Medical image quality enhancement system based on big data

Abstract

The invention relates to the technical field of machine learning, in particular to a medical image quality enhancement system based on big data, which comprises an image partition analysis module, a sparse feature positioning module, a texture compensation generation module, a boundary mutation recognition module and a fracture continuous filling module. According to the method, the texture missing region is screened through the product of the gray level change rate and the edge density, the detail degradation position in the image is accurately recognized, directional texture reconstruction is achieved based on multi-angle high-frequency residual signal convolution, the region structure reduction precision is improved, and gray level consistency and edge connectivity detection are fused; the method comprises the steps of enhancing the coordination of brightness distribution and edge morphology of an image, constructing a gray sequence difference model for boundary sudden change points, eliminating an abnormal boundary misrecognition risk, constructing smooth transition through intersection gray mean filling and linear interpolation, realizing continuous repair of a fracture area, improving the structural splitting feeling caused by a boundary fault, and improving the image quality. And the completeness of the image in the aspects of vision and diagnosis is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and in particular to a medical image quality enhancement system based on big data. Background Art

[0002] The field of machine learning technology includes technical methods that use algorithms to automatically extract patterns from data and make predictions and decisions. The core content of this technical field is to build models that can continuously optimize their performance through training data, so as to complete tasks such as classification, recognition and regression in different scenarios. In many fields such as image processing, speech recognition, and natural language processing, machine learning has formed a systematic research framework, covering many aspects such as feature extraction, model training, data set construction, and optimization strategies. With the improvement of computing power and the expansion of data scale, the technology in this field continues to deepen, and big data technology is widely integrated for the analysis and processing of complex data scenarios.

[0003] Among them, the medical image quality enhancement system based on big data refers to a technical solution that uses large-scale medical image data to train machine learning models to improve image clarity, contrast, structural integrity and other quality indicators. The technical matters targeted by this patent subject include problems such as noise artifacts and blurring in the acquisition process of medical images. Specifically, the mapping relationship between the original image and the high-quality image is trained by constructing a convolutional neural network, and the model is supervised and learned based on the labeled medical image data set in the big data sample library. The image enhancement algorithm is used to extract multi-layer features from the input image and generate quality-optimized image output. This method usually combines image processing methods such as residual structure attention mechanism image denoising and super-resolution reconstruction to complete the quality enhancement process.

[0004] Existing technologies rely on model-side training of the entire image, which makes it difficult to distinguish local degraded areas in the image, and is prone to uneven enhancement when dealing with problems such as missing details or structural breaks. There is a lack of a distinguishing mechanism for the non-structural mutation edges that appear in the image, and noise is easily misjudged as a boundary signal, resulting in repair deviations. The grayscale filling process lacks local statistical guidance, which can easily lead to problems such as discontinuous brightness and abrupt boundaries. Taking a noisy MRI image as an example, after enhancement, some areas still have blurred details and discontinuous edges, which affects structural recognition and limits the application of images in high-precision diagnostic needs. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a medical image quality enhancement system based on big data.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: A medical image quality enhancement system based on big data comprises: The sparse feature location module calculates the product of grayscale change rate and edge density based on the medical image partition feature matrix, selects partitions with a product lower than 3.2, records the texture missing area location information, and generates a texture missing area location information set; The texture compensation generation module extracts the high-frequency residual signal image of the corresponding area according to the texture missing area position information set, performs gradient convolution at 0°, 45°, 90°, and 135°, generates a texture enhancement matrix by normalized weighted fusion, superimposes the corresponding pixels of the original image, detects the grayscale distribution consistency and edge connectivity, and generates an image texture enhancement metric value; The boundary mutation recognition module scans the pixel grayscale difference based on the image texture enhancement measurement value, records the mutation position, extracts the grayscale sequence of 16 pixels before and after, calculates the mean difference and the variance difference, screens the potential fracture boundary, and generates a potential fracture boundary coordinate set; The fracture continuous filling module calls the potential fracture boundary coordinate set, extracts the grayscale sequences on both sides of the mutation point, calculates the grayscale range of the intersection, calculates the intersection mean to fill the mutation point, selects 4 pixels before and after the mutation point, performs linear grayscale interpolation to generate a smooth gradient transition, replaces the pixel values ​​of the corresponding area of ​​the image, and establishes a grayscale continuous filling sequence.

[0007] As a further solution of the present invention, the texture missing area location information set specifically includes texture sparse distribution points, structural discontinuous areas, and grayscale sudden drop areas. The image texture enhancement measurement value includes local texture consistency score, edge structure continuity score, and enhanced area coverage ratio. The potential fracture boundary coordinate set includes grayscale mutation points, edge structure fracture lines, and abnormal texture turning zones. The grayscale continuous filling sequence specifically refers to the smooth grayscale layer of the fracture area, the pixel value gradient transition zone, and the texture connection completion segment.

[0008] As a further solution of the present invention, the image partition analysis module includes: The image partitioning submodule obtains the image boundary range and pixel distribution density based on the soft tissue MRI image pixel array, performs partition operation on the image array according to the 64×64 pixel size, assigns partition numbers according to the coordinate area and locates the center point of the area, extracts the pixel point set in the corresponding area, and generates the regional pixel set matrix; The grayscale sequence extraction submodule calls each sub-region pixel set in the regional pixel set matrix, extracts and arranges the grayscale value sequence in row priority order, performs average difference operation on the grayscale value and the difference set between the grayscale values ​​of adjacent pixels, and calculates the regional grayscale mean difference; The edge pixel detection submodule detects edge pixels in multiple regions based on the regional grayscale mean difference and the pixel sets in multiple sub-regions, and identifies the pixel sets located at the edge boundary, using the formula: ; Calculate the regional edge strength index, identify the edge pixels and calculate the edge strength of the entire region based on the gray value difference to obtain the medical image partition feature matrix; in, Representative The edge strength index of the image sub-region, Representative In the region The gray value of edge pixels, Represents the average gray value of edge pixels in the region. Represents the standard deviation of the grayscale values ​​of edge pixels, represents a constant used to avoid zero denominator, Represents the adjustment index that controls the expansion degree of the grayscale difference index. Representative The total number of region edge pixels, Represents the sub-region number, Represents the sequence number of edge pixels in the region.

[0009] As a further solution of the present invention, the sparse feature positioning module includes: The texture grayscale feature extraction submodule obtains the image grayscale distribution in the multi-partition area based on the medical image partition feature matrix, calculates the grayscale mean and grayscale variance after extracting the grayscale values ​​of all pixels in the partition, constructs a two-dimensional array to represent the multi-partition texture features in a combined manner, and establishes an image texture parameter matrix; The grayscale edge calculation submodule calls the grayscale mean and grayscale variance of the corresponding partition in the image texture parameter matrix, and calculates the grayscale change rate and edge density value of multiple partitions according to the grayscale change degree and the number of edge pixels, and uses the formula: ; The grayscale-edge comprehensive values ​​of multiple partitions are obtained by operation, and the grayscale-edge comprehensive matrix is ​​established in combination with the partition positions; in, Represents the grayscale-edge comprehensive value of the jth partition, represents the grayscale mean of the jth partition, Represents the grayscale mean of the entire image, represents the grayscale variance of the jth partition, represents the pixel density value at the edge of the j-th partition, represents the texture directional diffusion of the jth partition, where j represents the sub-region index; The texture missing screening submodule screens the partitions with comprehensive values ​​lower than the threshold value 3.2 according to the grayscale edge comprehensiveness matrix, records the missing area position information according to the combination of the partition number and the coordinate information, and obtains the texture missing area position information set.

[0010] As a further solution of the present invention, the texture compensation generation module includes: The high-frequency extraction submodule locates the corresponding image area in the original image based on the texture missing area position information set, extracts the difference between the multi-pixel and the eight-neighborhood pixel mean in the regional pixel matrix, constructs a high-frequency distribution map in the region, and simultaneously calculates the residual signal values ​​of the multi-pixel points and stores the residual vector set to generate a high-frequency residual signal value; The gradient fusion submodule calls the high-frequency residual signal value, performs gradient convolution processing in four directions of 0°, 45°, 90°, and 135°, extracts the gradient response image corresponding to multiple directions, calculates the contrast, directional consistency, and response intensity of each direction in the local area, and performs normalization processing based on the response value, using the formula: ; The fused texture enhancement matrix is ​​obtained by operation, and is superimposed pixel by pixel on the corresponding area of ​​the original image to obtain an enhanced superimposed pixel matrix; in, Indicates The gradient response matrix in each direction, Indicates The local contrast value in each direction, Indicates The gradient direction consistency coefficient in each direction, Indicates the original grayscale value of the corresponding pixel in the direction, Represents the grayscale mean of all pixels in the area, represents the normalized smoothing factor constant, Represents the texture enhancement matrix after normalization and fusion; The texture connectivity assessment submodule calls the enhanced superimposed pixel matrix to detect the grayscale mean, grayscale standard deviation and edge pixel gradient changes of the enhanced area and the neighborhood area, judge the pixel continuity and grayscale consistency of the edge structure of the enhanced area, and calculate the ratio of the edge connected area to the total area of ​​the enhanced area, and generate the image texture enhancement metric value in combination with the grayscale variance.

[0011] As a further solution of the present invention, the boundary mutation recognition module includes: The texture enhancement recognition submodule performs intensive scanning of the image pixel area based on the image texture enhancement metric value, obtains the gray value variation range in the adjacent area of ​​multiple pixel points in the image, calculates the gray value difference sequence of adjacent pixels, and determines the image mutation point according to the maximum difference position in the sequence, and generates a mutation gray point position set; The grayscale difference extraction submodule calls the mutation grayscale point position set, extracts the grayscale sequence of 16 pixels before and after each mutation point, obtains the mean and variance of the corresponding grayscale sequence, and simultaneously calculates the grayscale mean difference and grayscale variance difference to form a mutation pixel grayscale change parameter set; The potential boundary screening submodule determines whether the grayscale mean difference and grayscale variance difference meet the target threshold according to the grayscale change parameter set of the mutation pixel, using the formula: ; Calculate the boundary mutation significance value and compare it with the set significance threshold to establish the potential fracture boundary coordinate set; in, Representative position The boundary mutation significance value of Represents the mutation point position The difference between the mean values ​​of the grayscale sequences of the previous and next 16 pixels, Represents the mutation point position The difference between the variances of the grayscale sequences of the previous and next 16 pixels, Represents the mutation point position The minimum grayscale value in the first 16 pixel grayscale sequence, Represents the mutation point position The minimum grayscale value in the last 16 pixel grayscale sequence.

[0012] As a further solution of the present invention, the system further includes: The image partition analysis module obtains the soft tissue MRI image pixel array, divides the image into 64×64 partitions, extracts the regional grayscale sequence and the mean of adjacent differences, counts the edge pixel ratio, and establishes the medical image partition feature matrix; The medical image partition feature matrix includes regional grayscale balance, edge texture density, and regional grayscale difference.

[0013] As a further solution of the present invention, the fracture continuous filling module comprises: The intersection grayscale extraction submodule calls the potential fracture boundary coordinate set, extracts the grayscale sequence on both sides of the mutation point, determines the grayscale value interval of the same pixel position in the sequence according to the corresponding relationship of the pixel position, screens the grayscale intersection area to calculate the mean, and obtains the intersection grayscale mean; The mutation point filling submodule locates the mutation point coordinate position in the image according to the intersection grayscale mean, performs grayscale filling processing on the mutation point pixels using the intersection grayscale mean, and outputs the mutation point filling result layer to generate a mutation point grayscale value set; The linear interpolation construction submodule selects 4 pixels before and after the mutation point to construct a grayscale sequence based on the grayscale value set of the mutation point, calculates the grayscale difference of adjacent pixels and the pixel spacing, and uses the formula: ; Calculate the smooth transition grayscale value, embed the linear interpolation grayscale sequence into the corresponding area of ​​the image, and establish a grayscale continuous filling sequence; in, Represents a smooth transition gray value, Represents the grayscale pixel value before and after the mutation point, Represents the mutation point and The pixel distance is pixels, represents the grayscale value generated by the interpolated pixel position, Represents the number of grayscale participation points before and after, Represents the number of interpolation sequences, s and t are index variables, and r is the current interpolation position.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, the texture missing area is screened by multiplying the grayscale change rate and the edge density, accurately identifying the location of detail degradation in the image, and avoiding the problem of global enhancement covering up local details. Directional texture reconstruction is achieved based on multi-angle high-frequency residual signal convolution to improve the accuracy of regional structure restoration. The grayscale consistency and edge connectivity detection are integrated to enhance the coordination of the brightness distribution and edge morphology of the image. A grayscale sequence difference model is constructed for the boundary mutation points to eliminate the risk of misidentification of abnormal boundaries. A smooth transition is constructed by filling the intersection grayscale mean and linear interpolation to achieve continuous repair of the broken area, improve the sense of structural fragmentation caused by the boundary fault, and improve the integrity of the image at the visual and diagnostic levels. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a system flow chart of the present invention; Figure 2 This is a flow chart of the image partition analysis module of the present invention; Figure 3 This is a flow chart of the sparse feature positioning module of the present invention; Figure 4 This is a flow chart of a texture compensation generation module of the present invention; Figure 5 This is a flow chart of the boundary mutation identification module of the present invention; Figure 6 This is a flow chart of the fracture continuous filling module of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0017] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0018] Embodiment 1: See also Figure 1 , a medical image quality enhancement system based on big data includes: The image partition analysis module obtains the soft tissue MRI image pixel array, divides the image into 64×64 partitions, extracts the regional grayscale sequence and the mean of adjacent differences, counts the edge pixel ratio, and establishes the medical image partition feature matrix; The sparse feature location module calculates the product of grayscale change rate and edge density based on the medical image partition feature matrix, selects partitions with a product lower than 3.2, records the texture missing area location information, and generates a texture missing area location information set; The texture compensation generation module extracts the high-frequency residual signal image of the corresponding area according to the position information set of the texture missing area, performs gradient convolution at 0°, 45°, 90°, and 135°, and generates a texture enhancement matrix by normalized weighted fusion. It superimposes the corresponding pixels of the original image, detects the consistency of grayscale distribution and edge connectivity, and generates an image texture enhancement metric value. The boundary mutation recognition module is based on the image texture enhancement measurement value, scans the pixel grayscale difference, records the mutation position, extracts the grayscale sequence of 16 pixels before and after, calculates the mean difference and variance difference, screens the potential fracture boundary, and generates the potential fracture boundary coordinate set; The fracture continuous filling module calls the potential fracture boundary coordinate set, extracts the grayscale sequences on both sides of the mutation point, calculates the grayscale range of the intersection, calculates the intersection mean to fill the mutation point, selects 4 pixels before and after the mutation point, performs linear grayscale interpolation to generate a smooth gradient transition, replaces the pixel values ​​of the corresponding area of ​​the image, and establishes a grayscale continuous filling sequence.

[0019] The feature matrix of medical image partitioning includes regional grayscale balance, edge texture density, and regional grayscale difference. The texture missing area location information set specifically includes texture sparse distribution points, structural discontinuity areas, and grayscale sudden drop areas. The image texture enhancement measurement values ​​include local texture consistency score, edge structure continuity score, and enhanced area coverage ratio. The potential fracture boundary coordinate set includes grayscale mutation points, edge structure fracture lines, and abnormal texture turning zones. The grayscale continuous filling sequence specifically refers to the smooth grayscale layer of the fracture area, the pixel value gradient transition zone, and the texture connection completion segment.

[0020] See also Figure 2 , the image partition parsing module includes: The image partitioning submodule obtains the image boundary range and pixel distribution density based on the soft tissue MRI image pixel array, performs partition operation on the image array according to the 64×64 pixel size, assigns partition numbers according to the coordinate area and locates the center point of the area, extracts the pixel point set in the corresponding area, and generates the regional pixel set matrix; First, determine the size information of the pixel array. For example, get the original image matrix as Pixels, mark the horizontal and vertical coordinate ranges of the pixels respectively (for example, the horizontal and vertical coordinate ranges are both 0 to 511), and then according to the pre-set area size (that is, the size of each sub-area is The specific operation is: according to the horizontal and vertical coordinates, the division is performed every 64 pixels. For example, the horizontal coordinate 0 to 63 pixels and the vertical coordinate 0 to 63 pixels are determined as the first partition, the horizontal coordinate 64 to 127 pixels and the vertical coordinate 0 to 63 pixels are determined as the second partition, and so on, until the entire The matrix is ​​divided into 64 sub-areas, numbered from 1 to 64, and the center coordinates of each sub-area are determined respectively, such as the coordinates of the center point of the first partition are (31.5, 31.5), the center point of the second partition is (95.5, 31.5), and so on. Subsequently, the pixel affiliation is determined according to the coordinate area. For example, the pixel coordinate set of the first partition is all pixel points from 0 to 63 horizontally and 0 to 63 vertically. The pixel sets in each area are classified by coordinates and stored in the regional pixel set matrix. For example, the coordinate data of a total of 4096 pixels in the first partition are completely recorded in the first sub-matrix to form the pixel set matrices of each sub-area, completing the division and storage of pixel data.

[0021] The grayscale sequence extraction submodule calls each partition pixel set in the regional pixel set matrix, extracts and arranges the grayscale value sequence in row priority order, performs average difference operation on the grayscale value and the difference set between the grayscale values ​​of adjacent pixels, and calculates the regional grayscale mean difference; Taking the first partition as an example, the grayscale values ​​of all 4096 pixels in the area are extracted in row priority order, starting from the pixel value of the first row and the first column (coordinate (0,0)), and then extracted to the 64th row and the 64th column (coordinate (63,63)), generating a grayscale value sequence of length 4096. Taking actual data as an example, for example, the grayscale value sequence extracted from a certain soft tissue MRI image area is [42, 45, 48, …, 96]. Then, a difference operation is performed between each grayscale value in the sequence and the grayscale value of the adjacent pixel in the row direction. For example, the difference between the grayscale value 42 of pixel (0,0) and the grayscale value 45 of pixel (0,1) is 3. The difference sequence is recorded as [3, 3, 2, …, 1]. Subsequently, the average value of all values ​​in the difference sequence is calculated, such as (3+3+2+ …+1) / 4095=2.75. The grayscale mean difference of the first area is 2.75. Similarly, the same operation is performed on other areas, and finally the calculation of the grayscale mean difference of all areas is completed.

[0022] The edge pixel detection submodule detects edge pixels in multiple regions based on the regional grayscale mean difference and the pixel sets in multiple sub-regions, and identifies the pixel sets located at the edge boundary using the formula: ; Calculate the regional edge strength index, identify the edge pixels and calculate the edge strength of the entire region based on the gray value difference to obtain the medical image partition feature matrix; in, Representative The edge strength index of the image sub-region, Representative In the region The gray value of edge pixels, Represents the average gray value of edge pixels in the region. Represents the standard deviation of the grayscale values ​​of edge pixels, represents a constant used to avoid zero denominator, Represents the adjustment index that controls the expansion degree of the grayscale difference index. Representative The total number of region edge pixels, Represents the sub-region number, Represents the sequence number of edge pixels in the region.

[0023] First, the calculated grayscale mean difference of the first region (such as 2.75) is called, and then the pixels with significant grayscale value differences among the 4096 pixels in the first region are identified based on the grayscale values ​​of all pixels in the region. For example, the grayscale difference between each pixel and its 8 neighboring pixels is calculated and compared one by one, and pixels with grayscale differences greater than the set threshold are selected (the threshold setting method is to determine the upper limit of the 95% confidence interval after calculating the grayscale difference of all pixels. For example, the upper limit threshold of the 95% confidence interval of the grayscale difference in the first region in the actual data is 8). If the grayscale difference between a pixel and its neighbors exceeds 8, it is marked as an edge pixel. Assuming that a total of 80 edge pixels are identified in the first region, the grayscale values ​​of these edge pixels are recorded to form a set The grayscale value sequence of the edge pixels in the first region is obtained through actual measurement as [120, 122, ..., 135], and the average grayscale value of the sequence is calculated. , Standard Deviation . Then according to the given formula: ; The parameters in the formula are described as follows: For the The edge strength index of the region; For the edge pixel grayscale values, such as , , and so on; is the average grayscale value of edge pixels, which is 128 obtained by the above calculation; is the standard deviation of grayscale value, which is 4.2 as calculated above; To avoid a constant with a denominator of zero, 0.01 is taken; is the grayscale difference expansion adjustment index, and its value is 1.5 according to experience; is the total number of region edge pixels, which is 80 in this example.

[0024] The specific calculation process is as follows (only some pixel operations are listed): ; ; The results show that the edge strength index of region 1 is 2.32, which is within the effective range of 1.0 to 5.0 for the edge strength index of typical soft tissue MRI image partitions, meaning that the edge pixels in this region have a moderate grayscale difference intensity. This value will be used to generate the medical image partition feature matrix.

[0025] The benefit of the formula is that by using both the grayscale value mean difference and the standard deviation as normalization parameters, it can effectively alleviate the calculation error caused by abnormal fluctuations in grayscale values ​​and improve the sensitivity and accuracy of the edge strength index to the actual image boundary conditions.

[0026] To further clarify the implementation data, as shown in Table 1: Table 1 Grayscale statistics of edge pixels in MRI image area ; As shown in Table 1, taking the first area as an example, the edge strength index is calculated to be 2.32. This value and other areas together constitute the soft tissue MRI image feature matrix data for subsequent medical image analysis.

[0027] See also Figure 3 , the sparse feature localization module includes: The texture grayscale feature extraction submodule obtains the image grayscale distribution in the multi-partition area based on the medical image partition feature matrix, extracts the grayscale values ​​of all pixels in the partition, calculates the grayscale mean and grayscale variance, constructs a two-dimensional array to represent the multi-partition texture features in a combined manner, and establishes the image texture parameter matrix; Taking brain MRI images as an example, a single 256×256 pixel MRI image is taken as the analysis object. First, the image is divided into 4×4 sub-regions to form 16 partitions. The partition number j increases from 1 to 16 in sequence. The size of each partition is 64×64 pixels. The information of each partition area corresponds one-to-one with the corresponding index j in the matrix. The specific execution actions of the texture grayscale feature extraction process are as follows: First, traverse each sub-region one by one, extract the grayscale values ​​of all pixels in each partition, and count the grayscale value distribution pixel by pixel. Taking the 5th partition as an example, extract the grayscale values ​​of 4096 pixels in the region. Suppose the actual measured grayscale value array is: ; Calculate the grayscale mean of the partition by calling the array Grayscale variance The calculation process of grayscale mean is to call the formula , where n=4096, add up the 4096 grayscale values ​​and divide by 4096, as shown in the example: ; Grayscale variance is calculated by calling formula , that is, the sum of the squares of the difference between the grayscale of each pixel and the grayscale mean, and then divided by 4096 to get the grayscale variance, as shown in the example: The grayscale means and variances of all partitions are combined according to the partition numbers to construct the medical image texture parameter matrix in the form of a two-dimensional array. Finally, the grayscale mean of the fifth partition in the parameter matrix is ​​132.5 and the grayscale variance is 45.7.

[0028] The grayscale edge calculation submodule calls the grayscale mean and grayscale variance of the corresponding partition in the image texture parameter matrix, and calculates the grayscale change rate and edge density value of multiple partitions according to the grayscale change degree and the number of edge pixels, and uses the formula: ; The grayscale-edge comprehensive values ​​of multiple partitions are obtained by operation, and the grayscale-edge comprehensive matrix is ​​established in combination with the partition positions; in, Represents the grayscale-edge comprehensive value of the jth partition, represents the grayscale mean of the jth partition, Represents the grayscale mean of the entire image, represents the grayscale variance of the jth partition, represents the pixel density value at the edge of the j-th partition, represents the texture directional diffusion of the jth partition, where j represents the sub-region index; The specific calculation is to call the grayscale mean of all 16 partitions for calculation: , assuming that the grayscale mean of the whole image is 128.0. Then calculate the grayscale change rate of each partition separately. The specific operation is to call the grayscale mean of each partition and the grayscale mean of the whole image to calculate the absolute difference: Take the 5th partition as an example: . Then call the edge pixel density of each partition separately , actually measure the number of edge pixels in the partition and calculate the density value. For example, 328 edge pixels are detected in 4096 pixels in the 5th partition, then the edge density The calculation is 328 / 4096=0.0801; then call the partition texture direction diffusion , according to the consistency of pixel direction statistics based on texture direction, the actual measured texture direction diffusion of the 5th partition is 0.85. Then call the formula for calculation: ; Substituting the above actual data, the specific calculation process is: ; The grayscale-edge comprehensive value of the 5th partition is 8.86. All partitions are calculated in the same way to obtain the grayscale-edge comprehensive matrix, and the corresponding value of the 5th partition in the matrix is ​​8.86.

[0029] The texture missing screening submodule screens the partitions with comprehensive values ​​lower than the threshold value of 3.2 according to the grayscale edge comprehensive matrix, records the missing area location information according to the combination of partition number and coordinate information, and obtains the texture missing area location information set.

[0030] In order to compare the comprehensive values ​​of all partitions with the preset threshold value 3.2 one by one, the judgment standard is that when the comprehensive value is less than 3.2, it is marked as a texture missing area. Taking the 9th partition as an example, by calling the actual measured value 2.85, the numerical comparison action 2.85<3.2 is performed, and the 9th partition is determined to belong to the texture missing area. The partition number and position coordinate information are recorded, and the partition position coordinates are called as the upper left corner coordinates (128,128) and the lower right corner coordinates (192,192) of the 9th partition. This is recorded as the missing area information, and the texture missing area position information set is finally obtained. .

[0031] See also Figure 4 , the texture compensation generation module includes: The high-frequency extraction submodule locates the corresponding image area in the original image based on the texture missing area position information set, extracts the difference between the multi-pixel and the eight-neighborhood pixel mean in the regional pixel matrix, constructs a high-frequency distribution map in the region, and calculates the residual signal values ​​of multiple pixel points and stores the residual vector set to generate high-frequency residual signal values; First, based on the texture missing area location information set, determine the corresponding area in the original image. This process can be achieved through image processing technology. By scanning the texture missing area in the image, locate the specific pixel area, and then extract the pixel matrix of the area. In this part, the key operation is to calculate the mean difference between each multi-pixel and the eight neighboring pixels. This can be done by traversing each pixel in the target area, calculating the mean difference between it and the eight surrounding neighboring pixels, and then constructing a high-frequency distribution map of the area. For example, for a certain area in an image, assuming that the pixel matrix of the area is ,in and is the pixel coordinate of the image, and the mean difference between each pixel and its eight neighbors can be expressed by the formula: ; To calculate, Represents pixels The high-frequency distribution map is constructed by these difference values ​​to help identify the texture information of the area. In addition, the residual signal value of each multi-pixel point needs to be calculated, and a residual vector set is formed through these residual signal values ​​to finally generate a high-frequency residual signal value. For example, when calculating the residual of a certain pixel point, suppose the residual value of the point is calculated to be , then the residual vector set will contain all the calculated residual values ​​for that pixel.

[0032] The gradient fusion submodule calls the high-frequency residual signal value, performs gradient convolution processing in four directions of 0°, 45°, 90°, and 135°, extracts the gradient response image corresponding to multiple directions, calculates the contrast, directional consistency, and response intensity of each direction in the local area, and performs normalization processing based on the response value, using the formula: ; The fused texture enhancement matrix is ​​obtained by operation, and is superimposed pixel by pixel on the corresponding area of ​​the original image to obtain an enhanced superimposed pixel matrix; in, Indicates The gradient response matrix in each direction, Indicates The local contrast value in each direction, Indicates The gradient direction consistency coefficient in each direction, Indicates the original grayscale value of the corresponding pixel in the direction, Represents the grayscale mean of all pixels in the area, represents the normalized smoothing factor constant, Represents the texture enhancement matrix after normalization and fusion; First, you need to call the high-frequency residual signal values, which will be gradient convolved in four different directions, including 0°, 45°, 90°, and 135°. The gradient response image in each direction is composed of the convolution results in that direction. The calculation process first applies the gradient operation to the image in each direction. This process can be achieved by using specific gradient operators, for example, the Sobel operator, Prewitt operator, or Scharr operator, which can help calculate the gradient information of the image in a given direction. For each direction, the specific gradient response matrix can be obtained by convolving the image with the gradient operator in the corresponding direction.

[0033] Taking the 0° direction as an example, assuming that the image The gradient response matrix By using the gradient operator with the 0° direction Convolution can be expressed as: ; in is the gradient operator in the 0° direction, is the original image, Represents a convolution operation. The gradient response matrix It reflects the degree of change of the image in the 0° direction. It is usually a floating value, which indicates the texture characteristics of the area in this direction. For example, if the texture in the image changes greatly, the value of the gradient response will be higher; if the image texture is smooth, the gradient response will be lower.

[0034] Next, we need to calculate the local contrast in each direction , Directional consistency coefficient and response strength. These parameters help characterize the quality of image texture in each direction. Local contrast It can be obtained by calculating the grayscale standard deviation of the pixels in the target area. If the grayscale value of the area changes greatly and the standard deviation is high, it means that the texture features of the area are more obvious and the local contrast is high. Higher. Assume that the gray value of this area is , then the standard deviation The calculation formula is: ; in For the region The gray value of a pixel, is the grayscale mean of the area, is the number of pixels in the region. The calculated standard deviation reflects the contrast in the region. For the grayscale value above, the grayscale mean is: ; Then, calculate the standard deviation for: ; ; Therefore, the local contrast About 10.21.

[0035] Directional consistency coefficient It is used to measure the consistency between pixels in that direction, which is usually determined by the directional characteristics of the image gradient. If the image has strong texture consistency in a certain direction, the consistency coefficient in that direction Higher. Directional consistency can be measured by calculating the correlation between the gradient directions in that direction.

[0036] Then, the normalized texture enhancement matrix is ​​calculated using the following formula: : ; In this formula, and denote the local contrast and directional consistency coefficients respectively, It is the direction The original gray value, is the mean grayscale value of all pixels in the region, is the normalized smoothing factor constant. The purpose of the normalization process is to balance the response intensity in all directions so that the texture enhancement effect is more uniform. Assume ,and and The difference is small, so we can assume , ,but: ; First calculate each term: ; Then: ; Through this formula, the texture enhancement matrix can be obtained , this value reflects the degree of texture enhancement in the image area after gradient fusion.

[0037] The texture connectivity assessment submodule calls the enhanced superimposed pixel matrix to detect the grayscale mean, grayscale standard deviation and edge pixel gradient changes in the enhanced area and the neighborhood area, judge the pixel continuity and grayscale consistency of the enhanced area on the edge structure, and calculate the ratio of the edge connected area to the total area of ​​the enhanced area, and generate the image texture enhancement metric value in combination with the grayscale variance.

[0038] First, we need to analyze the grayscale mean and standard deviation of the target area and its neighboring areas. By calculating the mean and standard deviation of the grayscale values ​​of the pixels in the area, assuming that the grayscale mean of a certain area is , the standard deviation is , it can be determined whether the area has a relatively consistent texture.

[0039] Next, the gradient change of edge pixels is calculated to evaluate the continuity and grayscale consistency of edge structure pixels in the enhanced area. The continuity of edge pixels is determined by gradient analysis. If the gradient change of adjacent edge pixels is large, it indicates that the texture enhancement effect may not be significant. Finally, the ratio of edge connected areas to the total area of ​​the enhanced area is calculated to further analyze its impact. Assume that the ratio of edge connected areas to the total area of ​​the enhanced area is , indicating that the edge connectivity of the enhanced area is good.

[0040] See also Figure 5 , the boundary mutation recognition module includes: The texture enhancement recognition submodule performs dense scanning on the image pixel area based on the image texture enhancement metric value, obtains the gray value variation range in the adjacent area of ​​multiple pixel points in the image, calculates the gray value difference sequence of adjacent pixels, and determines the image mutation point according to the maximum difference position in the sequence, and generates a mutation gray point position set; First, take the position in the image to be tested as the starting point, for example, start from the upper left corner pixel of the image (such as coordinate (0,0)), and use the sliding window size as The pixel area moves to the right, scanning all pixel areas of the image row by row, and collecting the grayscale values ​​of all pixels in each area one by one. Assuming that the grayscale sequence data of adjacent pixels in a typical area (such as the area with pixel coordinates (20,30)) is: [132,130,127,115,111,108], the grayscale difference sequence will be calculated as [2,3,12,4,3] in sequence, and then the position of the maximum grayscale difference is determined by directly comparing the difference data in the sequence. In this example, the maximum difference is 12, which is located at the third position. The image position corresponding to the pixel coordinate (20,30) is the mutation grayscale point, which is recorded in the mutation grayscale point position set. This step is continuously performed on the entire image area, and finally the mutation grayscale point position set in the entire image range is obtained, such as [(20,30), (21,45), (50,60)].

[0041] The grayscale difference extraction submodule calls the mutation grayscale point position set, extracts the grayscale sequence of 16 pixels before and after each mutation point, obtains the mean and variance of the corresponding grayscale sequence, and calculates the grayscale mean difference and grayscale variance difference to form the mutation pixel grayscale change parameter set; Taking a typical mutation position (20,30) in the mutation gray point position concentration as an example, the gray sequence data of the 16 pixel positions before and after the point as the center are extracted respectively. For example, the gray sequence of the 16 pixels before the mutation position is [130,131,130,…,132], and the gray sequence of the 16 pixels after the mutation position is [120,122,118,…,119]. Then, the gray mean and variance of the front and back sequences are calculated respectively: the front sequence mean is calculated as (130+131+…+132) / 16=129.4, and the variance is 4.12; the back sequence mean is (120+122+…+119) / 16=121.3, and the variance is 3.56. Then, the gray mean difference is obtained by subtracting the two sets of sequence values. , grayscale variance difference , forming a set of mutation pixel grayscale change parameters, and recorded as [(8.1,0.56),…].

[0042] The potential boundary screening submodule determines whether the grayscale mean difference and grayscale variance difference meet the target threshold according to the grayscale change parameter set of the mutation pixel, using the formula: ; Calculate the boundary mutation significance value and compare it with the set significance threshold to establish the potential fracture boundary coordinate set; in, Representative position The boundary mutation significance value of Represents the mutation point position The difference between the mean values ​​of the grayscale sequences of the previous and next 16 pixels, Represents the mutation point position The difference between the variances of the grayscale sequences of the previous and next 16 pixels, Represents the mutation point position The minimum grayscale value in the first 16 pixel grayscale sequence, Represents the mutation point position The minimum grayscale value in the last 16 pixel grayscale sequence.

[0043] Taking the data (8.1, 0.56) in the sudden pixel grayscale change parameter set as an example, perform the following operation: Square it to get 65.61, The square is 0.3136, and the squares of the two items are added to get 65.9236. Then the minimum grayscale value of the first 16 pixel sequence is obtained, such as , the minimum grayscale value of the next 16 pixel sequence, such as , by taking the smaller value of the two, the minimum grayscale value is determined to be 105, and finally the formula is used for calculation: ; The obtained boundary mutation significance value 0.622 is compared with the preset significance threshold 0.5. The comparison process is clearly performed, that is, the significance value minus the threshold, 0.622-0.5=0.122. If it is greater than 0, it means that the condition is met, so the mutation position is determined to be a potential fracture boundary position, and the coordinates are recorded in the potential fracture boundary coordinate set, such as [(20,30),…].

[0044] Detailed description of the formula: In the formula, the parameters Represents the image position The boundary mutation significance value of , whose value directly determines whether the position is a potential boundary mutation point; It represents the mean difference of the grayscale sequence of 16 pixels before and after the mutation point, which is used to reflect the change of the average grayscale level at the mutation position. The larger the value, the more drastic the grayscale change. It represents the variance difference of the grayscale sequence of 16 pixels before and after the mutation point, reflecting the change in the uniformity of the local grayscale distribution; It is the minimum grayscale value in the grayscale sequence of 16 pixels before the mutation point, indicating the lowest brightness of the grayscale area before the mutation; It is the minimum grayscale value in the grayscale sequence of 16 pixels after the mutation point, indicating the lowest brightness of the grayscale area after the mutation. The square operation in the formula reflects the combined effect of the mean difference and variance difference on the mutation intensity. Dividing by (1 + minimum grayscale value) is used to suppress the misjudgment of small changes in low-brightness backgrounds, ensuring that the calculation of saliency can better match the actual image situation.

[0045] Table 2 Typical mutation position grayscale sequence and calculation results ; As shown in Table 2, taking three typical positions as examples, their grayscale mean difference, variance difference and minimum grayscale value are calculated respectively, and substituted into the significance formula to obtain the significance value of each position. Among them, the significance of position (21,45) obviously exceeds the threshold of 0.5, which is an effective boundary mutation point, while the significance value of position (50,60) is far below the threshold and is excluded. The above calculation results show that when the significance value When , the corresponding image position is determined as a potential fracture boundary, and the corresponding position is recorded in the coordinate set; otherwise, it is not included in the candidate range, thereby realizing accurate identification and screening of the mutation boundary position.

[0046] See also Figure 6 , the fracture continuous filling module includes: The intersection grayscale extraction submodule calls the potential fracture boundary coordinate set, extracts the grayscale sequence on both sides of the mutation point, determines the grayscale value interval of the same pixel position in the sequence based on the corresponding relationship of the pixel position, screens the grayscale intersection area to calculate the mean, and obtains the intersection grayscale mean; During the execution process, the row and column data of several pixels before and after the mutation point are first extracted, and then the grayscale value of each position is determined according to the position information of the pixel. On this basis, according to the corresponding relationship of the pixel position, it is determined whether the grayscale values ​​of these positions fall in the same grayscale interval. If their grayscale values ​​fall in the same grayscale interval, the grayscale value of the position is considered to be relatively stable. If the grayscale value exceeds this interval, the system will judge it as a grayscale mutation point, and further screening will be performed. Next, the intersection grayscale area is screened out, that is, the intersection of the grayscale value intervals on both sides of the mutation point, and the average grayscale value of these areas is calculated and output as the intersection grayscale mean. For example, assuming that the grayscale sequence around the mutation point extracted in a certain image area is [120, 125, 130, 135, 140], and the grayscale sequence interval of the mutation points on both sides is [125, 135], then the intersection grayscale mean is the grayscale value in this interval (for example, it can be calculated by the mean value, and the grayscale mean of this area is 130). The grayscale mean of the intersection will provide data support for the subsequent filling mutation point operation.

[0047] The mutation point filling submodule locates the mutation point coordinates in the image according to the intersection grayscale mean, performs grayscale filling processing on the mutation point pixels using the intersection grayscale mean, and outputs the mutation point filling result layer to generate a mutation point grayscale value set; During this process, the image processing algorithm will find the location of each potential mutation point and fill the grayscale of the pixels of these mutation points based on the previously calculated intersection grayscale mean. For example, if a mutation point is located at a certain coordinate position of the image (such as (5,5)), the grayscale value of the point will be adjusted to the intersection grayscale mean of 130. This processing not only smoothes the grayscale changes in the image, but also alleviates the mutations at the edge of the image, improving the overall visual effect of the image. After the filling is completed, the system will generate a new image containing the filled mutation points and output the layer as the result, forming a set of mutation point grayscale values.

[0048] The linear interpolation construction submodule is based on the gray value set of the mutation point. It selects 4 pixels before and after the mutation point to construct a gray sequence, calculates the gray difference of adjacent pixels and the pixel spacing, and uses the formula: ; Calculate the smooth transition grayscale value, embed the linear interpolation grayscale sequence into the corresponding area of ​​the image, and establish a grayscale continuous filling sequence; in, Represents a smooth transition gray value, Represents the grayscale pixel value before and after the mutation point, Represents the mutation point and The pixel distance of pixels represents the grayscale value generated by the interpolated pixel position. Represents the number of grayscale participation points before and after, Represents the number of interpolation sequences, s and t are index variables, and r is the current interpolation position.

[0049] First, the linear interpolation construction submodule selects 4 pixels before and after the mutation point to construct a grayscale sequence based on the grayscale value set of the mutation point. For these pixels, their grayscale values ​​and positions (i.e., the distance between pixels) will be involved in the calculation to ensure a smooth grayscale transition of the mutation point. Specifically, assuming that the mutation point is located at a certain position in the image (e.g., position (5,5)), and 4 pixel positions before and after this position are selected, that is, the grayscale values ​​of the 4 pixels before and after the mutation point position are selected.

[0050] For example, suppose the grayscale values ​​of the four pixels before and after the mutation point are: First 4 pixels: ; The last 4 pixels: ; These grayscale values ​​constitute a grayscale sequence near the mutation point.

[0051] At the same time, the pixel spacing (i.e. the distance between the mutation point and these pixels) is determined according to the position. Assume that the distance between these pixels (pixel spacing) is as follows: First 4 pixels: (meaning the distance between the first pixel and the mutation point is 1 pixel, the second is 2 pixels, and so on); The last 4 pixels: (Indicates the distance between subsequent pixels and the mutation point, arranged according to the actual image); Next, you need to calculate the grayscale difference between adjacent pixels (that is, the difference between the grayscale values ​​of the previous and next pixels): For the first 4 pixels, the grayscale difference is: ; ; ; For the last 4 pixels, the grayscale difference is: ; ; ; These grayscale differences represent the degree of grayscale change between adjacent pixels.

[0052] The purpose of linear interpolation is to smooth the transition of grayscale values ​​through a certain calculation formula, making the image smoother and preventing sudden changes. According to the formula: ; in, is the smooth transition grayscale value we need to calculate, is the difference in grayscale value before and after the mutation point, is the distance between each pixel and the mutation point, is the grayscale value generated by the interpolation position, and are the number of front and back grayscale points and interpolation sequences respectively.

[0053] For the grayscale values ​​before and after and the pixel spacing, we combine the calculated grayscale difference with the pixel spacing and perform a weighted sum. For example, we combine the grayscale difference of the first 4 pixels (such as 10, 5, 10) with the corresponding pixel spacing (such as 1, 2, 1) for weighted calculation.

[0054] First, calculate each term: - For the first pixel, calculate as -For the second pixel, the calculation is -For the third pixel, the calculation is ; Next, similar calculations are performed on the grayscale difference and pixel spacing of the next four pixels: - For the first pixel, the calculation is -For the second pixel, the calculation is -For the third pixel, the calculation is ; Summing these calculated values ​​yields the weighted sum of the interpolated values: The weighted sum of the first 4 pixels is ; Next, we calculate the interpolation weight for each pixel. The interpolation weight for each pixel can be calculated by adding a constant 1 to the pixel spacing: The sum of the weights of the first 4 pixels is -The sum of the weights of the last 4 pixels is ; Then, substitute the formula to calculate the gray value of smooth transition; Substitute the calculated result into the formula for interpolation: ; First calculate: ; Finally, we get: ; Therefore, the calculated smooth transition gray value is 20.26, which will be used as the new gray value of the interpolation area.

[0055] The system will embed the calculated smooth transition grayscale value into the corresponding area in the image to ensure smooth transition of grayscale and avoid mutation. This interpolation process can eliminate the obvious grayscale difference caused by mutation points in the image, making the image smoother and more natural.

[0056] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A medical image quality enhancement system based on big data, characterized in that: The system comprises: The sparse feature location module calculates the product of grayscale change rate and edge density based on the medical image partition feature matrix, selects partitions with a product lower than 3.2, records the texture missing area location information, and generates a texture missing area location information set; The texture compensation generation module extracts the high-frequency residual signal image of the corresponding area according to the texture missing area position information set, performs gradient convolution at 0°, 45°, 90°, and 135°, generates a texture enhancement matrix by normalized weighted fusion, superimposes the corresponding pixels of the original image, detects the grayscale distribution consistency and edge connectivity, and generates an image texture enhancement metric value; The boundary mutation recognition module scans the pixel grayscale difference based on the image texture enhancement measurement value, records the mutation position, extracts the grayscale sequence of 16 pixels before and after, calculates the mean difference and the variance difference, screens the potential fracture boundary, and generates a potential fracture boundary coordinate set; The fracture continuous filling module calls the potential fracture boundary coordinate set, extracts the grayscale sequences on both sides of the mutation point, calculates the grayscale range of the intersection, calculates the intersection mean to fill the mutation point, selects 4 pixels before and after the mutation point, performs linear grayscale interpolation to generate a smooth gradient transition, replaces the pixel values ​​of the corresponding area of ​​the image, and establishes a grayscale continuous filling sequence.

2. The medical image quality enhancement system based on big data according to claim 1, characterized in that: The texture missing area location information set specifically includes texture sparse distribution points, structural discontinuity areas, and grayscale sudden drop areas. The image texture enhancement metric value includes local texture consistency score, edge structure continuity score, and enhanced area coverage ratio. The potential fracture boundary coordinate set includes grayscale mutation points, edge structure fracture lines, and abnormal texture turning zones. The grayscale continuous filling sequence specifically refers to the smooth grayscale layer of the fracture area, the pixel value gradient transition zone, and the texture connection completion segment.

3. The medical image quality enhancement system based on big data according to claim 2, characterized in that: The system further comprises: The image partition analysis module obtains the soft tissue MRI image pixel array, divides the image into 64×64 partitions, extracts the regional grayscale sequence and the mean of adjacent differences, counts the edge pixel ratio, and establishes the medical image partition feature matrix; The medical image partition feature matrix includes regional grayscale balance, edge texture density, and regional grayscale difference; The image partition analysis module includes: The image partitioning submodule obtains the image boundary range and pixel distribution density based on the soft tissue MRI image pixel array, performs partition operation on the image array according to the 64×64 pixel size, assigns partition numbers according to the coordinate area and locates the center point of the area, extracts the pixel point set in the corresponding area, and generates the regional pixel set matrix; The grayscale sequence extraction submodule calls each sub-region pixel set in the regional pixel set matrix, extracts and arranges the grayscale value sequence in row priority order, performs average difference operation on the grayscale value and the difference set between the grayscale values ​​of adjacent pixels, and calculates the regional grayscale mean difference; The edge pixel detection submodule detects edge pixels in multiple regions based on the regional grayscale mean difference and the pixel sets in multiple sub-regions, and identifies the pixel sets located at the edge boundary, using the formula: ； Calculate the regional edge strength index, identify the edge pixels and calculate the edge strength of the entire region based on the gray value difference to obtain the medical image partition feature matrix; in, Representative The edge strength index of the image sub-region, Representative In the region The gray value of edge pixels, Represents the average gray value of edge pixels in the region. Represents the standard deviation of the grayscale values ​​of edge pixels, represents a constant used to avoid zero denominator, Represents the adjustment index that controls the expansion degree of the grayscale difference index. Representative The total number of region edge pixels, Represents the sub-region number, Represents the sequence number of edge pixels in the region.

4. The medical image quality enhancement system based on big data according to claim 3 is characterized in that: The sparse feature positioning module includes: The texture grayscale feature extraction submodule obtains the image grayscale distribution in the multi-partition area based on the medical image partition feature matrix, calculates the grayscale mean and grayscale variance after extracting the grayscale values ​​of all pixels in the partition, constructs a two-dimensional array to represent the multi-partition texture features in a combined manner, and establishes an image texture parameter matrix; The grayscale edge calculation submodule calls the grayscale mean and grayscale variance of the corresponding partition in the image texture parameter matrix, and calculates the grayscale change rate and edge density value of multiple partitions according to the grayscale change degree and the number of edge pixels, and uses the formula: ； The grayscale-edge comprehensive values ​​of multiple partitions are obtained by operation, and the grayscale-edge comprehensive matrix is ​​established in combination with the partition positions; in, Represents the grayscale-edge comprehensive value of the jth partition, represents the grayscale mean of the jth partition, Represents the grayscale mean of the entire image, represents the grayscale variance of the jth partition, represents the pixel density value at the edge of the j-th partition, represents the texture directional diffusion of the jth partition, where j represents the sub-region index; The texture missing screening submodule screens the partitions with comprehensive values ​​lower than the threshold value 3.2 according to the grayscale edge comprehensiveness matrix, records the missing area position information according to the combination of the partition number and the coordinate information, and obtains the texture missing area position information set.

5. The medical image quality enhancement system based on big data according to claim 4, characterized in that: The texture compensation generation module comprises: The high-frequency extraction submodule locates the corresponding image area in the original image based on the texture missing area position information set, extracts the difference between the multi-pixel and the eight-neighborhood pixel mean in the regional pixel matrix, constructs a high-frequency distribution map in the region, and simultaneously calculates the residual signal values ​​of the multi-pixel points and stores the residual vector set to generate a high-frequency residual signal value; The gradient fusion submodule calls the high-frequency residual signal value, performs gradient convolution processing in four directions of 0°, 45°, 90°, and 135°, extracts the gradient response image corresponding to multiple directions, calculates the contrast, directional consistency, and response intensity of each direction in the local area, and performs normalization processing based on the response value, using the formula: ； The fused texture enhancement matrix is ​​obtained by operation, and is superimposed pixel by pixel on the corresponding area of ​​the original image to obtain an enhanced superimposed pixel matrix; in, Indicates The gradient response matrix in each direction, Indicates The local contrast value in each direction, Indicates The gradient direction consistency coefficient in each direction, Indicates the original grayscale value of the corresponding pixel in the direction, Represents the grayscale mean of all pixels in the area. represents the normalized smoothing factor constant, Represents the texture enhancement matrix after normalization and fusion; The texture connectivity assessment submodule calls the enhanced superimposed pixel matrix to detect the grayscale mean, grayscale standard deviation and edge pixel gradient changes of the enhanced area and the neighborhood area, judge the pixel continuity and grayscale consistency of the edge structure of the enhanced area, and calculate the ratio of the edge connected area to the total area of ​​the enhanced area, and generate the image texture enhancement metric value in combination with the grayscale variance.

6. The medical image quality enhancement system based on big data according to claim 5, characterized in that: The boundary mutation recognition module includes: The texture enhancement recognition submodule performs intensive scanning of the image pixel area based on the image texture enhancement metric value, obtains the gray value variation range in the adjacent area of ​​multiple pixel points in the image, calculates the gray value difference sequence of adjacent pixels, and determines the image mutation point according to the maximum difference position in the sequence, and generates a mutation gray point position set; The grayscale difference extraction submodule calls the mutation grayscale point position set, extracts the grayscale sequence of 16 pixels before and after each mutation point, obtains the mean and variance of the corresponding grayscale sequence, and simultaneously calculates the grayscale mean difference and grayscale variance difference to form a mutation pixel grayscale change parameter set; The potential boundary screening submodule determines whether the grayscale mean difference and grayscale variance difference meet the target threshold according to the grayscale change parameter set of the mutation pixel, using the formula: ； Calculate the boundary mutation significance value and compare it with the set significance threshold to establish the potential fracture boundary coordinate set; in, Representative position The boundary mutation significance value of Represents the mutation point position The difference between the mean values ​​of the grayscale sequences of the previous and next 16 pixels, Represents the mutation point position The difference between the variances of the grayscale sequences of the previous and next 16 pixels, Represents the mutation point position The minimum grayscale value in the first 16 pixel grayscale sequence, Represents the mutation point position The minimum grayscale value in the last 16 pixel grayscale sequence.

7. The medical image quality enhancement system based on big data according to claim 6, characterized in that: The fracture continuous filling module comprises: The intersection grayscale extraction submodule calls the potential fracture boundary coordinate set, extracts the grayscale sequence on both sides of the mutation point, determines the grayscale value interval of the same pixel position in the sequence according to the corresponding relationship of the pixel position, screens the grayscale intersection area to calculate the mean, and obtains the intersection grayscale mean; The mutation point filling submodule locates the mutation point coordinate position in the image according to the intersection grayscale mean, performs grayscale filling processing on the mutation point pixels using the intersection grayscale mean, and outputs the mutation point filling result layer to generate a mutation point grayscale value set; The linear interpolation construction submodule selects 4 pixels before and after the mutation point to construct a grayscale sequence based on the grayscale value set of the mutation point, calculates the grayscale difference of adjacent pixels and the pixel spacing, and uses the formula: ； Calculate the smooth transition grayscale value, embed the linear interpolation grayscale sequence into the corresponding area of ​​the image, and establish a grayscale continuous filling sequence; in, Represents a smooth transition gray value, Represents the grayscale pixel value before and after the mutation point, Represents the mutation point and The pixel distance is pixels, represents the grayscale value generated by the interpolated pixel position, Represents the number of grayscale participation points before and after, Represents the number of interpolation sequences, s and t are index variables, and r is the current interpolation position.

Images