A method and system for diabetic wound image analysis

Abstract

This application provides a method and system for analyzing diabetic foot lesion images, relating to the field of medical image processing and analysis technology. This application acquires a color image of the wound area of ​​a diabetic foot patient, along with infrared radiation energy distribution data aligned with the pixel coordinates in the color image. It then uses a gray-level co-occurrence matrix to perform texture distribution statistics on the color image to obtain energy gradient data. The joint probability of the texture distribution data and energy gradient data at each pixel coordinate in the color image is calculated to construct an association distribution map. Anisotropic convolution is performed on the association distribution map to extract the association gradient, resulting in a convergence boundary curve. Based on the association distribution map, the closed region enclosed by the convergence boundary curve is clustered and segmented. The area measurement is calculated based on the number of pixels in the necrotic and granulation tissue regions. This achieves accurate segmentation of the lesion boundary in diabetic foot lesions and automated quantitative identification of necrotic and granulation tissue.

Description

Technical Field

[0001] This application relates to the field of medical image processing and analysis technology, and in particular to a method and system for analyzing images of diabetic wounds. Background Technology

[0002] Diabetic foot ulcers, a serious complication of diabetes, require precise assessment for reducing amputation rates and improving prognosis. With the development of smart healthcare, the use of computer vision technology for automated quantitative analysis of wound tissue has become a core component of clinical disease monitoring and treatment planning.

[0003] In existing technologies, diabetic wound analysis typically employs color component recognition or edge detection algorithms based on RGB color images, supplemented by discretization analysis using skin surface temperature acquired by infrared thermal imagers. For example, some analysis systems attempt to distinguish necrotic tissue from granulation tissue in a two-dimensional wound by independently processing color features and thermal field changes.

[0004] However, due to interference from exudate, crusting, or ambient light at the wound edges, relying solely on visual texture or single-dimensional thermal data is insufficient to accurately capture subtle dynamic differences between lesion tissues, leading to significant boundary localization deviations and unstable tissue identification results. This deficiency in multi-source information fusion and edge extraction capability limits the accuracy of clinical assessment. Therefore, existing technologies suffer from inaccurate tissue segmentation and low reliability of quantitative identification in diabetic wounds under complex backgrounds. Summary of the Invention

[0005] The purpose of this application is to provide a method and system for analyzing images of diabetic wounds, in order to solve the technical problems of inaccurate tissue segmentation and low reliability of quantitative identification in diabetic wounds under complex backgrounds in the prior art.

[0006] In a first aspect, this application provides a method for analyzing images of diabetic wounds, including:

[0007] Acquire color images of the wound area in diabetic foot patients, and energy distribution data of infrared radiation aligned with pixel coordinates in the color images of the wound;

[0008] The texture distribution of the color image of the wound was statistically analyzed using the gray-level co-occurrence matrix to obtain texture distribution data, and the energy distribution data was normalized by gradient to obtain energy gradient data.

[0009] Calculate the joint probability of texture distribution data and energy gradient data for each pixel coordinate in the color image of the wound, and map the joint probability to the association space to construct an association distribution map;

[0010] Anisotropic convolution is performed on the correlation distribution map to extract the correlation gradient. A velocity variable is constructed based on the correlation gradient. The level set function constructed based on the preset initial curve is iteratively evolved using level set evolution to make the initial curve converge to the edge of the lesion area in the wound color image, thus obtaining the convergence boundary curve.

[0011] Based on the correlation distribution map, the closed region enclosed by the convergence boundary curve is clustered and segmented to identify the necrotic region and granulation region in the color image of the wound, and the area measurement value is calculated based on the number of pixels in the necrotic region and granulation region.

[0012] Optionally, after calculating the joint probability of texture distribution data and energy gradient data for each pixel coordinate in the wound color image and mapping the joint probability to the association space to construct an association distribution map, the method further includes:

[0013] The pixels in the color images of wounds with a joint probability greater than a preset correlation threshold in the correlation distribution map are combined to obtain the lesion pixel set.

[0014] Calculate the geometric center of the lesion pixel set, construct the minimum circumcircle surrounding all pixels in the lesion pixel set with the geometric center as the center, and use the minimum circumcircle as the preset initial curve.

[0015] Optionally, the texture distribution of the wound color image is statistically analyzed using the gray-level co-occurrence matrix to obtain texture distribution data, and the energy distribution data is gradient normalized to obtain energy gradient data, including:

[0016] Based on preset conversion coefficients, the RGB channel components of each pixel in the color image of the wound are converted to obtain a brightness distribution image;

[0017] The brightness value corresponding to each pixel in the brightness distribution image is divided into multiple brightness levels. With the preset pixel spacing and scanning angle as spatial constraints, the frequency of occurrence of the brightness level of each pair of pixels in the brightness distribution image is counted and normalized to obtain multiple co-occurrence probabilities.

[0018] Based on the co-occurrence probability, the local contrast and information entropy of the brightness distribution image are calculated, and the local contrast and information entropy are vector-combined to obtain the texture distribution data.

[0019] The difference between adjacent pixels in the energy distribution data is calculated to obtain multiple energy variations. Based on the fluctuation range of the energy variations, the energy variations are scaled proportionally to obtain energy gradient data.

[0020] Optionally, the joint probability of texture distribution data and energy gradient data at each pixel coordinate in the color image of the wound is calculated, and the joint probability is mapped to the association space to construct an association distribution map, including:

[0021] The texture distribution data and energy gradient data with the same pixel coordinates in the color image of the wound are numerically aligned to obtain multiple pairs of vectors.

[0022] The frequency of occurrence of the same vector pair in the color image of the wound is counted. An association space is constructed with texture distribution data and energy gradient data as coordinate axes, and the joint probability of each pixel coordinate in the association space is calculated based on the frequency of occurrence.

[0023] Using the pixel coordinates of the color image of the wound as a spatial index, an association distribution map is constructed based on the corresponding joint probability.

[0024] Optionally, anisotropic convolution is performed on the association distribution map to extract the association gradient, including:

[0025] Determine the evolution direction of each associated coordinate in the associated distribution map, and use anisotropic convolution kernels to perform convolution operations along the normal direction and tangent direction of the evolution direction respectively to obtain the normal direction gradient and tangent direction gradient.

[0026] Based on the magnitude ratio of the normal gradient to the tangent gradient, the local anisotropy index of each associated coordinate in the associated distribution map is calculated. Then, based on the local anisotropy index, the normal gradient and the tangent gradient are weighted and summed to obtain the association degree gradient.

[0027] Optionally, a velocity variable is constructed based on the correlation gradient, and the level set function constructed based on the preset initial curve is iteratively evolved using level set evolution, so that the initial curve converges to the edge of the lesion area in the color image of the wound, resulting in a convergence boundary curve, including:

[0028] Calculate the sign distance between each associated coordinate and the initial curve in the associated distribution map, construct a horizontal set function based on the sign distance, and use the pixel spacing between adjacent pixels in the wound color image as the spatial reference to calculate the second-order partial derivatives of each element coordinate in the horizontal and vertical directions to obtain the local curvature of each element coordinate.

[0029] The influence coefficient is calculated based on the correlation gradient and local curvature, and the velocity variable is obtained by weighted summation of the correlation gradient and local curvature based on the influence coefficient.

[0030] The level set function is updated based on the velocity variable. The intermediate evolution curve is determined based on the coordinates of the zero element in the updated elements. The offset of the intermediate evolution curve relative to the previous intermediate evolution curve is calculated.

[0031] When the offset is greater than the preset change threshold, the system returns to update each element in the level set function based on the velocity variable until the offset is less than or equal to the change threshold, and then determines the intermediate evolution curve as the convergence boundary curve.

[0032] Optionally, based on the correlation distribution map, the closed region enclosed by the convergence boundary curve is clustered and segmented to identify the necrotic and granulation regions in the color image of the wound, and the area measurement value is calculated based on the number of pixels in the necrotic and granulation regions, including:

[0033] Calculate the discrete deviation between the associated coordinates of different targets located within the convergence boundary curve in the associated distribution map, and classify the associated coordinates of the targets into the first feature cluster and the second feature cluster based on the discrete deviation;

[0034] Calculate the average value of the first feature cluster and the average value of the second feature cluster and compare them. Mark the region where the feature cluster with the smaller average value is located as the necrotic region and the region where the feature cluster with the larger average value is located as the granulation region.

[0035] The total number of first pixels in the necrotic area and the total number of second pixels in the granulation area are counted separately. The product of the total number of first pixels and the total number of second pixels with the preset single pixel area is calculated to obtain the area measurement value of the necrotic area and the area measurement value of the granulation area.

[0036] Secondly, this application provides a diabetic wound image analysis system, comprising:

[0037] The acquisition module is used to acquire color images of the wound area of ​​diabetic foot patients, as well as energy distribution data of infrared radiation aligned with the pixel coordinates in the color images of the wound.

[0038] The analysis module is used to perform texture distribution statistics on the color image of the wound using the gray-level co-occurrence matrix to obtain texture distribution data, and to perform gradient normalization on the energy distribution data to obtain energy gradient data.

[0039] The computation module is used to calculate the joint probability of texture distribution data and energy gradient data for each pixel coordinate in the color image of the wound, and to map the joint probability to the association space to construct the association distribution map;

[0040] The convolution module is used to perform anisotropic convolution on the correlation distribution map to extract the correlation gradient. Based on the correlation gradient, a velocity variable is constructed. The level set function constructed based on the preset initial curve is iteratively evolved using level set evolution, so that the initial curve converges to the edge of the lesion area in the wound color image, and the convergence boundary curve is obtained.

[0041] The identification module is used to cluster and segment the closed region enclosed by the convergence boundary curve based on the correlation distribution map, identify the necrotic region and granulation region in the color image of the wound, and calculate the area measurement value based on the number of pixels in the necrotic region and granulation region.

[0042] Thirdly, this application provides an electronic device, comprising:

[0043] Memory, used to store computer programs;

[0044] A processor is used to execute computer programs to implement the steps of a method for analyzing images of diabetic wounds as described in the first aspect above.

[0045] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the diabetic wound image analysis method described in the first aspect above.

[0046] This application provides a method for analyzing diabetic wound images. By acquiring color images of the wound and aligned energy distribution data, it ensures absolute spatial consistency between visual texture information and infrared thermal energy information. It suppresses interference from raw data caused by uneven ambient light or individual patient base temperature differences, improving the quality of feature representation. It achieves deep fusion of visual and thermal features, transforming previously isolated physical quantities into correlated feature fields characterizing tissue properties. This solves the problem of insufficient multi-source information fusion in existing technologies, shields against pseudo-edge interference caused by noise such as exudate and crusting, and drives the curve to adaptively converge to the true boundary, reducing boundary localization bias. It eliminates the subjective errors of manual assessment, providing highly reliable decision support for clinical evaluation.

[0047] Furthermore, this application constructs multiple vector pairs by numerically aligning texture distribution data and energy gradient data with the same pixel coordinates in the color image of the wound. An association space is then constructed using the texture distribution data and energy gradient data as coordinate axes. The joint probability of each pixel coordinate in the association space is calculated by statistically analyzing the frequency of occurrence of the same vector pair across the entire image. Finally, using the pixel coordinates of the color image of the wound as the spatial index, the joint probability is back-mapped to the image space to construct an association distribution map. This solves the problem of unstable recognition caused by poor feature capture in existing schemes in complex backgrounds. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 A flowchart illustrating a method for analyzing images of diabetic wounds provided in an embodiment of this application;

[0050] Figure 2 A flowchart illustrating a method for constructing an association distribution map provided in an embodiment of this application;

[0051] Figure 3 A flowchart illustrating a method for obtaining a convergence boundary curve provided in an embodiment of this application;

[0052] Figure 4 This is a schematic diagram of the structure of a diabetic wound image analysis system provided in an embodiment of this application;

[0053] Figure 5 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0054] To address the technical barriers in the environment of diabetic foot ulcer wounds, such as interference from exudate, scab obstruction, and uneven lighting, traditional RGB color recognition or independent thermal field analysis is unable to capture subtle dynamic differences between tissues, resulting in large boundary positioning deviations and unstable recognition results.

[0055] This application extracts highly robust low-level features through texture distribution statistics and gradient normalization. The core logic lies in mapping heterogeneous texture and energy information to a unified association space by calculating joint probabilities, constructing an association distribution map that can characterize the essential attributes of tissue, thereby eliminating logical gaps caused by insufficient fusion of multi-source information. Subsequently, to address the problem of weak edge extraction capability in complex backgrounds, this application utilizes anisotropic convolution to extract the association degree gradient with directional adaptability in the association distribution map, and uses this to drive the level set function to converge towards the true edge of the lesion, thereby obtaining a high-precision convergence boundary curve. Finally, based on the multidimensional features contained in the association distribution map, clustering and segmentation are performed in closed regions to achieve automated discrimination and accurate area measurement of necrotic and granulation areas, solving the problems of inaccurate wound tissue segmentation and low reliability of quantitative identification in existing technologies in complex backgrounds.

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] The core of this application is to provide a method for analyzing images of diabetic wounds, and a flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes:

[0058] Step 101: Obtain a color image of the wound area of ​​the diabetic foot patient, and energy distribution data of infrared radiation aligned with the pixel coordinates in the color image of the wound.

[0059] In this step, diabetic foot patients refer to subjects with diabetes and accompanying distal lower extremity nerve damage or peripheral vascular disease. The wound area refers to the site of skin damage and its surrounding tissues where the lesion is located, which may include ulcers, necrotic tissue, granulation tissue, and surrounding inflammatory infiltration areas. Wound color images refer to image data acquired using a visible light sensing unit that records the appearance, color, shape, and texture information of the lesion. Infrared radiation refers to electromagnetic waves emitted outward by an object due to internal thermal motion; in the context of wound monitoring, it specifically refers to the far-infrared frequency energy generated by the metabolic heat of the lesion tissue. Energy distribution data refers to the set of numerical values ​​describing the intensity of radiation at various points within the wound area.

[0060] In this embodiment, a visible light sensor is first used to image the lesion site of a diabetic foot patient, acquiring a color image of the wound that records the external contour information of the wound. Simultaneously, an infrared probe captures the infrared radiation emitted from the wound area, thereby obtaining energy distribution data that records changes in the thermal field of the lesion. The data acquired in this step includes a color pixel matrix with red, green, and blue channel information. and the numerical matrix reflecting the intensity of thermal radiation. Then, a spatial transformation model was used to transform the numerical matrix. The coordinate system is mapped to the color pixel matrix. In the physical space, the two types of data are aligned at the pixel level.

[0061] For example, in a diagnostic scenario at Hospital A, a set of color images of the wound on patient B's heel were acquired using a model A acquisition device. and the corresponding energy distribution data Color image The data showed obvious redness and swelling at the wound edges, while the energy distribution data... This shows that the thermal radiation intensity of the swollen area is significantly higher than that of the surrounding normal skin tissue. Finally, the data is calculated... Plane coordinates and data The coordinates of the sampling points are merged to obtain a composite dataset that includes not only color and texture information but also heat intensity.

[0062] Step 102: Use the gray-level co-occurrence matrix to perform texture distribution statistics on the color image of the wound to obtain texture distribution data, and perform gradient normalization on the energy distribution data to obtain energy gradient data.

[0063] In this step, the gray-level co-occurrence matrix (GLCM) refers to a mathematical statistical matrix that describes texture features by the frequency of occurrence of pixel gray-level pairs under specific spatial relationships in a statistical image. Texture distribution data refers to a set of numerical values ​​reflecting the strength of local detail changes on the wound surface, and may include feature parameters representing local contrast or energy distribution states. Energy gradient data refers to a numerical matrix obtained by spatially differencing the heat distribution and scaling it according to the range of variation; it can reflect the dynamic trend of temperature changes in the wound area.

[0064] Step 201: Based on the preset conversion coefficient, convert the RGB channel components of each pixel in the color image of the wound to obtain a brightness distribution image.

[0065] In this step, the preset conversion coefficient refers to a fixed weighting parameter set according to the difference in brightness contribution of each channel during color space dimensionality reduction. RGB channel components refer to the intensity values ​​of the three basic color information (red, green, and blue) that make up the pixels of a color image. The brightness distribution image refers to the aggregation of the three-channel pixel data of the original color image into a single-channel grayscale data matrix that reflects the reflectivity of the object's surface.

[0066] In this embodiment, the color component corresponding to each pixel position in the color image of the wound is first extracted. For the color image of patient A's wound, its local pixel distribution can be represented as a color pixel matrix. :

[0067]

[0068] Each element All include red components. Green components And the blue component Then, using the preset transformation coefficient vector... A linear weighted calculation is performed with the pixel components, where the coefficients satisfy the following condition to ensure that the converted brightness does not overflow: In this embodiment, We take empirical values ​​of 0.299, 0.587, and 0.114 respectively. The calculation formula is: Then, the calculation results at each pixel coordinate are filled into the corresponding grid. The final result is the brightness distribution image matrix. :

[0069]

[0070] Step 202: Divide the brightness value corresponding to each pixel in the brightness distribution image into multiple brightness levels, and use the preset pixel spacing and scanning angle as spatial constraints to count the frequency of occurrence of the brightness level of each pair of pixels in the brightness distribution image and perform normalization processing to obtain multiple co-occurrence probabilities.

[0071] In this step, brightness level refers to the discrete grayscale order obtained by dividing continuous brightness values ​​into intervals. Preset pixel spacing refers to the coordinate offset span between two target points selected during texture statistics. Preset scan angle refers to the search direction used to determine the relative spatial orientation of pixel pairs. Spatial constraint refers to the logical conditions, composed of spacing and angle, used to limit the selection range of pixel pairs. Occurrence frequency refers to the total number of pixel pairs satisfying a specific numerical combination within a specified range. Co-occurrence probability refers to the normalized frequency of occurrence of pixel pairs with a specific brightness level combination under specific spatial constraints.

[0072] In this embodiment of the application, the brightness distribution image matrix is ​​first... The element values ​​are divided into multiple brightness levels. Next, the preset pixel spacing is set. for Set the preset scanning angle They are respectively , , as well as Within the image region, count the pixel pairs that satisfy the above spatial constraints. Specifically, for brightness levels... and The total number of times a frequency occurs in a given direction is counted. This results in multiple frequency statistics matrices. :

[0073]

[0074] Then the frequency statistics matrix Each element in the matrix is ​​divided by the total number of pixels to perform normalization. This results in a probability matrix consisting of multiple co-occurrence probabilities. :

[0075]

[0076] probability matrix The spatial distribution patterns of different brightness combinations on the surface of the wound lesion were accurately recorded.

[0077] Step 203: Based on the co-occurrence probability, calculate the local contrast and information entropy of the brightness distribution image, and combine the local contrast and information entropy into a vector to obtain the texture distribution data.

[0078] In this step, local contrast refers to a feature quantity that measures the drasticness of local brightness changes in an image by calculating the expected value of the square of the brightness difference between pixels. Information entropy refers to a statistical metric calculated using probability distributions to measure the complexity or randomness of image texture distribution.

[0079] In the embodiments of this application, firstly based on the probability matrix Calculate the local contrast of the brightness distribution image and information entropy Local contrast The formula used to describe the degree of abrupt change in image grayscale is shown in formula (1):

[0080] (1)

[0081] in, The total number of preset brightness levels, and This represents the brightness level of a pixel pair. Simultaneously, to measure the complexity or randomness of the image texture distribution, information entropy is calculated. The calculation formula is shown in formula (2):

[0082] (2)

[0083] The calculated scalars are then spliced ​​together according to a preset combination order. For the wound area of ​​patient A, the resulting contrast is... With information entropy Encapsulation is performed. The final result is a texture distribution data vector representing the surface morphology features of the lesion. .

[0084] Step 204: Calculate the difference between adjacent pixels in the energy distribution data to obtain multiple energy variations. Based on the fluctuation range of the energy variations, scale the energy variations proportionally to obtain energy gradient data.

[0085] In this step, adjacent pixels refer to two sampling points that are physically adjacent in the spatial coordinate grid. Energy variation refers to the original numerical difference reflecting the fluctuation of local thermal radiation intensity. Fluctuation range refers to the difference between the maximum and minimum values ​​in the set of energy variations within a specified region.

[0086] In this embodiment of the application, the acquired energy distribution data matrix is ​​first read. :

[0087]

[0088] Next, the difference between the coordinates of adjacent pixels in the matrix is ​​calculated by subtraction to obtain the energy variation matrix. :

[0089]

[0090] Then in the matrix Identify the element with the largest value. and the element with the smallest value To determine the fluctuation range, a linear mapping is performed based on a preset scaling ratio. The calculation formula is as follows: Finally, the dimensionless energy gradient data matrix is ​​obtained. :

[0091]

[0092] in, It accurately reflects the relative intensity of infrared energy variation in space.

[0093] Step 103: Calculate the joint probability of texture distribution data and energy gradient data for each pixel coordinate in the color image of the wound, and map the joint probability to the association space to construct the association distribution map.

[0094] In this step, pixel coordinates refer to the two-dimensional row and column indices used to determine the physical location of each data point in the image data array. Joint probability refers to the statistical probability that texture features and energy gradient features simultaneously satisfy a specific numerical combination at a pixel coordinate location. Association space refers to a mathematical mapping space composed of multiple physical attribute dimensions, which can be a coordinate system composed of combinations of texture intensity and energy gradient strength. Association distribution map refers to the data matrix reflecting the inherent associations of the organization formed after back-mapping the joint probability corresponding to each pixel location to the image space grid.

[0095] like Figure 2 As shown, Figure 2 This is a flowchart illustrating a method for constructing an association distribution map, as provided in an embodiment of this application.

[0096] Step 301: Numerically align the texture distribution data and energy gradient data with the same pixel coordinates in the color image of the wound to obtain multiple pairs of vectors.

[0097] In this step, a vector pair refers to a composite feature vector formed by merging the texture component and the energy component at the same coordinate position.

[0098] In this embodiment of the application, the texture distribution data vector is first obtained. and energy gradient data matrix Pixel positions for the wound area. The corresponding contrast and entropy components are extracted, and the gradient value at that location is read. These values ​​are then encapsulated to obtain a vector pair matrix. :

[0099]

[0100] Each element is represented as a three-dimensional column vector. ,letter Indicates local contrast, letters Representing information entropy, the letter This represents the energy gradient.

[0101] For example, in the samples collected from Rehabilitation Center A, alignment operations were performed to ensure that each set of texture features could find its corresponding thermal field change features, thus forming a set of alignment features spanning the visual and energy dimensions.

[0102] Step 302: Count the frequency of occurrence of the same vector pair in the color image of the wound, construct an association space with texture distribution data and energy gradient data as coordinate axes, and calculate the joint probability of each pixel coordinate in the association space based on the frequency of occurrence.

[0103] In this step, the association space refers to a multi-dimensional mathematical space constructed using multiple feature attributes as coordinate axes to describe the joint distribution pattern of data. Frequency of occurrence refers to the total number of pixels with the same combination of numerical features within a specified image range. Total number of pixels refers to the sum of the products of all sampling points included in the effective domain of the wound color image. Joint probability refers to the normalized probability value of a specific feature combination appearing in the association space.

[0104] In this embodiment, an association space is constructed with texture components as the horizontal axis and energy gradient components as the vertical axis. Then, the vector pair matrix is ​​traversed. Count the frequency of occurrence of each type of vector. Simultaneously calculate the total number of pixels in the image. Then, the joint probability is calculated using division. The calculation formula is: Statistical analysis of the wound images of patient B revealed that pixel pairs with specific roughness and large thermal gradients occurred more frequently, and their corresponding joint probabilities... The corresponding increase.

[0105] For example, for a sequence of vector pairs The corresponding probability sequence is obtained through statistics. .

[0106] Step 303: Using the pixel coordinates of the color image of the wound as a spatial index, construct an association distribution map based on the corresponding joint probability.

[0107] In this step, spatial indexing refers to using the row and column numbers of the original pixel grid as the logical address for locating and accessing associated values. The association distribution map is a two-dimensional data array reflecting the inherent association strength of tissues, generated after mapping the joint probability back to the image's physical coordinates.

[0108] In this embodiment, the pixel coordinates of the color image of the wound are used as the spatial index. First, each coordinate position is retrieved. The corresponding original feature combination. Then, the joint probability corresponding to this combination is extracted from the association space. The probability value is then filled into the corresponding coordinate grid. Finally, the correlation distribution matrix is ​​constructed. :

[0109]

[0110] Among the elements This represents the numerical value indicating the correlation strength of the lesion at that coordinate location. Specifically, for active lesion areas, the mapped value is... The values ​​exhibit a clear clustered distribution pattern.

[0111] Step 1031: Combine the pixels in the color images of the wounds that have a joint probability greater than the preset correlation threshold in the correlation distribution map to obtain the lesion pixel set.

[0112] In this step, the preset correlation threshold refers to the critical value used to define the correlation probability between lesion tissue and background tissue during the initial screening of targets. The lesion pixel set refers to the discrete set of pixel coordinates that satisfy the high correlation constraint.

[0113] In this embodiment of the application, the constructed correlation distribution map matrix is ​​read. First, set a preset correlation threshold. Then iterate through each element in the correlation distribution matrix. And perform a size comparison. If the element value is greater than... If the coordinates are found to be true, then that coordinate is considered a candidate lesion point. Finally, all points that meet the criteria are spatially aggregated to obtain the lesion pixel set. : This set records the pixel coordinates of all suspected lesions in list form.

[0114] For example, when processing case data provided by Hospital A, a large number of low-probability normal skin areas in the background were removed, and the physical distribution range of the diseased tissue was initially defined.

[0115] Step 1032: Calculate the geometric center of the lesion pixel set, construct the minimum circumcircle surrounding all pixels in the lesion pixel set with the geometric center as the center, and use the minimum circumcircle as the preset initial curve.

[0116] In this step, the geometric center refers to the arithmetic mean position of the pixel set on the two-dimensional coordinate axes. The minimum circumcircle refers to the closed circular boundary that encloses the target set and has the smallest area. The preset initial curve refers to the initial trajectory of the zero-level set used to initiate the iterative evolution of the level set.

[0117] In this embodiment of the application, the lesion pixel set is first... The sequence of horizontal coordinates in and the ordinate sequence Calculate the average value for each. This yields the geometric center coordinate vector. Next, calculate the set. From each point in the middle to the geometric center The Euclidean distance is calculated, and the maximum distance value is selected as the radius. Then, with the geometric center... Construct the smallest circumcircle around the center. Finally, use this smallest circumcircle as the preset initial curve.

[0118] For example, for the wound distribution of patient B, a circular outline was generated that just covers all candidate lesions.

[0119] Step 104: Perform anisotropic convolution on the correlation distribution map to extract the correlation gradient, construct a velocity variable based on the correlation gradient, and use level set evolution to iteratively evolve the level set function constructed based on the preset initial curve, so that the initial curve converges to the edge of the lesion area in the wound color image, and obtain the convergence boundary curve.

[0120] In this step, the correlation gradient refers to the rate of change of the numerical values ​​in the correlation distribution map in terms of spatial orientation. The velocity variable refers to the vector driving force controlling the movement of the boundary curve towards the target edge; it can be a comprehensive vector combining image probability features and geometric constraints. Level set evolution refers to the numerical calculation process of continuously deforming the zero-level set interface in a two-dimensional plane using the dynamic iteration of a high-dimensional level set function. The preset initial curve refers to the closed trajectory surrounding the lesion body defined before the evolution begins. The level set function refers to a surface function defined on a two-dimensional grid whose function value corresponds to the distance to the evolution curve. The lesion region refers to the target lesion site in the wound that has abnormal texture and thermal characteristics. The convergence boundary curve refers to the final closed contour where deformation stops when the evolution process reaches a stable equilibrium state.

[0121] Step 401: Determine the evolution direction of each associated coordinate in the associated distribution map, and use anisotropic convolution kernels to perform convolution operations along the normal direction and tangent direction of the evolution direction to obtain the normal direction gradient and tangent direction gradient.

[0122] In this step, the evolution direction refers to the dominant orientation where the gradient change of the associated probability distribution at the associated coordinate point is most drastic. Anisotropic convolution kernels refer to filtering operators with inconsistent weight distributions across different spatial axes and capable of adapting to local structural deformations. The normal direction refers to the orientation perpendicular to the evolution trajectory of the lesion edge. The tangential direction refers to the orientation extending along the tangent at the lesion edge. The normal direction gradient is the numerical matrix reflecting the intensity of probability abrupt changes obtained after normal anisotropic filtering. The tangential direction gradient is the numerical matrix reflecting the continuity of the edge obtained after tangential anisotropic filtering.

[0123] In this embodiment of the application, the structure tensor is first used to analyze the correlation distribution graph matrix. Local geometric feature analysis is performed, and the evolution direction at each associated coordinate is determined by calculating the eigenvectors of the second-order moment matrix. Specifically, the eigenvectors corresponding to the large eigenvalues ​​of the structure tensor are calculated as the normal directions. Calculate the eigenvectors corresponding to the smallest eigenvalues ​​as the tangent directions. Then, asymmetric Gaussian convolution kernels are constructed along these two directions. and For the correlation distribution matrix respectively Perform convolution operation. Generate the gradient matrix of the normal direction. and the gradient matrix of the tangent direction :

[0124]

[0125] Among the elements Represents the gradient intensity along the normal direction, element This represents the gradient intensity along the tangent direction.

[0126] Step 402: Based on the magnitude ratio of the normal direction gradient to the tangent direction gradient, calculate the local anisotropy index of each associated coordinate in the associated distribution map, and based on the local anisotropy index, perform a weighted summation of the normal direction gradient and the tangent direction gradient to obtain the association degree gradient.

[0127] In this step, the magnitude ratio refers to the proportional relationship between the magnitude of the gradient in the normal direction and the magnitude of the gradient in the tangent direction at the same coordinate point. The local anisotropy index is a dimensionless metric used to represent the degree of significance of the structured features at the coordinate point location.

[0128] In this embodiment, the gradient matrix of the normal direction is first calculated. gradient matrix with tangent direction The modulus ratio of the corresponding elements. Then, the local anisotropy index is calculated using the exponential mapping function. For each coordinate point, the generated anisotropy index matrix... :

[0129]

[0130] Among the elements A value closer to 1 indicates that the point is closer to the lesion edge. Then, a weighted summation is performed on the gradients along the normal and tangent directions based on the local anisotropy index. This results in the enhanced correlation gradient matrix. :

[0131]

[0132] like Figure 3 As shown, Figure 3 This is a flowchart illustrating a method for obtaining a convergence boundary curve, provided in an embodiment of this application.

[0133] Step 411: Calculate the sign distance between each associated coordinate and the initial curve in the associated distribution map, construct a horizontal set function based on the sign distance, and use the pixel spacing between adjacent pixels in the wound color image as the spatial reference to calculate the second-order partial derivatives of each element coordinate in the horizontal and vertical directions to obtain the local curvature of each element coordinate.

[0134] In this step, element coordinates refer to the index pairs in the level set function matrix used to determine the position of each value. The second partial derivative is the second derivative of the rate of change of the function value in spatial orientation, used to capture local shape changes of the surface.

[0135] In this embodiment, the symbol distance value of pixels inside the initial curve is defined as negative, while that outside the curve is positive. First, the distance between each associated coordinate in the associated distribution map and the preset initial curve is calculated. The signed distance value is then used to construct the initial level set function matrix. :

[0136]

[0137] Then, based on the pixel spacing between adjacent pixels Using the central difference scheme as a spatial reference, the first-order partial derivatives are calculated. And second-order partial derivatives. Finally, calculate the local curvature at each element's coordinates. The calculation formula is shown in formula (3):

[0138] (3)

[0139] This formula accurately describes the degree of geometric curvature of the evolution interface at the edge of the lesion.

[0140] For example, for lesion region A, the generated local curvature matrix :

[0141]

[0142] Step 412: Calculate the influence coefficient based on the correlation gradient and local curvature, and obtain the velocity variable by weighted summation of the correlation gradient and local curvature based on the influence coefficient.

[0143] In this step, the influence coefficient refers to the weighting parameter used to adjust the proportion of the contribution of the correlation gradient and local curvature to the curve evolution.

[0144] In this embodiment of the application, the calculated correlation gradient matrix is ​​first read. and local curvature matrix Next, the influence coefficient is set according to the preset empirical parameters. The correlation gradient matrix is ​​calculated using a linear weighted algorithm. and local curvature matrix Perform a combined calculation. The calculation formula is: Finally, the velocity variable matrix is ​​obtained. :

[0145]

[0146] Specifically, by adjusting the influence coefficient The magnitude of this value allows the velocity variable to reference the correlation gradient more extensively at complex lesion edges, thereby guiding the curve to move accurately.

[0147] Step 413: Update each element in the level set function based on the velocity variable, determine the intermediate evolution curve according to the element coordinates corresponding to the zero element in the updated elements, and calculate the offset of the intermediate evolution curve relative to the previous intermediate evolution curve.

[0148] In this step, the zero element refers to the element with a value of zero in the level set function matrix, and their corresponding positions collectively constitute the current evolutionary interface. The offset refers to the geometric distance of the evolution curve in spatial position between two adjacent iterations.

[0149] In this embodiment of the application, the velocity variable matrix is ​​used. For the level set function matrix Perform iterative updates. The update process follows a time-step progression logic to obtain the updated function matrix. Next, the search matrix is ​​retrieved. The coordinates of the zero element are used to determine the current intermediate evolution curve. Then, the intermediate evolution curve is calculated. The offset scalar is calculated relative to the positional difference of the curve generated in the previous iteration. Specifically, it is obtained by calculating the average of the positional changes of all curve points. .

[0150] For example, for a set of evolution steps, record the resulting offset sequence. .

[0151] Step 414: When the offset is greater than the preset change threshold, return to update each element in the level set function based on the velocity variable until the offset is less than or equal to the change threshold, and determine the intermediate evolution curve as the convergence boundary curve.

[0152] In this step, the preset change threshold refers to the minimum critical value used to determine whether the evolution of the level set has reached a convergent equilibrium state.

[0153] In this embodiment of the application, the calculated offset is first... Compared with the preset change threshold Perform a logical comparison. If the offset... Greater than the preset change threshold If the function updates, it returns to the step where it was executed to continue driving the curve toward the edge. This process is repeated until the offset is reached. Less than or equal to the preset change threshold Finally, the iteration is stopped, and the intermediate evolution curve at this point is determined as the convergence boundary curve. Specifically, for the wound image of patient B, when the offset... When the value drops to within a preset fluctuation threshold range, the output convergence boundary curve is... It precisely defines the boundary between necrotic tissue and granulation tissue.

[0154] Step 105: Based on the correlation distribution map, cluster the closed region enclosed by the convergence boundary curve to identify the necrotic region and granulation region in the color image of the wound, and calculate the area measurement value based on the number of pixels in the necrotic region and granulation region.

[0155] In this step, the closed region refers to the set of pixels representing the interior of the lesion, enclosed by a converging boundary curve. The necrotic region refers to a pathological area in the wound that exhibits a low correlation probability distribution due to tissue inactivation. The granulation region refers to a healing area in the wound that exhibits active metabolic correlation characteristics due to good blood supply. The area measurement value refers to the tissue geometric size obtained by pixel counting combined with physical scaling, and can be a numerical value reflecting the actual physical scale of the lesion.

[0156] Step 501: Calculate the discrete deviation between the associated coordinates of different targets located within the convergence boundary curve in the associated distribution map, and classify the associated coordinates of the targets into the first feature cluster and the second feature cluster based on the discrete deviation.

[0157] In this step, the target associated coordinates refer to those located on the convergence boundary curve. Internal and in the correlation distribution matrix The corresponding position index pairs on the grid. Discrete bias refers to the geometric distance value used to measure the similarity or difference between the associated coordinates of two different targets in the associated attribute space. The first feature cluster and the second feature cluster refer to two non-overlapping subsets of feature coordinates formed by dividing pixels with similar associated attributes through a clustering algorithm.

[0158] In this embodiment, the convergence boundary curve is first extracted. The positional information of all pixels within the enclosed area. Associated coordinates for each target. From the correlation distribution matrix The corresponding correlation score is read from the database. Then, cluster analysis is used, with the correlation score as the clustering feature, to calculate the numerical dispersion deviation between different pixels. Through iterative calculation, the intra-cluster dispersion deviation is minimized, thereby dividing all target correlated pixels into two categories: the first feature cluster and the second feature cluster. Second feature cluster .

[0159] For example, the sequence of correlation values ​​extracted from the internal data of patient B's lesions can be represented as follows: Then, through iterative calculations, the intra-class discrete bias is minimized, thereby classifying all target associated coordinates into two categories. This yields the first feature cluster composed of coordinate vectors. and the second feature cluster .

[0160] Step 502: Calculate the average value of the first feature cluster and the average value of the second feature cluster and compare them. Mark the region where the feature cluster with the smaller average value is located as the necrotic region and the region where the feature cluster with the larger average value is located as the granulation region.

[0161] In this embodiment of the application, the first feature clusters are first obtained respectively. Second feature cluster The coordinates of each point in the correlation distribution matrix The corresponding values ​​are then calculated. Next, the average value for each set is calculated. For the first feature cluster... The first average value was calculated. For the second feature cluster The second average value was calculated. Then the first average value will be... Compared with the second average Perform a mean comparison.

[0162] For example, in the clinical analysis of Hospital A, if the mean vector is calculated... And satisfy It is on the side with the smaller value. Because necrotic tissue has a low metabolic level and exhibits a low probability distribution in the correlation distribution map, the first feature cluster is... The area in question is marked as a necrotic area. At the same time, the side with the larger average value is the second feature cluster. The area is marked as a granulation area. .

[0163] Step 503: Count the total number of first pixels in the necrotic area and the total number of second pixels in the granulation area, and calculate the product of the total number of first pixels and the total number of second pixels with the preset single pixel area to obtain the area measurement value of the necrotic area and the area measurement value of the granulation area.

[0164] In this step, the total number of first pixels refers to those belonging to the necrotic area. The total number of elements in the set of pixels. The second total number of pixels refers to those belonging to the granulation region. The number of elements in a pixel set. Preset single-pixel area refers to the actual physical space size represented by a single pixel grid, pre-set according to the resolution parameters of the imaging device. Area measurement is a quantitative physical indicator reflecting the degree of tissue growth or damage, obtained through pixel scale conversion.

[0165] In this embodiment of the application, the identified necrotic areas are first... Perform pixel accumulation statistics to obtain the total number of the first pixel. At the same time, for the granulation tissue area Perform pixel accumulation statistics to obtain the total number of the second pixel. Next, retrieve the preset single-pixel area. Then, the area measurements are calculated using multiplication. The specific calculation formula is as follows: ,in This represents the area measurement of the necrotic region. This represents the area measurement of the granulation tissue region.

[0166] For example, a sequence of pixel counts was obtained from the lesions on the feet of subject B. By comparing with the preset single pixel area Multiplying these results yields a quantitative record of the necrotic area and the granulation area.

[0167] This application's embodiments ensure absolute spatial consistency between visual texture information and infrared thermal energy information by acquiring color images of the wound and aligned energy distribution data. This suppresses interference from raw data caused by uneven ambient light or individual patient base temperature differences, improving the quality of feature representation. Deep fusion of visual and thermal features is achieved, transforming previously isolated physical quantities into correlated feature fields characterizing tissue properties. This solves the problem of insufficient multi-source information fusion in existing technologies, shields against pseudo-edge interference caused by noise such as exudate and crusting, and drives the curve to adaptively converge to the true boundary, reducing boundary positioning deviation. It eliminates the subjective errors of manual assessment, providing highly reliable decision support for clinical evaluation.

[0168] Figure 4 This is a schematic diagram illustrating a specific implementation of a diabetic wound image analysis system provided in this application. (Refer to...) Figure 4 The system may include:

[0169] The acquisition module 21 is used to acquire a color image of the wound area of ​​a diabetic foot patient, as well as energy distribution data of infrared radiation aligned with the pixel coordinates in the color image of the wound.

[0170] Analysis module 22 is used to perform texture distribution statistics on the color image of the wound using the gray-level co-occurrence matrix to obtain texture distribution data, and to perform gradient normalization on the energy distribution data to obtain energy gradient data;

[0171] Calculation module 23 is used to calculate the joint probability of texture distribution data and energy gradient data for each pixel coordinate in the wound color image, and to map the joint probability to the association space to construct the association distribution map;

[0172] Convolution module 24 is used to perform anisotropic convolution on the correlation distribution map to extract the correlation gradient, construct a velocity variable based on the correlation gradient, and use level set evolution to iteratively evolve the level set function constructed based on the preset initial curve, so that the initial curve converges to the edge of the lesion area in the wound color image to obtain the convergence boundary curve.

[0173] The recognition module 25 is used to cluster and segment the closed region enclosed by the convergence boundary curve based on the correlation distribution map, identify the necrotic region and granulation region of the wound color image, and calculate the area measurement value based on the number of pixels of the necrotic region and granulation region.

[0174] This application provides a diabetes wound image analysis system to implement the aforementioned diabetes wound image analysis method. Therefore, the specific implementation of the diabetes wound image analysis system can be found in the embodiment section of the diabetes wound image analysis method above. The specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0175] Figure 5 A schematic diagram of the hardware structure of the electronic device provided in an embodiment of this application is shown.

[0176] This application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the above-described methods for analyzing diabetic wound images.

[0177] The electronic device may include a processor 510 and a memory 520 storing computer program instructions.

[0178] Specifically, the processor 510 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0179] Memory 520 may include mass storage for data or instructions. For example, and not limitingly, memory 520 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 520 may include removable or non-removable (or fixed) media. Where appropriate, memory 520 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 520 is non-volatile solid-state memory.

[0180] Memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to the first aspect of this disclosure.

[0181] The processor 510 reads and executes computer program instructions stored in the memory 520 to implement any of the diabetic wound image analysis methods in the above embodiments.

[0182] In one example, the electronic device may also include a communication interface 530 and a bus 540. Wherein, such as Figure 5 As shown, the processor 510, memory 520, and communication interface 530 are connected through bus 540 and complete communication with each other.

[0183] The communication interface 530 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0184] Bus 540 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 540 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0185] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described methods for analyzing diabetic wound images.

[0186] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0187] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the embodiments of the diabetic wound image analysis method described above.

[0188] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0189] The above provides a detailed description of a method and system for analyzing diabetic wound images provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method of diabetic wound image analysis, characterized by, include: Acquire color images of the wound area of ​​a diabetic foot patient, and energy distribution data of infrared radiation aligned with the pixel coordinates in the color images of the wound; The texture distribution of the wound color image is statistically analyzed using the gray-level co-occurrence matrix to obtain texture distribution data, and the energy distribution data is gradient normalized to obtain energy gradient data. Calculate the joint probability of the texture distribution data and the energy gradient data for each pixel coordinate in the color image of the wound, and map the joint probability to the association space to construct an association distribution map; Anisotropic convolution is performed on the correlation distribution map to extract the correlation gradient. A velocity variable is constructed based on the correlation gradient. The level set function constructed based on the preset initial curve is iteratively evolved using level set evolution to make the initial curve converge to the edge of the lesion area in the wound color image, thus obtaining the convergence boundary curve. Based on the correlation distribution map, the closed region enclosed by the convergence boundary curve is clustered and segmented to identify the necrotic region and granulation region of the wound color image, and the area measurement value is calculated according to the number of pixels of the necrotic region and the granulation region.

2. The method of claim 1, wherein, After calculating the joint probability of the texture distribution data and the energy gradient data for each pixel coordinate in the color image of the wound, and mapping the joint probability to an association space to construct an association distribution map, the method further includes: The pixels in the color image of the wound corresponding to the joint probability greater than the preset correlation threshold in the correlation distribution map are combined to obtain the lesion pixel set; Calculate the geometric center of the lesion pixel set, construct the minimum circumcircle surrounding all pixels in the lesion pixel set with the geometric center as the center, and use the minimum circumcircle as the preset initial curve.

3. The method of claim 1, wherein, The texture distribution of the wound color image is statistically analyzed using the gray-level co-occurrence matrix to obtain texture distribution data. Then, the energy distribution data is gradient normalized to obtain energy gradient data, including: Based on preset conversion coefficients, the RGB channel components of each pixel in the color image of the wound are converted to obtain a brightness distribution image; The brightness value corresponding to each pixel in the brightness distribution image is divided into multiple brightness levels. With preset pixel spacing and scanning angle as spatial constraints, the frequency of occurrence of brightness levels of each pair of pixels in the brightness distribution image is counted and normalized to obtain multiple co-occurrence probabilities. Based on the co-occurrence probability, the local contrast and information entropy of the brightness distribution image are calculated, and the local contrast and information entropy are vector-combined to obtain texture distribution data; The difference between adjacent pixels in the energy distribution data is calculated to obtain multiple energy variations. Based on the fluctuation range of the energy variations, the energy variations are scaled proportionally to obtain energy gradient data.

4. The method of claim 1, wherein, Calculate the joint probability of the texture distribution data and the energy gradient data for each pixel coordinate in the color image of the wound, and map the joint probability to the association space to construct an association distribution map, including: The texture distribution data and energy gradient data with the same pixel coordinates in the color image of the wound are numerically aligned to obtain multiple pairs of vectors. The frequency of occurrence of the same vector pair in the color image of the wound is counted, an association space is constructed using the texture distribution data and the energy gradient data as coordinate axes, and the joint probability of each pixel coordinate in the association space is calculated based on the frequency of occurrence. Using the pixel coordinates of the color image of the wound as a spatial index, an association distribution map is constructed based on the corresponding joint probability.

5. The method of claim 1, wherein, Anisotropic convolution is performed on the correlation distribution map to extract the correlation gradient, including: The evolution direction of each associated coordinate in the associated distribution map is determined, and anisotropic convolution kernels are used to perform convolution operations along the normal direction and tangent direction of the evolution direction to obtain the normal direction gradient and the tangent direction gradient. Based on the magnitude ratio of the normal direction gradient to the tangent direction gradient, the local anisotropy index of each associated coordinate in the associated distribution map is calculated, and based on the local anisotropy index, the normal direction gradient and the tangent direction gradient are weighted and summed to obtain the association degree gradient.

6. The method of claim 5, wherein, Based on the correlation gradient, a velocity variable is constructed. The level set function, constructed based on a preset initial curve, is iteratively evolved using level set evolution. This causes the initial curve to converge towards the edge of the lesion region in the color image of the wound, resulting in a convergence boundary curve, including: Calculate the sign distance between each associated coordinate in the associated distribution map and the initial curve, construct a horizontal set function based on the sign distance, and use the pixel spacing between adjacent pixels in the wound color image as a spatial reference to calculate the second-order partial derivatives of each element coordinate in the horizontal and vertical directions to obtain the local curvature of each element coordinate. The influence coefficient is calculated based on the correlation gradient and the local curvature, and the velocity variable is obtained by weighted summation of the correlation gradient and the local curvature based on the influence coefficient. The level set function is updated based on the velocity variable. The intermediate evolution curve is determined according to the coordinates of the zero element in the updated elements. The offset of the intermediate evolution curve relative to the previous intermediate evolution curve is calculated. When the offset is greater than the preset change threshold, the system returns to update each element in the level set function based on the velocity variable until the offset is less than or equal to the change threshold, and the intermediate evolution curve is determined as the convergence boundary curve.

7. The method of claim 1, wherein, Based on the aforementioned correlation distribution map, the closed region enclosed by the convergence boundary curve is clustered and segmented to identify the necrotic and granulation regions in the color image of the wound. The area measurement is then calculated based on the number of pixels in the necrotic and granulation regions, including: Calculate the discrete deviation between the associated coordinates of different targets located within the convergence boundary curve in the associated distribution map, and classify the associated coordinates of the targets into a first feature cluster and a second feature cluster based on the discrete deviation; Calculate the average value of the first feature cluster and the average value of the second feature cluster and compare them. Mark the region where the feature cluster with the smaller average value is located as the necrotic region and the region where the feature cluster with the larger average value is located as the granulation region. The total number of first pixels in the necrotic region and the total number of second pixels in the granulation region are counted respectively. The product of the total number of first pixels and the total number of second pixels with the preset single pixel area is calculated to obtain the area measurement value of the necrotic region and the area measurement value of the granulation region.

8. A diabetic wound image analysis system, characterized by, include: The acquisition module is used to acquire a color image of the wound area of ​​a diabetic foot patient, and energy distribution data of infrared radiation aligned with the pixel coordinates in the color image of the wound. The analysis module is used to perform texture distribution statistics on the color image of the wound using the gray-level co-occurrence matrix to obtain texture distribution data, and to perform gradient normalization on the energy distribution data to obtain energy gradient data. The calculation module is used to calculate the joint probability of the texture distribution data and the energy gradient data for each pixel coordinate in the color image of the wound, and to map the joint probability to the association space to construct an association distribution map; The convolution module is used to perform anisotropic convolution on the correlation distribution map to extract the correlation gradient, construct a velocity variable based on the correlation gradient, and use level set evolution to iteratively evolve the level set function constructed based on the preset initial curve, so that the initial curve converges to the edge of the lesion area in the wound color image to obtain the convergence boundary curve. The identification module is used to perform clustering and segmentation of the closed region enclosed by the convergence boundary curve based on the correlation distribution map, identify the necrotic region and granulation region of the wound color image, and calculate the area measurement value according to the number of pixels of the necrotic region and the granulation region.

9. An electronic device, comprising: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of a method for analyzing diabetic wound images as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of a method for analyzing diabetic wound images as described in any one of claims 1 to 7.

Images