Medical image quality control and organ phenotype extraction system

Abstract

The invention provides a medical image quality control and organ phenotype extraction system, relates to the technical field of image analysis, and constructs a topological relation tree of organ phenotypes according to relevance of anatomical structures among organ phenotype features in a medical image. Determining a structure association graph of the organ phenotype in the medical image based on the topological relation tree and the distribution information of the organ contour in the organ phenotype feature; determining a semantic structure chart of the medical entities in the text report through a context relationship of clinical description statements in the text report and a logic association relationship between the medical entities; performing node-level alignment fusion on the structure association graph of the organ phenotype and the semantic structure graph of the medical entity to obtain an organ-entity feature association graph; and performing quality control judgment on the medical image data based on the feature correlation map, and if the medical image data is judged to have imaging quality abnormality, outputting a quality control warning. Based on the scheme, the multi-modal correlation quality control of the medical image data can be realized, so that the reliability of the data is improved.

Description

Technical Field

[0001] This application relates to the field of image analysis technology. More specifically, this application relates to a system for medical image quality control and organ phenotype extraction. Background Art

[0002] Medical image quality control is a key link to ensure the accuracy of image diagnosis. Traditional quality control relies on manual review, which has problems such as strong subjectivity and low efficiency, and is prone to detail omissions and diagnostic deviations. Image analysis technology automatically identifies image features through algorithms to achieve objective quantitative evaluation of quality parameters such as image clarity, noise, and artifacts, improving detection consistency and efficiency. Using methods such as deep learning and edge detection, it can also automatically detect equipment failures or operation errors, reducing human intervention. Its application makes up for the deficiencies of traditional quality control in standardization, real-time performance, and detail recognition ability, significantly enhancing the accuracy and stability of medical image quality control.

[0003] Existing medical image quality control technologies mainly rely on the independent evaluation of single-modal data. Usually, parameters such as image-level clarity, noise, and artifacts are used as the basis for quality control, ignoring the internal correlation between multi-modal data such as images and text reports, clinical descriptions, etc., resulting in the lack of clinical context support for quality control results and the problem of insufficient overall reliability despite accurate local evaluation. In addition, most existing methods do not fully utilize the correspondence between medical entity information in text reports and image data, and fail to assist in verifying image quality at the semantic level. The quality control process has the problem of over-reliance on a single data source, affecting the overall data credibility. Facing defects such as insufficient multi-modal data fusion, limited understanding of organ phenotype structures, and ineffective utilization of clinical semantic information, existing technologies are difficult to meet the requirements of highly reliable medical image quality control. Therefore, how to achieve multi-modal associated quality control of medical image data to improve data reliability has become a difficult problem faced by the industry. Summary of the Invention

[0004] This application provides a system for medical image quality control and organ phenotype extraction, which can achieve multi-modal associated quality control of medical image data, thereby improving data reliability.

[0005] In a first aspect, this application provides a system for medical image quality control and organ phenotype extraction. The system specifically includes: A data acquisition module for acquiring medical image data to be quality-controlled, where the medical image data includes medical images and text reports; An organ phenotype extraction module for constructing a topological relationship tree of organ phenotypes based on the anatomical structure correlation between organ phenotype features in the medical image, and then determining the structural association diagram of organ phenotypes in the medical image based on the topological relationship tree and the distribution information of organ contours in organ phenotype features; A medical entity association module, which is used to determine the semantic structure diagram of medical entities in the text report through the context relationship of clinical description statements in the text report and the logical association relationship between medical entities; A feature fusion module, which is used to perform node-level alignment and fusion on the structural association diagram of the organ phenotype and the semantic structure diagram of the medical entity to obtain a feature association map of organ-entity; A quality control determination module, which is used to perform quality control determination on the medical image data based on the feature association map. If it is determined that there is an abnormal imaging quality in the medical image data, a quality control warning is output.

[0006] In this embodiment, medical image data to be detected for quality control is collected from the hospital PACS system.

[0007] In this embodiment, the format of the medical image is DICOM format.

[0008] In this embodiment, the format of the text report is HL7 format.

[0009] In this embodiment, constructing a topological relationship tree of organ phenotypes according to the anatomical structure association between organ phenotype features in the medical image specifically includes: Using a pre-trained organ segmentation model to segment organs in the medical image and extract various organ phenotype features; Defining the parent-child association relationship between various organ phenotype features based on the spatial position and anatomical adjacency relationship of organs in the human body; Regarding each organ phenotype feature as a node of the tree and the parent-child association relationship between each organ phenotype feature as the edge of the tree, thereby obtaining the topological relationship tree of organ phenotypes.

[0010] In this embodiment, determining the structural association diagram of organ phenotypes in the medical image based on the topological relationship tree and the distribution information of organ contours in organ phenotype features specifically includes: Extracting the distribution information of organ contours in each organ phenotype feature; Determining the attribute information of each organ phenotype according to the distribution information of each organ contour; Determining the attribute relationship between each organ phenotype feature through all the attribute information; Adjusting the edges in the topological relationship tree through all the attribute relationships, and using the adjusted topological relationship tree as the structural association diagram of organ phenotypes in the medical image.

[0011] In this embodiment, determining the semantic structure diagram of medical entities in the text report through the context relationship of clinical description statements in the text report and the logical association relationship between medical entities specifically includes: Perform text entity recognition on the text report and extract medical entities in the text report; Perform dependency syntactic analysis on the clinical description statements based on all medical entities to obtain the context relationship between medical entities; Determine the relationship coefficient between medical entities through the context relationship and the logical association relationship between medical entities; Take medical entities as nodes and the relationship coefficients between medical entities as edges, thereby obtaining the semantic structure diagram of medical entities in the text report.

[0012] In this embodiment, performing node-level alignment and fusion on the structural association diagram of the organ phenotype and the semantic structure diagram of the medical entity to obtain the feature association map of organ-entity specifically includes: Align the corresponding nodes in the structural association diagram of the organ phenotype and the semantic structure diagram of the medical entity to obtain multiple alignment and fusion points; Perform semantic matching on the aligned structural association diagram and semantic structure diagram to obtain the feature association values of each alignment and fusion point; Determine the feature association map of organ-entity through all the feature association values.

[0013] In this embodiment, performing quality control determination on the medical image data based on the feature association map specifically includes: Perform modal association evaluation on the imaging quality of the medical image data through the feature association map to obtain the quality evaluation value of the medical image data; If the quality evaluation value is lower than the preset imaging abnormality threshold, it is determined that the medical image data has imaging quality abnormality.

[0014] In this embodiment, if it is determined that the medical image data has imaging quality abnormality, outputting a quality control warning means that when the quality control determination module determines that the medical image data has imaging quality abnormality, it prompts relevant personnel that there is a quality problem with the image data by means of a warning notice.

[0015] The technical solutions provided by the disclosed embodiments of this application have the following beneficial effects: First, collect medical image data to be quality-controlled, where the medical image data includes medical images and text reports; construct a topological relationship tree of organ phenotypes based on the anatomical structure relevance between the phenotypic characteristics of each organ in the medical images, and then determine the structural association graph of organ phenotypes in the medical images based on the topological relationship tree and the distribution information of organ contours in the phenotypic characteristics of the organs; determine the semantic structure graph of medical entities in the text report through the context relationship of clinical description statements in the text report and the logical association relationship between medical entities; perform node-level alignment and fusion on the structural association graph of organ phenotypes and the semantic structure graph of medical entities to obtain a feature association graph of organ-entities; perform quality control determination on the medical image data based on the feature association graph. If it is determined that the medical image data has abnormal imaging quality, output a quality control warning.

[0016] It can be seen that the solution of this application performs node-level alignment and fusion on the structural association graph of organ phenotypes and the semantic structure graph of medical entities to obtain a feature association graph of organ-entities, and then performs quality control determination on the medical image data based on the feature association graph. First, the organ phenotype extraction module systematically models the phenotypic characteristics between organs by constructing a topological relationship tree of organ phenotypes and combining anatomical structure relevance, effectively solving the problem of incomplete understanding of organ structures in the prior art. Further, a structural association graph is generated through the distribution information of organ contours, realizing the precise identification of hidden imaging abnormalities such as contour missing or dislocation, and improving the meticulousness of image-level quality control. Second, the medical entity association module constructs a semantic structure graph based on the context relationship in the text report and the logical association of medical entities, making up for the shortcoming of not using text semantics for quality control verification in traditional methods, enabling the image quality control results to have clinical context support, and enhancing the clinical applicability and interpretability of quality control. Then, the feature fusion module fuses the structural association graph of organ phenotypes and the semantic structure graph of medical entities through node-level alignment, establishes a cross-modal feature association graph of organ-entities, breaks through the information barrier between image data and text data, solves the problem of insufficient multi-modal data fusion, and realizes information complementarity. Finally, the quality control determination module performs quality control analysis based on the feature association graph, can comprehensively consider the image quality and text semantic consistency, precisely identify imaging quality abnormalities, avoid the deviation caused by single data source determination, and improve the reliability and intelligent level of overall quality control determination.

[0017] In summary, the technical solution of this application can achieve multi-modal associated quality control of medical image data, thereby improving the reliability of the data. Brief Description of the Drawings

[0018] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only the embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0019] Figure 1 is a flowchart of a system for medical image quality control and organ phenotype extraction provided by the present application; Figure 2 is an exemplary flowchart for determining a topological relationship tree provided by the present application; Figure 3 is an exemplary flowchart for determining a semantic structure graph provided by the present application. Detailed implementation manners

[0020] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0021] The embodiments of the present application provide a system for medical image quality control and organ phenotype extraction. The core is to construct a topological relationship tree of organ phenotypes based on the anatomical structure correlation between the organ phenotype features in medical images, and determine the structural association graph of organ phenotypes in medical images based on the topological relationship tree and the distribution information of organ contours in organ phenotype features; determine the semantic structure graph of medical entities in the text report through the context relationship of clinical description statements in the text report and the logical association relationship between medical entities; perform node-level alignment and fusion on the structural association graph of organ phenotypes and the semantic structure graph of medical entities to obtain a feature association map of organ-entities; perform quality control determination on medical image data based on the feature association map. If it is determined that the medical image data has abnormal imaging quality, a quality control warning is output. Based on the above solution, multi-modal associated quality control of medical image data can be realized, thereby improving the reliability of the data.

[0022] In Embodiment 1, to better understand the above technical solution, the following will describe the above technical solution in detail with reference to the accompanying drawings of the specification and specific implementation manners. Refer to Figure 1 As shown, this figure is an exemplary flowchart of a system for medical image quality control and organ phenotype extraction according to an embodiment of the present application. The system specifically includes: A data acquisition module 100, configured to acquire medical image data to be quality-controlled, where the medical image data includes medical images and text reports.

[0023] It should be noted that the medical image data to be quality-controlled can be collected from the Picture Archiving and Communication System (PACS) of a hospital. The medical image data includes medical images and text reports. Among them, the format of the medical images is DICOM (Digital Imaging and Communications in Medicine) format, and the format of the text reports is HL7 (Health Level Seven) format.

[0024] In specific implementation, first, medical images are obtained from the hospital PACS system through the DICOM protocol, and DICOM service classes (such as C-FIND, C-MOVE) are used to query and transmit the medical images to ensure that the medical images are stored in the standard DICOM format. Then, text reports are extracted from the PACS system or the hospital information system through the HL7 protocol, and HL7 messages (such as ORU, ADT) are parsed to obtain text information such as diagnostic reports. Finally, the collected DICOM images and HL7 text reports are transmitted to the local or quality control system, and data verification and preprocessing are performed to ensure data integrity and availability.

[0025] The organ phenotype extraction module 200 is used to construct a topological relationship tree of organ phenotypes according to the anatomical structure relevance between the organ phenotype features in the medical image, and then determine the structural association graph of the organ phenotypes in the medical image based on the topological relationship tree and the distribution information of the organ contours in the organ phenotype features.

[0026] In this embodiment, refer to Figure 2 As shown, this figure is an exemplary flowchart for determining the topological relationship tree in the embodiment of the present application. The construction of the topological relationship tree of organ phenotypes according to the anatomical structure relevance between the organ phenotype features in the medical image in this embodiment can be implemented by the following steps: In step S21, an organ segmentation model trained in advance is used to segment the medical image to extract each organ phenotype feature. In step S22, the parent-child association relationship between each organ phenotype feature is defined based on the spatial position and anatomical adjacency relationship of the organs in the human body. In step S23, each organ phenotype feature is used as a node of the tree, and the parent-child association relationship between each organ phenotype feature is used as the edge of the tree, thereby obtaining the topological relationship tree of organ phenotypes.

[0027] It should be noted that the organ phenotypic features in this application refer to the features that can characterize the external morphology and structural state of organs in medical images; the parent-child association relationship in this application refers to the spatial subordination relationship between organs in the human anatomical structure, where the "parent" organ is the organ in the upper or inclusion relationship in the structure, and the "child" organ is the organ attached to or included in the "parent" organ. The parent-child association relationship can reflect the hierarchical relationship of organs in terms of spatial position and anatomical function. For example, the liver is the parent organ tissue of the gallbladder, and the lung is the parent organ of the fetus. Constructing the association between organs through this relationship helps to form a topological structure model with human anatomical logic; the topological relationship tree in this application refers to a directed tree-like data structure constructed based on the anatomical structure and spatial relationship between organs, and this topological relationship tree is used to reflect the spatial dependence, anatomical genus, and tissue characteristics of the connection between organs.

[0028] In specific implementation, a pre-trained organ segmentation model (such as the U-Net, nnUNet, or DeepLab deep learning model) can be used to segment organs in medical images. By inputting medical images in DICOM format, the organ segmentation model outputs the organ labels of each pixel, and the phenotypic features of each organ are extracted; secondly, based on medical anatomy knowledge, the spatial position and anatomical adjacency relationship between organs (such as the adjacency relationship between the liver and the gallbladder, and the connection relationship between the heart and the large blood vessels) are defined, and the spatial position and anatomical adjacency relationship are mapped into a relationship vector, and this relationship vector is used as the parent-child association relationship between the phenotypic features of each organ; finally, each organ phenotypic feature is used as a node of the tree, and the parent-child association relationship between the organ phenotypic features is used as the edge of the tree. Using graph theory or tree structure algorithms (such as the minimum spanning tree or hierarchical clustering) to construct the topological relationship tree of the organ phenotype can intuitively represent the anatomical structure correlation between organs. It should be noted that the embodiments of this application combine deep learning segmentation technology and anatomical knowledge to achieve the automatic construction of the organ topological relationship tree from medical images.

[0029] In this embodiment, determining the structural association graph of the organ phenotype in the medical image based on the distribution information of the organ contours in the topological relationship tree and the organ phenotypic features can be implemented by the following steps: Extract the distribution information of the organ contours in each organ phenotypic feature; Determine the attribute information of each organ phenotype according to the distribution information of each organ contour; Determine the attribute relationship between each organ phenotypic feature through all the attribute information; Adjust the edges in the topological relationship tree through all the attribute relationships, and use the adjusted topological relationship tree as the structural association graph of the organ phenotype in the medical image.

[0030] It should be noted that the structural association diagram in this application refers to an undirected relational graph structure constructed based on the characteristic attributes of the organ table. Each node represents the table phenotypic type of different organs in the medical relationship image, and each edge represents the connection established between organs based on attribute relationships and structural relationships. The overall graph not only retains the anatomical hierarchy of the organs but also reflects the actual connections between organs in terms of morphology, space, and function, and is used for visualizing and analyzing the overall structural layout and interconnection of organs, providing comprehensive and reliable data support for medical image quality control or disease analysis.

[0031] In specific implementation, first, use existing medical image processing techniques (such as a segmentation network based on deep learning) to extract the segmentation information of each organ. Models such as U-Net can accurately segment organs in medical images, thereby extracting the external boundaries of each organ and obtaining the segmentation information of each organ (such as the modeled shape); second, the size, shape, and spatial distribution parameters in the distribution information of the contours of each organ can be used as the attribute information of the phenotypic types of each organ; then, for every two organ phenotypic types, the Euclidean distance between the attribute information of the two organ phenotypic types can be used as the attribute relationship between the two organ phenotypic features, and thus the attribute relationship between every two organ phenotypic features can be obtained; finally, the natural exponential function value of the opposite number of the attribute relationship between the two organ phenotypic features (nodes) can be used as the adjustment coefficient of the edge between the two organ phenotypic features (nodes), and then the product of the adjustment coefficient and the corresponding edge (i.e., the parent-child association relationship) can be used as the adjusted edge in the topological relationship tree. Through the above method, all adjusted edges can be obtained, and then all the adjusted edges are used to replace the edges before adjustment. The topological relationship tree after the edges are replaced is used as the structural association diagram of the organ phenotypic types in the medical image.

[0032] The medical entity association module 300 is used to determine the semantic structure diagram of the medical entities in the text report through the context relationship of the clinical description statements in the text report and the logical association relationship between medical entities.

[0033] In this embodiment, refer to Figure 3 As shown, this figure is an exemplary flowchart for determining the semantic structure diagram in the embodiment of this application. In this embodiment, the semantic structure diagram of the medical entities in the text report can be determined through the context relationship of the clinical description statements in the text report and the logical association relationship between medical entities, and can be implemented by the following steps: In step S31, perform text entity recognition on the text report to extract medical entities in the text report; In step S32, perform dependency syntactic analysis on the clinical description statements based on all medical entities to obtain the context relationship between medical entities; In step S33, determine the relationship coefficient between medical entities through the context relationship between medical entities and the logical association relationship between medical entities; In step S34, the medical entities are used as nodes, and the relationship coefficients between the medical entities are used as edges, thereby obtaining the semantic structure graph of the medical entities in the text report.

[0034] It should be noted that the medical entities in this application refer to words with clear medical meanings in medical texts and can individually refer to specific medical concepts, usually including disease names, symptom manifestations, examination indicators, drug names, anatomical parts, surgical operations, and pathological states; the relationship coefficients in this application are indicators for measuring the strength of the relationship between medical entities; the semantic structure graph in this application refers to a graphical structure that connects each node with edges representing the semantic relationships between medical entities, aiming to comprehensively present the semantic associations and logical connections between the medical entities in the text.

[0035] In specific implementation, first, an existing pre-trained entity recognition model (such as BioBERT or MedNER) can be used to perform entity recognition on the clinical description statements in the text report, and extract medical entity words related to diseases, symptoms, examination indicators, and drugs; second, a tool based on dependency syntactic analysis (such as a medical extension model based on SpaCy or Stanza) is used to perform dependency syntactic parsing on each clinical description statement in the text report to clarify the syntactic dependency relationships of each medical entity in the sentence, and extract the context relationships of subject-predicate-object, modification, and coordination; then, combined with the entity logical associations in the medical knowledge graph (such as the causal, concurrent, and subordinate relationships in UMLS or a self-built medical knowledge base), the dependency relationships and logical associations between each pair of medical entities are scored through a graph convolutional network (GCN) or a relation classifier (such as an RE model), and the score is used as the relationship coefficient between the medical entities. The strength of the association between medical entities can be measured through the relationship coefficient; finally, all the extracted medical entities are used as the nodes of the graph, and the relationship coefficients between the medical entities are used as the edges of the graph. A graph construction tool (such as NetworkX) is used to construct the semantic structure graph of the medical entities, that is, all the nodes and all the graphs can be substituted into the graph construction tool, and the output result is used as the semantic structure graph of the medical entities in the text report, thereby realizing the visualization and structuring of semantic associations.

[0036] The feature fusion module 400 is used to perform node-level alignment and fusion on the structure association graph of the organ phenotype and the semantic structure graph of the medical entities to obtain an organ-entity feature association map.

[0037] In this embodiment, the node-level alignment and fusion of the structure association graph of the organ phenotype and the semantic structure graph of the medical entities to obtain an organ-entity feature association map can be implemented by the following steps: Align the structural association graph of the organ phenotype with the corresponding nodes in the semantic structure graph of medical entities to obtain multiple alignment and fusion points; Perform semantic matching on the aligned structural association graph and semantic structure graph to obtain the feature association values of each alignment and fusion point; Determine the organ-entity feature association map through all the feature association values.

[0038] It should be noted that the alignment and fusion points in this application refer to the corresponding medical entity nodes with the same or similar semantics in the structural association graph of the organ phenotype and the semantic structure graph of medical entities, and are represented by nodes in the graph; the feature association value in this application refers to a quantitative index of the association strength between the organ phenotype node and the medical entity node in terms of semantics and structure in the alignment and fusion point; the feature association map in this application is a graph structure used to express and analyze the multi-level and multi-dimensional association relationships between different medical entities and organ phenotypes, aiming to reveal the semantic and structural connections between entities through the quantitative feature association values.

[0039] In specific implementation, first, a graph matching algorithm (such as graph isomorphism detection based on the adjacency matrix or similarity calculation method based on node embedding) can be used to align the nodes with the same or similar semantics in the structural association graph of the organ phenotype and the semantic structure graph of medical entities. Usually, GraphSAGE or Node2Vec is used to vectorize the nodes of the two graphs, and then the cosine similarity or Euclidean distance is used to calculate the node similarity, and multiple high-similarity alignment and fusion points are screened out; second, apply a cross-graph neural network (such as Graph Matching Network, GMN) to perform feature comparison on the local structure and context around the alignment and fusion points to achieve fine-grained semantic matching of the alignment and fusion points. Further, combined with the prior knowledge in the medical ontology (such as UMLS or self-built phenotype-entity mapping library), calculate the feature association value of each alignment and fusion point, and the connection strength and type between the organ phenotype and the medical entity can be quantified through the feature association value; finally, based on the feature association values of all alignment and fusion points, adopt a graph fusion strategy (such as weighted edge fusion or attention mechanism fusion) to integrate the two graphs (structural association graph and semantic structure graph) into a unified organ-entity feature association map, and through the organ-entity feature association map, a systematic modeling of the multi-level and multi-dimensional semantic relationships between the organ phenotype and the medical entity can be realized.

[0040] The quality control determination module 500 is used to perform quality control determination on the medical image data based on the feature association map. If it is determined that there is an abnormal imaging quality in the medical image data, a quality control warning is output.

[0041] In this embodiment, the quality control determination of the medical image data based on the feature association map can be implemented by the following steps: Perform modal correlation assessment on the imaging quality of the medical image data through the feature correlation map to obtain the quality assessment value of the medical image data; If the quality assessment value is lower than the preset imaging abnormality threshold, it is determined that the medical image data has imaging quality abnormalities.

[0042] It should be noted that the quality assessment value in this application is an indicator for measuring the quality of image data; the preset imaging abnormality threshold in this application can be set by analyzing the quality distribution of historical medical image data to set a value that can effectively distinguish normal and abnormal image quality.

[0043] In specific implementation, first, a multi-modal graph neural network (such as Graph Convolutional Network, GCN) can be used to analyze the feature correlation map, which contains medical entity and organ phenotype information in image data and text reports. The multi-modal graph neural network can fuse data of different modalities through graph embedding technology, combine the medical entities and features extracted from image data and text reports to capture the correlation between the two. In the modal correlation assessment stage, graph convolution operations can be used to model the interaction relationship between image data and text data, thereby generating a quality assessment value. This quality assessment can reflect the consistency between image quality and text description. To improve the assessment accuracy, an attention mechanism (such as Graph Attention Network, GAT) can also be introduced to weight the information according to the feature importance of each node and enhance the influence of key features; then, by comparing with the preset imaging abnormality threshold, if the quality assessment value is lower than the preset imaging abnormality threshold, it is determined that the image data has imaging quality abnormalities. Through this determination method, the quality control system can automatically identify potential problems in the images, thereby improving the quality control efficiency and accuracy.

[0044] It should be noted that in this application, if it is determined that the medical image data has imaging quality abnormalities, outputting a quality control warning means that when the quality control determination module determines that the medical image data has imaging quality abnormalities, relevant personnel are prompted through a warning notification that the image data has quality problems and needs further inspection or processing. The warning is usually presented in visual or audible ways, such as pop-up prompts, color markings, alarm sounds, etc., to ensure that quality control personnel can discover problems in a timely manner and take corresponding measures.

[0045] It can be seen that the solution of this application performs node-level alignment and fusion on the structural association graph of organ phenotypes and the semantic structure graph of medical entities to obtain an organ-entity feature association graph, and then performs quality control determination on the medical image data based on the feature association graph. First, the organ phenotype extraction module constructs an organ phenotype topological relationship tree, and systematically models the phenotypic features between organs in combination with anatomical structure relevance, effectively solving the problem of incomplete understanding of organ structures in the prior art. Further, a structural association graph is generated through organ contour distribution information to achieve accurate identification of hidden imaging abnormalities such as contour loss or dislocation, and improve the meticulousness of image-level quality control. Secondly, the medical entity association module constructs a semantic structure graph based on the context relationship in the text report and the logical association of medical entities, making up for the shortcoming of not using text semantics for quality control verification in traditional methods, enabling the image quality control results to have clinical context support, and enhancing the clinical applicability and interpretability of quality control. Then, the feature fusion module fuses the organ phenotype structural association graph and the medical entity semantic structure graph through node-level alignment, establishes a cross-modal organ-entity feature association graph, breaks through the information barrier between image data and text data, solves the problem of insufficient multi-modal data fusion, and realizes information complementarity. Finally, the quality control determination module performs quality control analysis based on the feature association graph, can comprehensively consider the image quality and text semantic consistency, accurately identify imaging quality abnormalities, avoid the deviation caused by single data source determination, and improve the reliability and intelligent level of overall quality control determination.

[0046] In summary, the technical solution of this application can achieve multi-modal associated quality control of medical image data, thereby improving the reliability of the data.

[0047] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of this application. It should be understood that each process and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions, and the combination of processes and / or blocks in the flowcharts and / or block diagrams can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or a device for implementing the functions specified in multiple blocks.

[0048] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing relevant hardware through a program, and this program can be stored in a computer-readable storage medium. The storage medium includes read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc memories, magnetic disc memories, tape memories, or any other medium that can be used to carry or store data and is computer-readable.

[0049] It should also be noted that the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent in such a process, method, commodity or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, commodity or device including the element.

Claims

1. A system for medical image quality control and organ phenotype extraction, characterized in that, The system specifically includes: A data acquisition module, which is used to acquire medical image data to be quality controlled, where the medical image data includes medical images and text reports; An organ phenotype extraction module, which is used to construct a topological relationship tree of organ phenotypes according to the anatomical structure relevance among the organ phenotype features in the medical image, and then determine the structural association graph of organ phenotypes in the medical image based on the topological relationship tree and the distribution information of organ contours in the organ phenotype features; A medical entity association module, which is used to determine the semantic structure graph of medical entities in the text report through the context relationship of clinical description statements in the text report and the logical association relationship among medical entities; A feature fusion module, which is used to perform node-level alignment and fusion on the structural association graph of organ phenotypes and the semantic structure graph of medical entities to obtain a feature association graph of organ-entity; A quality control determination module, which is used to perform quality control determination on the medical image data based on the feature association graph. If it is determined that the medical image data has abnormal imaging quality, a quality control warning is output.

2. The system for medical image quality control and organ phenotype extraction according to claim 1, characterized in that, Collect medical image data to be quality controlled from the hospital PACS system.

3. The system for medical image quality control and organ phenotype extraction according to claim 1, characterized in that, The format of the medical image is DICOM format.

4. A system for medical image quality control and organ phenotype extraction according to claim 1, characterized in that, The format of the text report is HL7 format.

5. The system for medical image quality control and organ phenotype extraction according to claim 1, wherein Specifically, constructing a topological relationship tree of organ phenotypes according to the anatomical structure relevance among the organ phenotype features in the medical image includes: Using a pre-trained organ segmentation model to segment the organs in the medical image and extract each organ phenotype feature; Defining the parent-child association relationship among each organ phenotype feature based on the spatial position and anatomical adjacency relationship of the organs in the human body; Taking each organ phenotype feature as a node of the tree and taking the parent-child association relationship among each organ phenotype feature as the edge of the tree, so as to obtain the topological relationship tree of organ phenotypes.

6. The system for medical image quality control and organ phenotype extraction according to claim 1, wherein Specifically, determining the structural association graph of organ phenotypes in the medical image based on the topological relationship tree and the distribution information of organ contours in the organ phenotype features includes: Extracting the distribution information of organ contours in each organ phenotype feature; Determining the attribute information of each organ phenotype according to the distribution information of each organ contour; Determining the attribute relationship among each organ phenotype feature through all the attribute information; Adjusting the edges in the topological relationship tree through all the attribute relationships, and taking the adjusted topological relationship tree as the structural association graph of organ phenotypes in the medical image.

7. The system for medical image quality control and organ phenotype extraction according to claim 1, characterized in that, Specifically, determining the semantic structure graph of medical entities in the text report through the context relationship of clinical description statements in the text report and the logical association relationship among medical entities includes: Performing text entity recognition on the text report to extract medical entities in the text report; Performing dependency syntactic analysis on the clinical description statements based on all the medical entities to obtain the context relationship among medical entities; Determining the relationship coefficient among medical entities through the context relationship among medical entities and the logical association relationship among medical entities; Taking medical entities as nodes and taking the relationship coefficient among medical entities as edges, so as to obtain the semantic structure graph of medical entities in the text report.

8. A system for medical image quality control and organ phenotype extraction according to claim 1, characterized in that, Performing node-level alignment and fusion on the structural association graph of the organ phenotype and the semantic structure graph of the medical entity to obtain the feature association map of organ-entity specifically includes: Aligning the corresponding nodes in the structural association graph of the organ phenotype and the semantic structure graph of the medical entity to obtain multiple alignment and fusion points; Performing semantic matching on the aligned structural association graph and semantic structure graph to obtain the feature association values of each alignment and fusion point; Determining the feature association map of organ-entity through all the feature association values.

9. A system for medical image quality control and organ phenotype extraction according to claim 1, characterized in that, Performing quality control determination on the medical image data based on the feature association map specifically includes: Performing modal association evaluation on the imaging quality of the medical image data through the feature association map to obtain the quality evaluation value of the medical image data; If the quality evaluation value is lower than the preset imaging abnormality threshold, it is determined that the medical image data has imaging quality abnormality.

10. A system for medical image quality control and organ phenotype extraction according to claim 1, characterized in that, If it is determined that the medical image data has imaging quality abnormality, outputting a quality control warning means that when the quality control determination module determines that the medical image data has imaging quality abnormality, it prompts relevant personnel that there is a quality problem with the image data by means of a warning notice.

Images