conduction link covered in a whole domain; and extracting node association data, conduction path information and defect matching results in the defect feature conduction link, and integrating according to medical image report specifications to form a structural image identification report, wherein the structural image identification report comprises a feature conduction map, defect association description and anatomical positioning labels. In still another aspect, an embodiment of the present invention further provides a medical image feature conduction identification system, including: The medical image feature conduction identification method comprises the steps of storing a medical image feature conduction identification method, a processor, a machine-readable storage medium and a machine-executable instruction of the processor, wherein the processor is configured to execute the medical image feature conduction identification method through executing the machine-executable instruction. In yet another aspect, embodiments of the present invention further provide a computer program product comprising machine-executable instructions stored in a computer-readable storage medium, from which a processor of a medical image feature conduction identification system reads the machine-executable instructions, the processor executing the machine-executable instructions, causing the medical image feature conduction identification system to perform the medical image feature conduction identification method described above. Based on the above aspects, the defect knowledge graph built in the large model is called and analyzed to generate the image feature conduction reference,