Medical Knowledge Graph-based Imaging Diagnosis Aided Decision-making System

Abstract

The present invention discloses an image diagnosis auxiliary decision-making system based on a medical knowledge graph, which relates to the technical field of medical auxiliary diagnosis. It includes an image data set establishment module, an image data set analysis module, a 24-dimensional bounding box acquisition module, a fuzzy processing module, and a fracture type judgment module. By analyzing the 24-dimensional vector forms of different fracture types, the 24-dimensional bounding boxes corresponding to each fracture type are obtained, providing clear boundary intervals. By performing fuzzy processing on the boundary intervals of the 24-dimensional bounding boxes, the possible similarities and variabilities between different fracture types are considered, improving the flexibility and fault tolerance when judging the fracture type of the image to be detected, being able to make reasonable judgments on complex or unclear diagnostic images, providing a wider reference range for the judgment of fracture types, reducing the risk of misjudgment caused by eigenvalue approaching the boundary, and improving the accuracy of diagnosis.

Description

Technical Field

[0001] The present invention belongs to the field of medical auxiliary diagnosis technology, and specifically is an image diagnosis auxiliary decision-making system based on medical knowledge graph. Background Art

[0002] In recent years, with the continuous advancement of medical imaging technology, medical knowledge graphs, as an emerging medical information technology, have provided more systematic and knowledge-based support for imaging diagnosis. Medical knowledge graphs can structure a large amount of medical data, medical knowledge, and clinical experience, allowing machine learning and deep learning algorithms to better mine valuable information from them, thereby providing more accurate decisions for imaging diagnosis systems.

[0003] Patent publication number CN110911009A discloses a clinical diagnosis decision support system and a method for accumulating medical knowledge graphs. The clinical diagnosis decision support system includes: a vital sign acquisition module, a medical image acquisition module, a central control module, a diagnosis and analysis module, a treatment plan recommendation module, a medical knowledge retrieval module, a graph creation module, a data storage module, and a display module. The present invention uses the historical data of various diseases to train the model through the diagnosis and analysis module, so that only one model can be used to pre-diagnose multiple diseases. It is very suitable for clinical diagnosis management and maintenance, and the diagnosis and analysis results are accurate. At the same time, a medical knowledge graph is created through the graph creation module, and medical data is managed through the medical knowledge graph. Therefore, when using medical data, medical data can be extracted through the medical knowledge graph, which can improve the convenience of using medical data to a certain extent.

[0004] However, at present, most applications of medical knowledge graphs are mainly concentrated on the processing and analysis of text information, and the in-depth analysis and diagnostic support for medical images are relatively weak. In some special cases, especially in the classification of fracture types, they only rely on the doctor's experience and visual judgment and simple feature matching or rough pattern recognition, which leads to the shortcomings of diagnostic results being easily affected by human factors and low diagnostic efficiency. At the same time, due to the inability to deeply explore the texture features and subtle differences in the image, it is impossible to accurately judge different types of fractures and unable to assist doctors in making decisions in real time. Based on this, an image diagnosis decision support system based on medical knowledge graphs is proposed. Summary of the invention

[0005] The purpose of the present invention is to provide an image diagnosis decision-making support system based on a medical knowledge graph, which solves the technical problem that it is impossible to deeply explore the texture features and slight differences in the image, resulting in the inability to accurately judge different types of fractures and the inability to assist doctors in making decisions in real time.

[0006] An image diagnosis auxiliary decision-making system based on a medical knowledge graph, comprising:

[0007] An image dataset establishment module that establishes an image dataset through multiple distal radius fracture images;

[0008] An image dataset analysis module that extracts and analyzes the texture information corresponding to each pixel point in each preprocessed distal radius fracture image to obtain the 24-dimensional vector form of each distal radius fracture image;

[0009] A 24-dimensional bounding box acquisition module that analyzes each dimension value in the 24-dimensional vector form corresponding to each distal radius fracture image under different fracture types to obtain the 24-dimensional bounding boxes of different fracture types and the boundary intervals of each dimension of the 24-dimensional bounding boxes of different fracture types;

[0010] A fuzzy processing module that performs fuzzy processing on the boundary intervals of each dimension of the 24-dimensional bounding boxes of different fracture types to obtain the fuzzy intervals corresponding to the boundary intervals of each dimension of the 24-dimensional bounding boxes of different fracture types;

[0011] A fracture type judgment module that compares and analyzes the dimension values in the real-time 24-dimensional vector form of the image to be detected and diagnosed with the boundary intervals and fuzzy intervals of each dimension of the 24-dimensional bounding boxes of different fracture types to judge the fracture type corresponding to the image to be detected and diagnosed.

[0012] As a further solution of the present invention: The specific method for obtaining the 24-dimensional vector form of each distal radius fracture image is as follows:

[0013] First, a feature value extraction unit obtains the texture feature values corresponding to each pixel point in each distal radius fracture image at different scales and directions according to the texture information of each distal radius fracture image. Then, a feature value analysis unit analyzes the texture feature values corresponding to each pixel point in each distal radius fracture image at different scales and directions, and obtains the calculated values corresponding to each distal radius fracture image at different scales and directions according to the analysis results. Then, through a normalization processing unit, the mean and variance in the calculated values corresponding to each distal radius fracture image at different scales and directions are both normalized to obtain the calibrated values of the calculated values corresponding to each distal radius fracture image at different scales and directions. Finally, through a 24-dimensional vector form acquisition unit, the calculated values of each distal radius fracture image are vectorized to obtain the 24-dimensional vector form of each distal radius fracture image. Here, the number of directions is selected as 4, namely 0°, 45°, 90°, and 135°, and the different scales are 1, 2, and 3 respectively.

[0014] The specific method for obtaining the calculated values corresponding to each distal radius fracture image at different scales and directions is as follows:

[0015] Select a fixed scale and direction from different scales and directions as the target scale direction, obtain the texture feature values corresponding to each pixel point in each distal radius fracture image within the target scale direction, and take the mean and variance of the texture feature values in each distal radius fracture image as the calculated values corresponding to each distal radius fracture image within the target scale direction;

[0016] Using the same analysis method, obtain the calculated values Tnij (Anij, Bnij) corresponding to each distal radius fracture image at different scales and directions, where Anij is the mean of the texture feature values corresponding to each distal radius fracture image at different scales and directions, Bnij represents the variance of the texture feature values corresponding to each distal radius fracture image at different scales and directions, n represents different distal radius fracture images, i represents different direction labels, i = 1, 2, 3, and 4, and j represents different scales, j = 1, 2, and 3.

[0017] The specific method for obtaining the calibration values of the calculated values corresponding to each distal radius fracture image at different scales and directions is as follows:

[0018] Through the formula: XAnij = (Anij - Anij min ) / (Anij max - Anij min ); XBnij = (Bnij - Bnij min ) / (Bnij max - Bnij min ), calculate and obtain the calibration values XTnij (XAnij, XBnij) of the calculated values corresponding to each distal radius fracture image at different scales and directions, where Anij max and Anij min are the maximum and minimum values in Anij respectively, and Bnij max and Bnij min are the maximum and minimum values in Bnij respectively.

[0019] The specific method for obtaining the 24-dimensional vector form of each distal radius fracture image is as follows:

[0020] According to the calibration values XTnij (XAnij, XBnij) of the calculated values corresponding to each distal radius fracture image at different scales and directions, the calculated values of each distal radius fracture image are vectorized to obtain the 24-dimensional vector form Ln (XAn11, XBn11, XAn12, XBn12, XAn13, XBn13,..., XAn43, XBn43) of each distal radius fracture image, and each vector form contains 24-dimensional numerical values.

[0021] As a further solution of the present invention: The specific method for obtaining the 24-dimensional bounding boxes of different fracture types and the boundary intervals of each dimension of the 24-dimensional bounding boxes of different fracture types is as follows:

[0022] The fracture types include type A, type B, and type C fracture types. First, the vector forms of multiple distal radius fracture images corresponding to type A fracture type are obtained as Le, where e represents different distal radius fracture images of type A fracture type. The mean value of the maximum and minimum values of each dimension value in the vector form Le of type A fracture type is marked as the reference value of each dimension, and the absolute value of the difference between the reference value of each dimension and the maximum and minimum values in the corresponding dimension value is added as the upper bound value corresponding to each dimension of type A fracture type. Then, the absolute value of the difference between the reference value of each dimension and the maximum and minimum values in the corresponding dimension value is subtracted from the reference value of each dimension as the lower bound value corresponding to each dimension of type A fracture type. Through the upper bound value WAr and the lower bound value WBr corresponding to each dimension of type A fracture type, the boundary interval QAr[WBr, WAr] corresponding to each dimension of type A fracture type is established. According to the boundary intervals corresponding to each dimension of type A fracture type, the 24-dimensional bounding box HA corresponding to type A fracture type is obtained, where r represents different dimensions, r = 1, 2,..., 24. The vector forms of multiple distal radius fracture images corresponding to type B and type C fracture types are analyzed in the same analysis method to obtain the 24-dimensional bounding boxes HB and HC corresponding to type B and type C fracture types respectively.

[0023] As a further solution of the present invention: The specific method for obtaining the fuzzy intervals corresponding to the boundary intervals of each dimension of the 24-dimensional bounding boxes of different fracture types is as follows:

[0024] Obtain the standard deviation Zr corresponding to each dimension value in the 24-dimensional vector morphology of type A fracture. Take the ratio between the standard deviation Zr of each dimension and the reference value as the buffer coefficient Cr of each dimension. Take the product between the buffer coefficient Cr of each dimension and the standard deviation Zr as the buffer value Fr of the boundary interval of each dimension. Take the sum of the upper bound value WAr corresponding to each dimension boundary interval of type A fracture and the buffer value Fr as the buffer upper limit value VAr corresponding to each dimension boundary interval of type A fracture. Take the absolute value of the difference between the lower bound value WBr corresponding to each dimension boundary interval of type A fracture and the buffer value Fr as the buffer lower limit value VBr corresponding to each dimension boundary interval of type A fracture. Further obtain the fuzzy interval MHr[VBr, VAr] corresponding to each dimension boundary interval of type A fracture. Use the same analysis method to perform fuzzy processing on the boundary intervals of each dimension of the 24-dimensional boundary boxes of type B and type C fractures, and obtain the fuzzy intervals corresponding to the boundary intervals of each dimension of the 24-dimensional boundary boxes of type B and type C fractures. When judging the dimension values in the real-time 24-dimensional vector morphology of the image to be detected and diagnosed, first judge whether its dimension value is within the boundary intervals of each dimension of the 24-dimensional boundary boxes of different fracture types. If it is, judge that the corresponding dimension value belongs to the boundary interval of the corresponding fracture type, and add 1 to the matching number of the boundary interval of the corresponding fracture type. If it is not, judge whether it is within the fuzzy intervals of each dimension of the 24-dimensional boundary boxes of each fracture type. If it is, judge that the corresponding dimension value belongs to the fuzzy interval of the corresponding fracture type, and add 1 to the matching number of the fuzzy interval of the corresponding fracture type. Otherwise, do nothing.

[0025] As a further solution of the present invention: The specific method for judging the fracture type corresponding to the image to be detected and diagnosed is:

[0026] Obtain the sum of the products of the number of dimension values in the real-time 24-dimensional vector morphology of the image to be detected and diagnosed that are respectively within the boundary intervals and fuzzy intervals of each dimension of the 24-dimensional boundary boxes of different fracture types and the preset coefficients β1 and β2, and take it as the selection coefficient corresponding to the image to be detected and diagnosed in the 24-dimensional boundary boxes of different fracture types. Take the fracture type corresponding to the maximum selection coefficient as the fracture type of the image to be detected and diagnosed. Here, 1 = β1 + β2, and 1 > β1 > 0.5 > β2 > 0.

[0027] When the selection coefficients of multiple fracture types are the same and the maximum, obtain the number of dimension values in the 24-dimensional vector morphology of the image to be detected and diagnosed that are respectively within the boundary intervals of the 24-dimensional boundary boxes with the same selection coefficient. Take the fracture type with the largest number as the fracture type corresponding to the image to be detected and diagnosed. If the numbers are still the same, output the fracture types corresponding to the same numbers simultaneously.

[0028] When the number of values in each dimension of the real-time 24-dimensional vector form of the diagnostic image to be detected located in the boundary intervals and fuzzy intervals of each dimension of the 24-dimensional boundary boxes corresponding to different fracture types is 0, the manual review process is triggered.

[0029] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0030] (1) In the present invention, by analyzing the 24-dimensional vector forms of different fracture types, 24-dimensional boundary boxes corresponding to each fracture type are obtained, providing clear boundary intervals, providing standardized and structured data support for fracture type classification, improving the reliability of classification judgment, and further enhancing the diagnostic accuracy of the system by setting clear boundary intervals, ensuring the efficiency and precision of the diagnostic process;

[0031] (2) In the present invention, by performing fuzzy processing on the boundary intervals of the 24-dimensional boundary boxes, the possible similarities and variabilities between different fracture types are considered, improving the flexibility and fault tolerance in judging the fracture type of the image to be detected, being able to make reasonable judgments on complex or unclear diagnostic images, providing a wider reference range for the judgment of fracture types, reducing the risk of misjudgment caused by eigenvalue approaching the boundary, and improving the diagnostic accuracy;

[0032] (3) In the present invention, by analyzing the 24-dimensional vector form of the image to be detected in real time and comparing it with the preset 24-dimensional boundary boxes and fuzzy intervals, the fracture type corresponding to the image to be detected can be automatically and accurately judged, improving the intelligence and automation level of fracture type classification, and enhancing the diagnostic speed and accuracy;

[0033] (4) In the present invention, when the selection coefficients of multiple fracture types are the same, by further comparing the number of dimensions in the boundary interval to determine the fracture type, it can effectively handle the diagnostic problems in complex situations. This multi-level judgment logic enables the system to make more reasonable decisions in the face of complex situations. At the same time, when the eigenvalue of the image to be detected cannot match the boundary interval or fuzzy interval of any fracture type, the manual review process can be triggered to ensure the final accuracy and reliability of the diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is a schematic diagram of the system framework structure of the present invention;

[0035] Figure 2 is a schematic diagram of the process for judging the number of values in each dimension of the real-time 24-dimensional vector form of the diagnostic image to be detected located in the boundary intervals and fuzzy intervals of each dimension of the 24-dimensional boundary boxes corresponding to different fracture types in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0036] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative work belong to the scope of protection of the present invention.

[0037] Embodiment 1: Please refer to Figure 1 - Figure 2 , the present application provides an image diagnosis auxiliary decision-making system based on a medical knowledge graph, including;

[0038] An image data set building module, configured to obtain a plurality of distal radius fracture images and build an image data set;

[0039] It should be noted that the acquisition source of the distal radius fracture images can be retrieved from the hospital database or extracted from some distal radius fracture images and related case information accumulated in previous studies. At the same time, when acquiring the distal radius fracture images, the basic information of the patients, that is, information such as the patient's name and ID number, was desensitized in advance to protect the privacy of the patients; the above are all existing and mature technologies, so they will not be elaborated here;

[0040] It should be noted that each fracture image has been labeled with the fracture type by experienced orthopedic surgeons;

[0041] According to the internationally accepted AO classification standard, the fracture types are divided into three types, type A, type B, and type C:

[0042] Type A: Fracture outside the joint, such as the bone breaks outside the joint.

[0043] Type B: Simple, minor or small part of the medial joint fracture, such as a small part of the bone breaks inside the joint.

[0044] Type C: Complex medial joint fracture, such as multiple fracture fragments of the bone inside the joint;

[0045] The internationally accepted AO classification standard is an existing internationally accepted classification standard and the current default classification standard. This classification is used by default in this embodiment, so it will not be elaborated here;

[0046] This module can quickly construct a high-quality and complete image data set by extracting distal radius fracture images from the hospital database or previous studies, protect the privacy of patients through desensitization processing, ensure compliance with data protection regulations, provide accurate and comprehensive input for subsequent data analysis, reduce the error rate in the data collection process, and lay a solid foundation for the accuracy and stability of the system.

[0047] An image dataset analysis module is used to preprocess each distal radius fracture image in the image dataset, extract and analyze the texture information corresponding to each pixel point in each distal radius fracture image, and obtain the 24-dimensional vector form of each distal radius fracture image according to the analysis results. The specific method is as follows:

[0048] Before analyzing the image dataset, it is necessary to preprocess each distal radius fracture image in the image dataset to improve the image quality. The preprocessing includes image enhancement, noise removal and other processing methods, which are all existing and mature technologies, so they will not be elaborated here.

[0049] A feature value extraction unit extracts the texture information of each distal radius fracture image at different scales and directions by setting different Gabor filters, and then obtains the texture feature values corresponding to each pixel point in each distal radius fracture image at different scales and directions;

[0050] It should be noted that: the number of directions selected here is 4, namely: such as 0°, 45°, 90°, 135°, and the number of scales selected is 3. Here, the different scales are 1, 2, and 3 respectively;

[0051] A feature value analysis unit is used to analyze the texture feature values corresponding to each pixel point in each distal radius fracture image at different scales and directions, and obtain the calculated values corresponding to each distal radius fracture image at different scales and directions according to the analysis results;

[0052] It should be noted that the texture feature value of each pixel point is the result obtained by convolving the Gabor filter with the corresponding area in the image. The texture feature values of each pixel point are calculated from the filter response. Specifically, in order to extract texture features, the Gabor filter slides across the entire image with a certain size (usually a small window), and this process is called convolution. During the convolution process, each time the filter covers one pixel of the image, that is, the window size, and calculates the response of the pixel area, and finally stores the calculation result at the corresponding pixel position. The method for obtaining the texture feature values by the Gabor filter belongs to existing and mature technologies, so no elaboration will be made here;

[0053] Select a fixed scale and direction from different scales and directions as the target scale direction;

[0054] Obtain the texture feature values corresponding to each pixel point in each distal radius fracture image within the target scale direction, calculate the mean and variance of the texture feature values in each distal radius fracture image, and use them as the calculated values corresponding to each distal radius fracture image within the target scale direction;

[0055] It should be noted that the calculation methods for obtaining the mean and variance are both existing and mature calculation methods, so no further elaboration will be made here;

[0056] The overall intensity or brightness of the texture is reflected by the mean, and the complexity or roughness of the texture is reflected by the variance;

[0057] Using the same analysis method, the calculated values Tnij (Anij, Bnij) corresponding to each distal radius fracture image at different scales and directions can be obtained, where Anij is the mean value of the texture feature values corresponding to each distal radius fracture image at different scales and directions, Bnij represents the variance of the texture feature values corresponding to each distal radius fracture image at different scales and directions, n represents different distal radius fracture images, and n is a positive integer, i represents different direction labels, i = 1, 2, 3, and 4, j represents different scales, and j = 1, 2, and 3;

[0058] The normalization processing unit normalizes both the mean value Anij and the variance Bnij in the calculated values corresponding to each distal radius fracture image at different scales and directions, and then the calibrated values of the calculated values corresponding to each distal radius fracture image at different scales and directions can be obtained;

[0059] The specific method is: through the formula: XAnij = (Anij - Anij min ) / (Anij max - Anij min ) ; XBnij = (Bnij - Bnij min ) / (Bnij max - Bnij min ), the calibrated values XTnij (XAnij, XBnij) of the calculated values corresponding to each distal radius fracture image at different scales and directions are calculated;

[0060] where Anij max and Anij min are respectively the maximum and minimum values in Anij, and Bnij max and Bnij min are respectively the maximum and minimum values in Bnij;

[0061] Normalizing the numerical values scales the calculated values to the range of [0, 1], which is convenient for the later analysis of the calculated values;

[0062] The 24 - dimensional vector form acquisition unit vectorizes the calculated values of each distal radius fracture image to obtain the 24 - dimensional vector form of each distal radius fracture image;

[0063] According to the calibration values XTnij (XAnij, XBnij) of the calculated values corresponding to each distal radius fracture image at different scales and directions, the calculated values of each distal radius fracture image are vectorized to obtain the 24-dimensional vector form Ln (XAn11, XBn11, XAn12, XBn12, XAn13, XBn13,..., XAn43, XBn43) of each distal radius fracture image. Each vector form contains 24-dimensional numerical values, that is, 24 = 4×3×2. Then the vector form of the distal radius fracture image is a 24-dimensional vector form.

[0064] By preprocessing, texture feature extraction and analysis of the distal radius fracture images, the key information in the images is effectively extracted and converted into a quantifiable 24-dimensional vector form. This not only improves the data processing efficiency but also provides accurate feature input for subsequent classification analysis, enabling better capture of the subtle differences between fracture types and significantly enhancing the accuracy of imaging diagnosis.

[0065] The 24-dimensional bounding box acquisition module analyzes each dimension value in the 24-dimensional vector form corresponding to each distal radius fracture image under different fracture types, and obtains the 24-dimensional bounding boxes corresponding to different fracture types according to the analysis results. The specific method is as follows:

[0066] Obtain the vector forms of multiple distal radius fracture images corresponding to fracture type A, denoted as Le, where e represents different distal radius fracture images of fracture type A.

[0067] Obtain the mean value of the maximum and minimum values of each dimension value in the 24-dimensional vector forms of fracture type A, denoted as the reference value of each dimension. Then add the absolute value of the difference between the reference value of each dimension and the maximum and minimum values in the corresponding dimension value as the upper bound value corresponding to each dimension of fracture type A. Subtract the absolute value of the difference between the reference value of each dimension and the maximum and minimum values in the corresponding dimension value from the reference value of each dimension as the lower bound value corresponding to each dimension of fracture type A.

[0068] Taking an example for illustration, mark the maximum and minimum values of each dimension value in the 24-dimensional vector forms of fracture type A as KAr and KBr respectively, where r represents different dimensions, r = 1, 2,..., 24. Calculate the upper bound value WAr and lower bound value WBr corresponding to each dimension of fracture type A through WAr = (KAr + KBr) / 2 + |KAr - KBr| and WBr = (KAr + KBr) / 2 - |KAr - KBr|.

[0069] By using the upper bound value WAr and the lower bound value WBr corresponding to each dimension of the type A fracture, the boundary interval QAr[WBr, WAr] corresponding to each dimension of the type A fracture is established. According to the boundary intervals corresponding to each dimension of the type A fracture, the 24-dimensional boundary box HA corresponding to the type A fracture is obtained. This boundary box contains all the characteristic numerical ranges of the type A fracture images in the 24-dimensional space;

[0070] After analyzing the vector morphologies of multiple distal radius fracture images corresponding to the type B and type C fractures in the same analysis manner, the 24-dimensional boundary boxes HB and HC corresponding to the type B and type C fractures are obtained respectively;

[0071] This module analyzes the 24-dimensional vector morphologies of different fracture types, obtains the 24-dimensional boundary boxes corresponding to each fracture type, provides clear boundary intervals, provides standardized and structured data support for the fracture type classification of the system, effectively avoids ambiguity and errors, improves the reliability of classification judgment, and further improves the diagnostic accuracy of the system by setting clear boundary intervals, ensuring the efficiency and accuracy of the diagnostic process.

[0072] The fuzzy processing module performs fuzzy processing on the boundary intervals of each dimension of the 24-dimensional boundary boxes of different fracture types to obtain the fuzzy intervals corresponding to the boundary intervals of each dimension of the 24-dimensional boundary boxes of different fracture types. The specific method is as follows:

[0073] Obtain the standard deviation Zr corresponding to each dimension value in the vector morphology Le of the type A fracture, where r represents different dimensions, r = 1, 2,..., 24;

[0074] Take the ratio between the standard deviation Zr of each dimension and the reference value as the buffer coefficient Cr of each dimension for expanding the boundary interval, and take the product between the buffer coefficient Cr of each dimension and the standard deviation Zr as the buffer value Fr of the boundary interval of each dimension;

[0075] Take the sum of the upper bound value WAr corresponding to each dimension boundary interval of the type A fracture and the buffer value Fr as the buffer upper limit value VAr corresponding to each dimension boundary interval of the type A fracture, and take the absolute value of the difference between the lower bound value WBr corresponding to each dimension boundary interval of the type A fracture and the buffer value Fr as the buffer lower limit value VBr corresponding to each dimension boundary interval of the type A fracture. Furthermore, obtain the fuzzy interval MHr[VBr, VAr] corresponding to each dimension boundary interval of the type A fracture, providing a basis for fuzzy judgment in fracture type judgment;

[0076] Use the same analysis method to perform fuzzy processing on the boundary intervals of each dimension of the 24-dimensional bounding boxes for fracture types B and C, and obtain the fuzzy intervals corresponding to the boundary intervals of each dimension of the 24-dimensional bounding boxes for fracture types B and C;

[0077] By performing fuzzy processing on the boundary intervals of the 24-dimensional bounding box, the system is allowed to be more flexible and fault-tolerant when processing image data, taking into account the possible similarities and variabilities between different fracture types. Through the introduction of fuzzy intervals, the system can make reasonable judgments on complex or ambiguous diagnostic images, avoiding the rigid rule limitations of traditional models, improving the adaptability and robustness of the system when facing complex cases. At the same time, by expanding the boundary intervals, a wider reference range is provided for the judgment of fracture types, reducing the risk of misjudgment caused by eigenvalue proximity to the boundary. The existence of fuzzy intervals enables the system to more comprehensively consider the distribution of eigenvalues, thereby improving the accuracy of diagnosis.

[0078] The fracture type judgment module obtains the real-time 24-dimensional vector form of the diagnostic image to be detected, and compares and analyzes the numerical values of each dimension in the real-time vector form with the boundary intervals and fuzzy intervals of each dimension of the 24-dimensional bounding boxes of different fracture types. According to the analysis results, it judges the fracture type corresponding to the diagnostic image to be detected. The specific method is as follows:

[0079] Obtain the number of numerical values of each dimension in the real-time 24-dimensional vector form of the diagnostic image to be detected that are respectively located in the boundary intervals and fuzzy intervals of each dimension of the 24-dimensional bounding boxes of different fracture types, and mark them as AS1, AS2, BS1, BS2, CS1, and CS2 respectively;

[0080] AS1, BS1, and CS1 refer to the number of numerical values of each dimension in the real-time 24-dimensional vector form of the diagnostic image to be detected that are respectively located in the boundary intervals of each dimension of the 24-dimensional bounding boxes of different fracture types, and AS2, BS2, and CS2 refer to the number of numerical values of each dimension in the real-time 24-dimensional vector form of the detected diagnostic image that are respectively located in the fuzzy intervals of each dimension of the 24-dimensional bounding boxes of different fracture types;

[0081] It should be noted that when judging the numerical values of each dimension in the real-time 24-dimensional vector form of the image to be detected and diagnosed, first judge whether the numerical value of the dimension is within the boundary interval of each dimension of the 24-dimensional boundary box of different fracture types. If it is, it is judged that the corresponding dimension value belongs to the boundary interval of the corresponding fracture type, and the matching quantity of the boundary interval of the corresponding fracture type is incremented by 1. For the dimension values that are not within the boundary interval of each dimension of the 24-dimensional boundary box of different fracture types, judge whether they are within the fuzzy interval of each dimension of the 24-dimensional boundary box of each fracture type. If it is, it is judged that the corresponding dimension value belongs to the fuzzy interval of the corresponding fracture type, and the matching quantity of the fuzzy interval of the corresponding fracture type is incremented by 1. Otherwise, no processing is performed, and the quantities of the dimension values in the real-time 24-dimensional vector form of the image to be detected and diagnosed that are respectively within the boundary interval and the fuzzy interval of each dimension of the 24-dimensional boundary box of different fracture types are obtained;

[0082] Obtain the sum of the products of the quantities of the dimension values in the real-time 24-dimensional vector form of the image to be detected and diagnosed that are respectively within the boundary interval and the fuzzy interval of each dimension of the 24-dimensional boundary box of different fracture types and the preset coefficients β1 and β2, and use it as the selection coefficient corresponding to the image to be detected and diagnosed in the 24-dimensional boundary box of different fracture types respectively. Take the fracture type corresponding to the maximum selection coefficient as the fracture type of the image to be detected and diagnosed. The specific numerical values of the preset coefficients β1 and β2 are determined by relevant personnel according to actual needs. Here, 1 = β1 + β2, satisfying 1 > β1 > 0.5 > β2 > 0;

[0083] Through the formulas: XZA = AS1×β1 + AS2×β2, XZB = BS1×β1 + BS2×β2, and XZC = CS1×β1 + CS2×β2, calculate the selection coefficients XZA, XZB, and XZC corresponding to the image to be detected and diagnosed in the 24-dimensional boundary box of different fracture types respectively;

[0084] By analyzing the 24-dimensional vector form of the image to be detected in real time and comparing it with the preset 24-dimensional boundary box and fuzzy interval, it is possible to automatically and accurately judge the fracture type corresponding to the image to be detected. It integrates a comprehensive knowledge graph and multi-dimensional analysis, greatly improving the intelligence and automation level of fracture type classification, reducing manual intervention, and enhancing the diagnostic speed and accuracy. This module can also further compare the quantities of dimension values when the selection coefficients are the same to ensure the most appropriate diagnosis in the case of multiple selections. It can greatly improve the accuracy and efficiency of image analysis, reduce the errors of manual operations, and enhance the fault tolerance ability of the system through fuzzy processing. In the classification of fracture types, the system can accurately classify according to the 24-dimensional vector form and the boundary box.

[0085] Embodiment 2: As Embodiment 2 of the present invention, when this application is specifically implemented, compared with Embodiment 1, the difference in the technical solution of this embodiment from that of Embodiment 1 is only that in this embodiment, analysis is performed when the selection coefficients for more than one fracture type are the largest and the number of values in each dimension of the real-time 24-dimensional vector form of the image to be detected and diagnosed located in the boundary intervals and fuzzy intervals of each dimension of the 24-dimensional boundary boxes of different fracture types is 0;

[0086] When the selection coefficients for more than one fracture type are the largest, the number of values in each dimension of the 24-dimensional vector form of the image to be detected and diagnosed located in the boundary intervals of the 24-dimensional boundary boxes with the same selection coefficient is obtained, and the fracture type with the largest number is taken as the fracture type corresponding to the image to be detected and diagnosed. If there are still the same numbers, the fracture types corresponding to the 24-dimensional boundary boxes with the same number are output simultaneously;

[0087] When the number of values in each dimension of the real-time 24-dimensional vector form of the image to be detected and diagnosed located in the boundary intervals and fuzzy intervals of each dimension of the 24-dimensional boundary boxes of different fracture types is 0, an artificial review process is triggered.

[0088] When the selection coefficients of multiple fracture types are the same, by further comparing the number of dimensions in the boundary intervals to determine the fracture type, it can effectively handle the diagnostic problems in complex situations. This multi-level judgment logic enables the system to make more reasonable decisions when facing complex situations, improving the robustness and reliability of the system. When the eigenvalue of the image to be detected cannot match the boundary interval or fuzzy interval of any fracture type, the artificial review process can be triggered to ensure the final accuracy of the diagnosis. By introducing artificial intervention when necessary, the reliability of the diagnostic result is ensured.

[0089] In the case of multiple identical selection coefficients, the principle of giving priority to the quantity is adopted to further improve the diagnostic accuracy. In addition, the system also has automation and real-time performance, greatly reducing the waiting time and the need for artificial intervention, and providing reliable decision support for clinicians.

[0090] Embodiment 3: As Embodiment 3 of the present invention, when this application is specifically implemented, compared with Embodiment 1 and Embodiment 2, the technical solution of this embodiment is to combine and implement the solutions of the above Embodiment 1 and Embodiment 2.

[0091] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to obtain a formula closest to the real situation. The preset parameters and threshold selection in the formulas are set by those skilled in the art according to the actual situation.

[0092] As described above, it is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims described above.

Claims

1. The image diagnosis decision-making support system based on medical knowledge graph is characterized by: include: An image data set building module, which builds an image data set through multiple distal radius fracture images; The image data set analysis module extracts and analyzes the texture information corresponding to each pixel point in each preprocessed distal radius fracture image to obtain the 24-dimensional vector form of each distal radius fracture image; The 24-dimensional bounding box acquisition module obtains the vector form of multiple distal radius fracture images corresponding to the type A fracture as Le, where e refers to different distal radius fracture images of the type A fracture. The upper bound value WAr and the lower bound value WBr corresponding to each dimension of the type A fracture are calculated by WAr=(KAr+KBr) / 2+|KAr-KBr| and WBr=(KAr+KBr) / 2-|KAr-KBr|, where KAr and KBr are the maximum and minimum values ​​of each dimension in the 24-dimensional vector form of the type A fracture, respectively. The type A fracture is obtained by The upper bound value WAr and the lower bound value WBr corresponding to each dimension are used to establish the boundary interval QAr[WBr,WAr] ​​corresponding to each dimension of the type A fracture. According to the boundary interval corresponding to each dimension of the type A fracture, the 24-dimensional bounding box HA corresponding to the type A fracture is obtained, where r refers to different dimensions, r=1, 2, ..., 24. The same analysis method is used to analyze the vector morphology of multiple distal radius fracture images corresponding to the types B and C fractures, and the 24-dimensional bounding boxes corresponding to the types B and C fractures are obtained. The fracture types include types A, B and C fractures. The fuzzy processing module obtains the standard deviation Zr corresponding to each dimension value in each 24-dimensional vector form of type A fracture, through the formula: , calculate and obtain the buffer value Fr of the boundary interval of each dimension; At the same time, the formulas: VAr=WAr+Fr and VBr=|WBr-Fr| are used to obtain the buffer upper limit VAr and buffer lower limit VBr corresponding to the boundary intervals of each dimension of the A fracture type, and then the fuzzy intervals MHr[VBr,VAr] corresponding to the boundary intervals of each dimension of the A fracture type are obtained. The same analysis method is used to obtain the fuzzy intervals corresponding to the boundary intervals of each dimension of the 24-dimensional bounding box of the B and C fracture types. The fracture type judgment module compares and analyzes the numerical values ​​of each dimension in the real-time 24-dimensional vector form of the diagnostic image to be detected with the boundary intervals and fuzzy intervals of each dimension of the 24-dimensional bounding box of different fracture types to determine the fracture type corresponding to the diagnostic image to be detected.

2. The image diagnosis decision support system based on medical knowledge graph according to claim 1 is characterized in that: The specific method of obtaining the 24-dimensional vector form of each distal radius fracture image is as follows: Firstly, the texture feature values ​​corresponding to each pixel point in each distal radius fracture image at different scales and directions are obtained by a feature value extraction unit according to the texture information of each distal radius fracture image at different scales and directions. Then, the texture feature values ​​corresponding to each pixel point in each distal radius fracture image at different scales and directions are analyzed by a feature value analysis unit. According to the analysis results, the calculated values ​​corresponding to each distal radius fracture image at different scales and directions are obtained. Then, the mean and variance of the calculated values ​​corresponding to each distal radius fracture image at different scales and directions are normalized by a normalization processing unit to obtain the calibration values ​​of the calculated values ​​corresponding to each distal radius fracture image at different scales and directions. Finally, the calculated values ​​of each distal radius fracture image are vectorized by a 24-dimensional vector morphology acquisition unit to obtain the 24-dimensional vector morphology of each distal radius fracture image. Here, the number of directions is selected as 4, namely 0°, 45°, 90°, and 135°, and the different scales are 1, 2, and 3, respectively.

3. The image diagnosis decision support system based on medical knowledge graph according to claim 2 is characterized in that: The specific method for obtaining the calculated values ​​corresponding to each distal radius fracture image at different scales and directions is as follows: A fixed scale and direction is selected from different scales and directions as the target scale direction, and the texture feature values ​​corresponding to each pixel point in each distal radius fracture image in the target scale direction are obtained, and the mean and variance of the texture feature values ​​in each distal radius fracture image are used as the corresponding calculation values ​​of each distal radius fracture image in the target scale direction; The same analysis method was used to obtain the calculated values ​​Tnij (Anij, Bnij) corresponding to each distal radius fracture image at different scales and directions, where Anij is the mean value of the texture feature values ​​corresponding to each distal radius fracture image at different scales and directions, Bnij refers to the variance of the texture feature values ​​corresponding to each distal radius fracture image at different scales and directions, n refers to different distal radius fracture images, i refers to different direction numbers, i=1, 2, 3 and 4, and j refers to different scales, j=1, 2 and 3.

4. The image diagnosis decision support system based on medical knowledge graph according to claim 3 is characterized in that: The specific method for obtaining the calibration values ​​of each distal radius fracture image corresponding to the calculated values ​​at different scales and directions is as follows: By formula: XAnij=(Anij-Anij min ) / (Anij max -Anij min );XBnij=(Bnij-Bnij min ) / (Bnij max -Bnij min ), calculate the calibration value XTNij (XAnij, XBnij) of the calculated value corresponding to each distal radius fracture image at different scales and directions, where Anij max and Anij min are the maximum and minimum values ​​in Anij, Bnij max and Bnij min are the maximum and minimum values ​​in Bnij respectively.

5. The image diagnosis decision support system based on medical knowledge graph according to claim 4 is characterized in that: The specific method of obtaining the 24-dimensional vector form of each distal radius fracture image is as follows: According to the calibration values ​​XTNij (XAnij, XBnij) of the calculated values ​​corresponding to each distal radius fracture image in different scales and directions, the calculated values ​​of each distal radius fracture image are vectorized to obtain the 24-dimensional vector morphology Ln (XAn11, XBn11, XAn12, XBn12, XAn13, XBn13, …, XAn43, XBn43) of each distal radius fracture image. Each vector morphology contains 24 dimensional values.

6. The image diagnosis decision support system based on medical knowledge graph according to claim 5 is characterized in that: The specific method for determining the fracture type corresponding to the diagnostic image to be detected is: The number of values ​​of each dimension in the real-time 24-dimensional vector form of the diagnostic image to be detected that are located in the boundary interval and fuzzy interval of each dimension of the 24-dimensional bounding box of different fracture types is obtained, and the sum of the products between them and the preset coefficients β1 and β2 is used as the selection coefficient corresponding to the 24-dimensional bounding box of the diagnostic image to be detected in different fracture types, and the fracture type corresponding to the maximum selection coefficient is used as the fracture type of the diagnostic image to be detected. Here, 1=β1+β2, satisfying 1>β1>0.5>β2>0.

7. The image diagnosis decision support system based on medical knowledge graph according to claim 6 is characterized in that: When the selection coefficients of multiple fracture types are the same and the largest, the number of values ​​of each dimension in the 24-dimensional vector form of the diagnostic image to be detected that are respectively located in the boundary interval of the 24-dimensional bounding box with the same selection coefficient is obtained, and the fracture type with the largest number is taken as the fracture type corresponding to the diagnostic image to be detected. If the number is still the same, the corresponding fracture types are output at the same time; When the number of values ​​of each dimension in the real-time 24-dimensional vector form of the diagnostic image to be detected that are located in the boundary interval and fuzzy interval of each dimension of the 24-dimensional bounding box of different fracture types is 0, the manual review procedure is triggered.

8. The image diagnosis decision support system based on medical knowledge graph according to claim 6 is characterized in that: When judging the values ​​of each dimension in the real-time 24-dimensional vector form of the diagnostic image to be detected, first determine whether its dimension value is within the boundary interval of each dimension of the 24-dimensional bounding box of different fracture types. If it is, the corresponding dimension value is judged to belong to the boundary interval of the corresponding fracture type, and the number of boundary interval matches of the corresponding fracture type is increased by 1. If it is not, determine whether it is within the fuzzy interval of each dimension of the 24-dimensional bounding box of each fracture type. If it is, the corresponding dimension value is judged to belong to the fuzzy interval of the corresponding fracture type, and the number of fuzzy interval matches of the corresponding fracture type is increased by 1. Otherwise, no processing is performed.

Images