Orthopedic auxiliary examination system based on image processing

Abstract

The application relates to the technical field of medical image processing and computer-aided diagnosis, in particular to an orthopedic auxiliary examination system based on image processing; the system comprises a tensor field mapping, a texture flow construction, diffusion enhancement, singularity analysis and a quantization unit; the system uses a structure tensor to construct a texture flow field to suppress background noise and metal artifacts; the core is to use a diffusion equation evolution flow field to amplify texture dislocation, identify topological singular points serving as fracture endpoints by calculating Poincare indexes, and reconstruct a lesion boundary by using a geodesic algorithm; the application solves the problem of microtexture loss under low bone density or metal interference, and realizes high-signal-to-noise ratio lesion accurate positioning and healing quantization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing and computer-aided diagnosis, in particular to an orthopedic auxiliary examination system based on image processing. BACKGROUND

[0002] With the development of fine orthopedic diagnosis and treatment technology, medical image analysis plays an increasingly important role in fracture determination and rehabilitation assessment; at present, orthopedic clinics usually rely on subjective interpretation of two-dimensional gray-scale images by physicians or auxiliary examination with conventional edge detection algorithms based on gradient steps and pixel gray-scale differences; however, the existing technology has significant limitations in processing complex images with low bone density or metal implant artifact interference; due to the excessive dependence of traditional algorithms on brightness changes, small fracture lines and hidden cracks are easily submerged in background noise, and conventional noise reduction processing often leads to loss of fine texture details; in addition, the assessment of callus healing status has long been in the qualitative stage of visual experience judgment, and there is a lack of objective physical indicators that can accurately represent the order degree of bone trabecular texture; therefore, how to accurately identify hidden fractures in a low signal-to-noise ratio and strong interference environment, and objectively quantify the callus healing degree, has become a technical problem that needs to be solved in the field. SUMMARY

[0003] To solve the above technical problems, the present application provides an orthopedic auxiliary examination system based on image processing, specifically, the technical scheme of the present application includes:

[0004] An anisotropic tensor field mapping unit is configured to receive original input two-dimensional medical gray-scale image data, perform domain transformation based on structure tensor construction theory, generate a structure tensor field describing the coherence and dominant direction of a local area, and suppress background noise at the mathematical level;

[0005] A texture flow vector construction unit is configured to receive the structure tensor field and perform feature decomposition, converting the tensor field into a texture flow vector field forced to be parallel to the texture direction of bone trabeculae;

[0006] A flow field chaotic diffusion enhancement unit is configured to receive the texture flow vector field, perform anisotropic evolution operation based on curvature driving using an embedded dynamic diffusion equation, magnify small texture misalignments into discontinuous points where the flow field direction changes abruptly in iterative evolution, and output an evolved flow field;

[0007] A topological singularity analysis unit is configured to calculate Poincare indices by traversing the evolved flow field, and locate topological singularities corresponding to fracture endpoints or stress concentration points by identifying points with non-zero indices;

[0008] A lesion reconstruction and quantification unit is configured to reconstruct lesion boundaries using geodesic algorithm with topological singularities as nodes, and calculate consistency entropy of the flow field to generate a quantitative healing degree score.

[0009] Preferably, the specific logic of the anisotropic tensor field mapping unit for generating the structure tensor field is:

[0010] The outer product of the input image gradient is calculated, and the outer product is convolved and integrated using a preset integral scale kernel, so as to obtain a structure tensor field containing matrix data;

[0011] Wherein, the coherence is determined by the ratio of the eigenvalues of the tensor, and the logic based on the noise performance of isotropy, i.e. the coherence tends to 0, filters out the background noise; at the same time, for the case of metal implant edge also showing high coherence interference, the unit introduces a texture frequency domain scale screening mechanism, only retains the anisotropy signal in the spatial frequency range corresponding to 3 to 10 pixel width of the bone trabecula, and eliminates the high gradient metal artifact signal, so as to ensure that the subsequent unit provides high signal-to-noise ratio and truly reflects the direction data of bone tissue texture.

[0012] Preferably, the transformation process of the texture flow vector construction unit is as follows:

[0013] The tensor of each pixel point in the structure tensor field is subjected to eigenvalue decomposition, the eigenvector corresponding to the minimum eigenvalue is extracted, the L2 norm normalization processing is performed on the eigenvector to obtain a normalized eigenvector, and the direction of the normalized eigenvector is defined as the direction of the texture flow vector, so as to construct the texture flow vector field as the evolution basis.

[0014] Preferably, the dynamic diffusion equation built in the flow field chaotic diffusion enhancement unit is defined as: the partial derivative of the flow field vector with respect to time is equal to the divergence of the product of the diffusion coefficient and the gradient of the flow field vector; the equation is based on the nonlinear anisotropic diffusion theory, and uses the curvature feedback mechanism to simulate the self-organizing evolution process of the flow field, rather than chaotic motion;

[0015] Wherein, the diffusion coefficient is constructed as a monotonically decreasing function of the local curvature mode of the flow field line, and the local curvature mode is used to represent the degree of order of the arrangement of the bone trabecula.

[0016] Preferably, the anisotropic evolution mechanism of the flow field chaotic diffusion enhancement unit includes:

[0017] When the local area flow field curvature is small, i.e. the bone trabecula arrangement is ordered, the diffusion coefficient tends to 1, so as to perform strong smoothing to make the flow field tend to be consistent;

[0018] When there is texture disorder in the local area flow field, resulting in high curvature characteristics, the diffusion coefficient is sharply attenuated to approach 0, so as to block the smoothing diffusion at the lesion, and force the small texture disorder to be relatively squeezed and enlarged in the smoothing process of the surrounding area.

[0019] Preferably, the process that the topological singularity resolving unit identifies the topological singularity is specifically:

[0020] The total number of rotations of the vector field around the current pixel point is calculated to obtain the Poincare index;

[0021] Points with a Poincare index of +1 are marked as vortex points, and points with a Poincare index of -1 are marked as saddle points, and the vortex points and saddle points are collected into a discrete singularity coordinate set.

[0022] Preferably, the process that the lesion reconstruction and quantification unit reconstructs the lesion boundary is specifically:

[0023] The identified singularity is used as a graph theory node, and a geodesic algorithm is used to search for the minimum energy path connecting the singularities;

[0024] The path energy function in the search process is defined as the sum of the inverse of the local consistency of the flow field and a constant to prevent the denominator from being zero; the local consistency of the flow field is specifically defined as the weighted average of the dot product of the texture flow vector of the current node and the texture flow vectors of its eight neighborhood nodes, which quantifies the continuity of the texture direction in the local area; the lower the flow field consistency, the smaller the energy cost of the corresponding region, so that the trajectory with the most disordered texture along the way is automatically locked as the actual boundary of the implicit fracture line or the lesion.

[0025] Preferably, the healing degree score generated by the lesion reconstruction and quantification unit is specifically the consistency entropy of the flow field, and its physical meaning is defined as follows:

[0026] The consistency entropy is used to represent the ordered degree of the flow field vector in the region of interest; the lower the entropy value, the more regular the arrangement of the trabeculae in the callus, and the higher the healing degree; the higher the entropy value, the more disordered the callus is in a cloudy state.

[0027] Compared with the prior art, the present application has the following advantages:

[0028] 1、The system automatically suppresses quantum noise at the mathematical level by constructing an anisotropic structure tensor field, taking advantage of the physical difference that noise is isotropic and trabeculae are anisotropic; at the same time, combined with the texture frequency domain scale screening mechanism, it can effectively eliminate high gradient metal implant artifact signals and only retain effective signals within the spatial frequency range corresponding to 3 to 10 pixel widths; this solves the problem of microtexture loss caused by excessive reliance on denoising algorithms when dealing with low bone density or metal interference in the prior art, providing a high signal-to-noise ratio data basis for subsequent analysis;

[0029] 2、The system does not rely on the sharp change of pixel gray scale, but extracts the minimum eigenvalue eigenvector of structure tensor, constructs the texture flow vector field forced to be parallel to the trabecular texture direction, and converts the analysis dimension from the gradient domain to the flow field domain; combined with the curvature-driven flow field chaos diffusion enhancement mechanism, the system can use anisotropic evolution to force the imperceptible micro-texture dislocation to be relatively squeezed and enlarged in iteration, and finally form the topological discontinuous point of flow field direction mutation; this effectively solves the technical problem that the traditional edge detection algorithm is difficult to identify the implicit fracture line with no obvious brightness change and submerged in the background;

[0030] 3、The system introduces the topological invariant Poincare index, which can accurately identify the fracture end point or stress concentration point as a vortex point or saddle point by calculating the number of vector field rotation circles, realizes the discretization positioning of the lesion characteristics, and maintains high robustness even when the image is blurred; further using the geodesic algorithm and the path energy function constructed based on the local consistency inverse of the flow field, the system can automatically search for the minimum energy path connecting each singular point, so as to automatically lock the trajectory with the most disordered texture as the actual physical boundary of the implicit fracture line or lesion, and improve the objectivity and accuracy of diagnosis;

[0031] 4、In view of the problem that bone callus healing evaluation has long relied on subjective experience, the system proposes a physical index of flow field consistency entropy; the index can accurately represent the order degree of the flow field vector in the region of interest, and convert the arrangement regularity degree of the trabecula inside the callus into a quantifiable numerical score by calculating the entropy value; low entropy value corresponds to high healing degree of regular arrangement, and high entropy value corresponds to low healing degree of disordered cloud shape, thereby realizing the leap from qualitative screening to quantitative evaluation, and providing scientific and objective data support for the adjustment of clinical rehabilitation plan. BRIEF DESCRIPTION OF DRAWINGS

[0032] The application will be further explained below in conjunction with the drawings and embodiments:

[0033] Figure 1 is a structural diagram of the system of the application. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical scheme and advantages of the application more clear and obvious, the application will be further described in detail below in conjunction with specific embodiments.

[0035] Embodiment 1:

[0036] Please refer to Figure 1 , an orthopedic auxiliary examination system based on image processing, comprising:

[0037] An anisotropic tensor field mapping unit is configured to receive the original input two-dimensional medical grayscale image data, perform a domain transformation based on a structure tensor construction theory, and generate a structure tensor field describing the coherence and dominant direction of a local region to suppress background noise at a mathematical level;

[0038] A texture flow vector construction unit is configured to receive the structure tensor field and perform a characteristic decomposition to convert the tensor field into a texture flow vector field forced to be parallel to the texture direction of the trabecular bone;

[0039] A flow field chaotic diffusion enhancement unit is configured to receive the texture flow vector field and perform a curvature-driven anisotropic evolution operation based on a built-in dynamic diffusion equation to magnify a tiny texture misplacement into a discontinuous point where the flow field direction changes abruptly, thereby outputting an evolved flow field;

[0040] A topological singularity analysis unit is configured to calculate Poincare indices by traversing the evolved flow field and locate topological singularities corresponding to fracture endpoints or stress concentration points by identifying points with non-zero indices;

[0041] A lesion reconstruction and quantification unit is configured to reconstruct a lesion boundary using a geodesic algorithm with the topological singularities as nodes and calculate a consistency entropy of the flow field to generate a quantitative healing degree score.

[0042] In this embodiment, the image processing-based orthopedic auxiliary examination system aims to solve the technical problem that it is difficult to identify and quantify tiny fractures, hidden cracks, and early callus healing states in traditional orthopedic image diagnosis. The system starts the anisotropic tensor field mapping unit, which is the input stage of the system and is specially configured to process original two-dimensional medical grayscale image data without external preprocessing. Unlike conventional algorithms that directly process pixel grayscale values, the core logic of this unit is to complete the transformation from the pixel domain to the structure tensor domain. By analyzing the geometric features of the local region of the image, a structure tensor field containing matrix data is outputted to provide direction information with high signal-to-noise ratio for subsequent processing. The data flow is transferred to the texture flow vector construction unit, which extracts vector information representing the growth trend of the trabecular bone based on the received structure tensor field data through characteristic decomposition, thereby generating a texture flow vector field forced to be parallel to the texture direction of the trabecular bone. The flow field chaotic diffusion enhancement unit receives the texture flow vector field and performs a special evolution operation using a built-in dynamic diffusion equation. This operation is not a simple smoothing process but a curvature-driven anisotropic evolution. The purpose is to use the physical evolution mechanism of the flow field to force the tiny texture misplacement that is difficult to detect in the original image to be preserved and relatively magnified in the iteration process, and finally form a discontinuous point where the flow field direction changes abruptly, i.e., the output evolved flow field.

[0043] On this basis, the topological singularity analysis unit performs traversal analysis on the evolution flow field, and the core logic is to calculate the Poincare index of each point in the flow field; According to the principle of topology, the unit identifies special points with non-zero index, which are determined as topological singularities, which physically correspond to the breakpoints or stress concentration points of the skeletal mechanical conduction structure; The lesion reconstruction and quantification unit takes the identified topological singularities as the nodes of graph theory, and uses the geodesic algorithm to search and reconstruct the actual boundary of the lesion; The unit also calculates the consistency entropy of the flow field to generate a quantitative healing degree score, thus completing the whole process from qualitative screening to quantitative evaluation;

[0044] Through the cooperative work of the above units, the system breaks through the dependence of traditional edge detection algorithms on gradient steps and pixel gray scale differences; By mapping image information into texture flow field and performing topological analysis, the system can still accurately identify hidden fractures by capturing the topological consistency changes of the texture flow field under extremely low bone density or in the presence of metal artifact interference; The system realizes the leap from subjective visual experience to objective physical indicators, effectively solving the clinical pain points of invisible fractures and inaccurate measurement of callus healing state.

[0045] Embodiment 2:

[0046] The specific logic of the anisotropic tensor field mapping unit to generate the structure tensor field is as follows:

[0047] The outer product of the input image gradient is calculated, and a pre-set integral scale kernel is used to convolve and integrate the outer product, thereby obtaining a structure tensor field containing matrix data;

[0048] Wherein, the coherence is determined by the ratio of the eigenvalues of the tensor, and the unit filters out background noise based on the logic that noise behaves isotropically, i.e. the coherence tends to 0; At the same time, for the case of metal implant edge also showing high coherence interference, the unit introduces a texture frequency domain scale screening mechanism, only retaining the anisotropy signals within the spatial frequency range corresponding to 3 to 10 pixel widths of the bone trabecula, and eliminating the high gradient metal artifact signals, thereby ensuring that the subsequent units are provided with high signal-to-noise ratio and true reflection of the direction data of the bone tissue texture.

[0049] The specific steps are as follows: perform two-dimensional fast Fourier transform on the input local image block to convert it to the frequency domain; Apply a ring band-pass filter, the passband frequency range of the filter is Set to the spatial frequency corresponding to the width of the bone trabecula, in this embodiment, the spatial scale corresponding to this frequency range is 3 to 10 pixel widths, so as to filter out low-frequency background and high-frequency metal artifacts; Restore the image through inverse Fourier transform; The anisotropic tensor field mapping unit receives the image data restored through the above steps, and calculates the outer product of the input image gradient;

[0050] The integral scale kernel is specifically a Gaussian kernel The expression is The gradient calculation scale is defined as the standard deviation of Gaussian smoothing before calculating the image gradient The integral scale is set to 1.5 to 3 times the gradient calculation scale to ensure that the main directionality of the structure is retained while smoothing the noise.

[0051] In the specific configuration of the embodiment, the anisotropic tensor field mapping unit performs an improved structure tensor construction logic; the unit receives the original medical image data, calculates the outer product of the gradient of each pixel point in the image, and constructs an initial tensor matrix; in order to eliminate the influence of local random noise and extract stable structural features, the unit performs convolution integral operation on the gradient outer product result by using a preset integral scale kernel; the output of this process is the structure tensor field, wherein each point in the field corresponds to a structure tensor matrix; in this process, the system introduces the key parameter of coherence; the coherence is a quantitative index between 0 and 1, which is calculated from the ratio of the eigenvalues of the tensor, and the specific calculation formula is: Wherein, and are the maximum eigenvalue and the minimum eigenvalue of the structure tensor respectively; the coherence is used to represent the uniformity of the texture direction of the local region; the unit has built-in noise suppression logic: based on the physical properties, the quantum noise in medical images usually exhibits isotropy, and the calculated coherence value tends to 0; while the trabecular bone and other anatomical structures exhibit significant anisotropy, and the coherence value tends to 1; through this filtering mechanism based on the ratio of eigenvalues, the unit automatically suppresses the interference of background noise at the mathematical level, and only retains and enhances the texture direction information representing the bone structure;

[0052] Through the construction of the tensor field based on the gradient outer product integral and the coherence analysis, the embodiment can effectively suppress the inherent quantum noise of the image while retaining the fine bone texture structure; this processing method avoids the loss of small lesion details caused by traditional denoising algorithms, and provides a high-purity and high signal-to-noise ratio directional data basis for subsequent flow field evolution.

[0053] Embodiment 3:

[0054] The transformation process of the texture flow vector construction unit is specifically:

[0055] The tensor of each pixel point in the structural tensor field is subjected to eigenvalue decomposition, a feature vector corresponding to the minimum eigenvalue is extracted, the feature vector is subjected to L2 norm normalization processing to obtain a normalized feature vector, and a direction of the normalized feature vector is defined as a direction of the texture flow vector, so as to construct the texture flow vector field as the evolution basis.

[0056] In this embodiment, the texture flow vector construction unit receives the structural tensor field output by the previous unit, and the core task of the texture flow vector construction unit is to convert the mathematical tensor matrix into the flow field vector with clear physical meaning. The texture flow vector construction unit performs eigenvalue decomposition operation on the tensor matrix of each pixel point in the structural tensor field, so as to obtain the corresponding eigenvalue and feature vector. Unlike the prior art which generally tracks the direction with the most dramatic change in gray scale, the texture flow vector construction unit specially extracts the feature vector corresponding to the minimum eigenvalue. The texture flow vector construction unit clearly defines the direction of the feature vector as the direction of the texture flow vector. Since the feature vector direction corresponding to the minimum eigenvalue is orthogonal to the gradient direction in mathematical properties, the vector field generated by the processing logic is forced to be geometrically parallel to the actual texture direction of the trabeculae. The texture flow vector construction unit converges the vectors extracted from all pixel points to construct a texture flow vector field which can truly reflect the texture direction inside the bone.

[0057] By extracting the feature vector corresponding to the minimum eigenvalue to construct the flow field, the embodiment successfully converts the medical image from the gradient field to the texture flow field. This conversion makes the subsequent analysis no longer subject to the brightness change at the fracture, but directly based on the continuity of the trabeculae, thereby providing a physical basis for identifying the implicit fracture which has no obvious change in brightness but has interrupted texture.

[0058] Embodiment 4:

[0059] The dynamic diffusion equation built in the flow field chaotic diffusion enhancement unit is defined as: the partial derivative of the flow field vector with respect to time is equal to the divergence of the product of the diffusion coefficient and the gradient of the flow field vector. The equation is based on the nonlinear anisotropic diffusion theory, and uses the curvature feedback mechanism to simulate the self-organizing evolution process of the flow field, rather than chaotic motion.

[0060] The diffusion coefficient is constructed as a monotone decreasing function of the local curvature modulus of the flow field line, and the local curvature modulus is used to represent the order degree of the arrangement of the trabeculae.

[0061] The anisotropic evolution mechanism of the flow field chaotic diffusion enhancement unit specifically includes:

[0062] When the flow field curvature in the local region is small, that is, the trabeculae are arranged in order, the diffusion coefficient tends to 1, so that strong smoothing is performed to make the flow field consistent.

[0063] When the flow field in a local area has texture disorder that leads to high curvature, the diffusion coefficient decreases sharply and approaches 0, thereby blocking smooth diffusion at the lesion and forcing the small texture disorder to be relatively squeezed and amplified during the smoothing process of the surrounding area.

[0064] Wherein, diffusion coefficient Constructed as a local curvature mode of the flow field lines The monotonically decreasing function is given by the following mathematical expression: ,in, The curvature scaling constant used to determine diffusion sensitivity is set to 1.2 to 2.0 times the average local curvature modulus of all pixels in the entire flow field, serving as an adaptive threshold to distinguish between regular textured backgrounds and lesion-related disordered textures.

[0065] To implement the above process in the digital system, the flow field chaotic diffusion enhancement unit uses the finite difference method to discretize and solve the dynamic diffusion equation. The system presets a time step, and in each iteration step, it updates the diffusion coefficient using the current flow field data and calculates the flow field state at the next moment. The specific discretization iterative update formula is as follows: ,in, For the flow field at the current moment, The time step is 0.1 to 0.25. The iterative process continues until the maximum pixel change value of the flow field between the two iterations is less than the preset convergence threshold, or the number of iterations reaches the preset maximum limit. At this time, the flow field output by the system is the final evolved flow field.

[0066] To further clarify the core mechanism of the flow field chaotic diffusion enhancement unit, this embodiment describes its internal dynamic evolution logic in detail. The unit has a built-in dynamic diffusion equation, the mathematical essence of which describes the evolution law of the flow field vector with time, that is, the partial derivative of the flow field vector with respect to time is equal to the divergence of the product of the diffusion coefficient and the gradient of the flow field vector. In this equation, the diffusion coefficient is not a constant, but is designed as a controlled variable, which is a monotonically decreasing function of the local curvature mode of the flow field line.

[0067] The specific calculation logic of the local curvature modulus is as follows: For each point in the flow field, the partial derivatives of the flow field vector in the horizontal and vertical coordinate directions of the image are calculated respectively. The squares of these partial derivatives are summed, and the result is used as the flow field gradient modulus value at that point. In this embodiment, since the texture flow field has been normalized, the gradient modulus value is directly defined as the local curvature modulus, which is used to approximate the degree of geometric curvature of the flow field at that point.

[0068] Here, the local curvature modulus is a physical quantity used to characterize the degree of order of the trabecular arrangement: the smaller the curvature modulus, the straighter and more ordered the texture lines; the larger the curvature modulus, the more curved and disordered the texture lines; based on this definition, the unit executes an anisotropic evolution mechanism:

[0069] Forward smoothing logic: when the local region's flow field curvature modulus is detected to be small, the diffusion coefficient is calculated as a value close to 1; at this time, the system performs strong smoothing operations, making the flow field vector direction of the region further consistent and regular;

[0070] Blocking and enhancement logic: when the local region flow field is detected to have a small texture disorder resulting in high curvature features, the diffusion coefficient will sharply decay and tend to 0; this change is mathematically equivalent to cutting off the diffusion path at that point, thereby blocking the smoothing operation at the lesion;

[0071] This curvature-driven nonlinear processing mechanism produces a unique coupling effect: the texture of normal tissue becomes more regular due to smoothing, while the small texture disorder at the lesion is relatively squeezed and magnified due to the inability to be smoothed, against the backdrop of the surrounding regular texture; this process forces the originally difficult-to-detect small phase dislocations to become topological discontinuity points with significant mutations in flow direction after iterative evolution, thereby greatly improving the system's sensitivity to small lesions.

[0072] Embodiment 5:

[0073] The process of identifying topological singularities by the topological singularity analysis unit is as follows:

[0074] Calculate the total number of rotations of the vector field around the currently traversed pixel point to obtain the Poincare index;

[0075] Mark the point with a Poincare index of +1 as a vortex point, and mark the point with a Poincare index of -1 as a saddle point, and collect the vortex points and saddle points into a discrete singularity coordinate set.

[0076] In specific calculations, the flow field vectors of the eight-neighbor pixels of the point are extracted in counterclockwise order, the angle difference between the two adjacent vectors is calculated, all angle differences are accumulated and divided by 2π, and the obtained integer is the Poincare index;

[0077] In this embodiment, the topological singularity analysis unit locks the lesion location through strict topological algorithms; the unit traverses each point in the evolving flow field, calculates the total number of turns of the vector field around the point, and defines the weighted average value as the Poincare index; the system has a clear determination and classification logic: when the Poincare index calculation result of a point is +1, the point is marked as a vortex point; when the calculation result is -1, the point is marked as a saddle point; and the point with an index of 0 is considered as a normal field point; the unit collects all identified vortex points and saddle points to form a set of discrete singularity coordinate sets output; these singularities, in the physical sense, are not meaningless mathematical symbols, but the key positions where the trabecular texture flow field is interrupted, bifurcated or converged, i.e. the core area of the fracture end point or stress concentration;

[0078] By introducing the Poincare index, a topological invariant, this embodiment realizes the dimension reduction of the lesion characteristics; instead of being obsessed with ambiguous edge pixels, the system directly locks the dead knots of the flow field structure, thereby realizing high-precision and discrete positioning of the fracture end point, and maintaining high robustness even in the case of image blur or extremely low contrast.

[0079] Embodiment 6:

[0080] The process of reconstructing the lesion boundary by the lesion reconstruction and quantification unit is as follows:

[0081] The identified singularities are used as graph theory nodes, and the geodesic algorithm is used to search for the minimum energy path connecting the singularities;

[0082] The path energy function in the search process is defined as the sum of the inverse of the local consistency of the flow field and a constant to prevent the denominator from being zero; the local consistency of the flow field is specifically defined as the weighted average value of the dot product of the texture flow vector of the current node and the texture flow vectors of its eight neighborhood nodes, which quantifies the continuity of the texture direction in the local area; the lower the flow field consistency, the smaller the energy cost of the corresponding area, thereby automatically locking the trajectory with the most disordered texture along the way as the implicit fracture line or the actual boundary of the lesion.

[0083] The healing degree score generated by the lesion reconstruction and quantification unit is specifically the consistency entropy of the flow field, and its physical meaning is defined as follows:

[0084] The consistency entropy is used to represent the order degree of the flow field vector in the region of interest; the lower the entropy value, the more regular the arrangement of the trabeculae in the callus, and the higher the healing degree; the higher the entropy value, the more disordered the callus is in a cloudy state.

[0085] The identified singularities are used as graph theory nodes, and the fast marching algorithm is used to solve the eikonal equation to search for the minimum energy path connecting the singularities on the pixel-level grid;

[0086] The specific calculation process of the consistency entropy is: the local consistency value of the flow field of all pixel points in the region of interest is counted, the value range of the local consistency value of the flow field is linearly divided into discrete intervals, the number of pixel points falling in each interval is counted to obtain a probability distribution, the product of the probability of each interval and its natural logarithm is calculated, the sum of the products of all intervals is taken and the negative value is obtained, and the result is the consistency entropy;

[0087] This embodiment details how the lesion reconstruction and quantification unit completes the final diagnosis output;

[0088] In terms of lesion reconstruction, the unit uses a graph theory method, taking the topological singularities identified in the previous steps as nodes in the network; the geodesic algorithm is used to search for connection paths between these nodes; in order to ensure that the searched path truly reflects the fracture line, the unit constructs a special path energy function, the value of which is defined as the sum of the reciprocal of the local consistency of the flow field and a small constant; according to this mathematical definition, the more disordered the texture of an area, the smaller the energy cost calculated by it; in the process of finding the minimum energy path, the geodesic algorithm will automatically tend to follow the trajectory with the most disordered texture and the most incoherence; this trajectory is finally determined by the system as the actual physical boundary of the implicit fracture line or lesion;

[0089] In terms of quantitative scoring, the unit calculates the consistency entropy of the flow field as the healing degree score; the consistency entropy is a quantitative index based on the information entropy theory, which is specially used to represent the ordered degree of the flow field vector in the region of interest; the physical meaning of which is defined as follows in this embodiment: the lower the calculated entropy value, the more regular the arrangement of the trabeculae within the callus and the more uniform the direction, representing a higher degree of fracture healing; on the contrary, the higher the entropy value, the more disordered the direction of the flow field within the callus, still in a disordered cloud shape, representing a lower degree of healing;

[0090] Through the reverse design of the path energy function, the system can automatically delineate the fracture line that is difficult to identify with the naked eye, providing intuitive visual assistance; at the same time, through the innovative index of consistency entropy, the system first converts the maturity of the callus from the subjective adjectives of doctors to precise and comparable numerical standards, providing scientific and objective data support for the adjustment of clinical rehabilitation programs.

[0091] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. An image processing-based orthopedic auxiliary examination system, characterized in that, include: An anisotropic tensor field mapping unit is used to receive the original input two-dimensional medical grayscale image data, perform domain transformation based on the structural tensor construction theory, and generate a structural tensor field describing the coherence and dominant direction of the local region in order to suppress background noise at the mathematical level. The texture flow vector construction unit is configured to receive the structural tensor field and perform feature decomposition, transforming the tensor field into a texture flow vector field that is forced to be parallel to the texture direction of the trabecular bone. The flow field chaos diffusion enhancement unit is used to receive the texture flow vector field and perform anisotropic evolution operation based on curvature driven by the built-in dynamic diffusion equation. This amplifies the tiny texture misalignments into discontinuities where the flow field direction changes abruptly during the iterative evolution, thereby outputting the evolved flow field. The topological singularity resolution unit is configured to traverse the evolving flow field to calculate the Poincaré index and locate the topological singularity corresponding to the fracture endpoint or stress concentration point by identifying points with non-zero indices. The lesion reconstruction and quantification unit is used to reconstruct the lesion boundary using the topological singularity as a node and the geodesic algorithm, and to calculate the consistency entropy of the flow field to generate a quantified healing score.

2. The image processing-based orthopedic auxiliary examination system according to claim 1, characterized in that, The specific logic for generating the structure tensor field by the anisotropic tensor field mapping unit is as follows: The outer product of the gradient of the input image is calculated, and the outer product is convolved and integrated using a preset integral scale kernel to obtain the structure tensor field containing matrix data. The coherence is determined by the ratio of the eigenvalues ​​of the tensor. This unit filters out background noise based on the logic that the noise is isotropic, i.e., the coherence approaches 0. At the same time, for the interference situation where the edge of the metal implant also exhibits high coherence, this unit introduces a texture frequency domain scale screening mechanism, which retains only the anisotropic signals within the spatial frequency range of 3 to 10 pixels wide corresponding to the trabecular bone, and removes high-gradient metal artifact signals, thereby ensuring that subsequent units are provided with directional data with high signal-to-noise ratio and that truly reflects the texture of bone tissue.

3. The image processing-based orthopedic auxiliary examination system according to claim 1, characterized in that, The transformation process of the texture flow vector construction unit is as follows: The tensor of each pixel in the structure tensor field is decomposed into features to extract the feature vector corresponding to the minimum feature value. The feature vector is then normalized by L2 norm to obtain a normalized feature vector. The direction of the normalized feature vector is defined as the direction of the texture flow vector, thereby constructing the texture flow vector field as the basis for evolution.

4. The image processing-based orthopedic auxiliary examination system according to claim 1, characterized in that, The dynamic diffusion equation built into the flow field chaotic diffusion enhancement unit is defined as follows: the partial derivative of the flow field vector with respect to time is equal to the divergence of the product of the diffusion coefficient and the gradient of the flow field vector. This equation is based on the nonlinear anisotropic diffusion theory and uses the curvature feedback mechanism to simulate the self-organized evolution process of the flow field, rather than the disordered chaotic motion. The diffusion coefficient is constructed as a monotonically decreasing function of the local curvature mode of the flow field lines, and the local curvature mode is used to characterize the regularity of the trabecular arrangement.

5. The image processing-based orthopedic auxiliary examination system according to claim 4, characterized in that, In the anisotropic evolution mechanism of the flow field chaotic diffusion enhancement unit, the calculation formula of the diffusion coefficient is as follows: wherein, is a curvature scale constant, is a local curvature modulus; when the local curvature modulus is less than , the diffusion coefficient tends to 1, performing a smoothing operation on the flow field; when the local curvature modulus is greater than , the diffusion coefficient tends to 0, blocking the smoothing diffusion at the lesions.

6. The image processing based orthopedic auxiliary examination system according to claim 1, wherein, The process by which the topological singularity resolution unit identifies topological singularities is as follows: Calculate the total number of rotations of the vector field around the currently traversed pixel to obtain the Poincaré index; Points with Poincare index of +1 are marked as vortex points, points with Poincare index of -1 are marked as saddle points, and the vortex points and saddle points are collected into a discrete set of singular point coordinates.

7. The image processing-based orthopedic auxiliary examination system according to claim 1, characterized in that, The process of reconstructing and quantifying the lesion boundary by the lesion reconstruction and quantification unit is specifically: The identified singular points are taken as graph theory nodes, and a geodesic algorithm is used to search for the minimum energy path connecting the singular points; Wherein, the path energy function in the search process is defined as the sum of the reciprocal of the local consistency of the flow field and a constant to prevent the denominator from being zero; The local consistency of the flow field is specifically defined as the weighted average of the dot product of the texture flow vector of the current node and the texture flow vector of its eight neighborhood nodes, which quantifies the continuity of the texture direction in the local area; The lower the flow field consistency, the smaller the energy cost of the corresponding area, so as to automatically lock the trajectory with the most turbulent texture along the way as the actual boundary of the hidden fracture line or lesion.

8. The image processing based orthopedic auxiliary examination system according to claim 1, wherein, The healing degree score generated by the lesion reconstruction and quantification unit is specifically the consistency entropy of the flow field, and the calculation process is: statistics of the local consistency of the flow field of all pixel points in the region of interest, linearly divides the value range of the local consistency of the flow field into N discrete intervals, statistics of the proportion of pixel points falling in each interval to obtain the probability distribution, calculation of the product of the probability of each interval and its natural logarithm, summation of the products of all intervals and taking the negative value, and the result is the consistency entropy.

Images