A skeletal muscle ultrasonic image processing system and method for ultrasonic detection

Abstract

The present invention discloses a skeletal muscle ultrasound image processing system and method for ultrasonic testing, belonging to the technical field of medical ultrasound image processing. The system comprises: after denoising the skeletal muscle ultrasound image using non-local mean dynamic filtering, enhancing the contrast of the skeletal muscle ultrasound image through nonlinear grayscale transformation, and segmenting the skeletal muscle region using the spatial continuity and similarity of the skeletal muscle; based on the segmented skeletal muscle region, binarizing the skeletal muscle ultrasound image, calculating the fractal dimension using grid sliding statistics and weighted processing, constructing and smoothing a tensor model, calculating the main eigenvalue and eigenvector, and locating the skeletal muscle atrophy area through threshold judgment and cluster analysis; and performing post-annotation processing on the skeletal muscle ultrasound image to generate a skeletal muscle ultrasound image annotation report. The system aims to solve technical problems such as denoising, contrast enhancement, region segmentation, texture complexity quantification, atrophy area location, and annotation report generation in skeletal muscle ultrasound images.

Description

Technical Field

[0001] The present invention relates to the technical field of medical ultrasonic image processing, and more particularly to a skeletal muscle ultrasonic image processing system and method for ultrasonic detection. Background Art

[0002] Skeletal muscle atrophy is a common medical condition with complex causes, encompassing neurogenic factors such as motor neuron disease and peripheral nerve damage; myogenic factors such as muscular dystrophy; and disuse factors such as prolonged bed rest or limb immobilization. Muscle atrophy not only severely impairs patients' mobility and quality of life but can also lead to a host of complications, notably changes in bone density. When skeletal muscle atrophy occurs, the mechanical stress on the bones is significantly reduced. Normal muscle contraction and relaxation exert certain pressures and tensions on bones, and this mechanical stimulation is crucial for maintaining bone density. Once muscle atrophy occurs, the bones lack sufficient mechanical stimulation, leading to increased osteoclast activity and suppressed osteoblast activity. This results in greater bone resorption than formation, gradual bone loss, and consequently, decreased bone density. Clinical studies have shown that patients with skeletal muscle atrophy due to prolonged bed rest may experience significant bone density loss within weeks, particularly in weight-bearing bones such as the hip and spine. Furthermore, decreased bone density further increases the risk of fractures. Therefore, accurate and timely detection and assessment of skeletal muscle atrophy is crucial for early diagnosis, treatment planning, and disease monitoring. Ultrasound testing technology, with its significant advantages such as ease of use, lack of radiation, and real-time dynamic observation, has been widely used in the diagnosis of skeletal muscle diseases. Ultrasound testing can capture information such as skeletal muscle morphology, structure, and echogenicity, providing a crucial basis for the diagnosis of muscle atrophy.

[0003] However, existing techniques for assessing skeletal muscle morphology, echogenicity, and texture using skeletal muscle ultrasound imaging rely heavily on the operator's experience and subjective judgment. Different doctors or technicians may reach different conclusions, resulting in a lack of a standardized assessment system, which impacts diagnostic accuracy and consistency. Furthermore, skeletal muscle image processing typically relies solely on single denoising and enhancement methods, which struggle to effectively address complex noise interference and insufficient image contrast. Traditional denoising methods, while removing noise, can easily blur detailed muscle features, resulting in loss of important information. Simple linear grayscale transformations have limited effect on enhancing image contrast, failing to highlight key muscle features. Regarding skeletal muscle segmentation, existing segmentation algorithms often struggle to accurately distinguish skeletal muscle from surrounding tissue, especially when muscle boundaries are blurred and image quality is poor due to muscle atrophy. Segmentation accuracy is low, impacting the accuracy of subsequent analysis. Quantitative assessment of skeletal muscle lacks comprehensive, objective quantitative indicators and effective analytical methods. Currently, methods rely primarily on the measurement of simple geometric parameters such as muscle thickness and cross-sectional area, often performed manually or semi-automatically, subject to significant human error. At the same time, these parameters cannot fully reflect the microstructural and functional changes within muscles, and lack a quantitative description of the complexity and irregularity of muscle texture. When assessing pathologies such as muscle atrophy, there is a lack of effective means to accurately locate the affected area and quantify the extent of the disease, making it difficult to meet the needs of clinical diagnosis and disease monitoring. Therefore, to overcome these limitations, the present invention proposes a skeletal muscle ultrasonic image processing system and method for ultrasonic testing. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide a skeletal muscle ultrasound image processing system and method for ultrasonic detection, which solves the technical problems of skeletal muscle ultrasound image denoising, contrast enhancement, region segmentation, texture complexity quantification, atrophy area positioning and automatic labeling, and improves the analysis accuracy and diagnostic efficiency of skeletal muscle ultrasound images.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A skeletal muscle ultrasonic image processing system for ultrasonic testing, comprising an image preprocessing module, a quantitative analysis module, and an intelligent labeling module;

[0007] The image preprocessing module is used to denoise the skeletal muscle ultrasound image using non-local mean dynamic filtering, enhance the contrast of the denoised skeletal muscle ultrasound image through nonlinear grayscale transformation, and segment the skeletal muscle area in the skeletal muscle ultrasound image using the spatial continuity and similarity of the skeletal muscle;

[0008] The quantitative analysis module is used to quantify the complexity and irregularity of skeletal muscle texture based on the segmented skeletal muscle area. By binarizing the skeletal muscle ultrasound image and calculating the fractal dimension using grid sliding statistics and weighted processing, the module also calculates the gradient amplitude, gradient direction, and texture direction of the skeletal muscle area, constructs and smoothes the tensor model, calculates the main eigenvalue and eigenvector, and locates the skeletal muscle atrophy area through threshold judgment and cluster analysis.

[0009] The intelligent annotation module is used to perform annotation and post-processing on skeletal muscle ultrasound images, highlight the skeletal muscle area in the skeletal muscle ultrasound image, mark the fractal dimension of the skeletal muscle area, and locally magnify the skeletal muscle atrophy area to generate a skeletal muscle ultrasound image annotation report.

[0010] Specifically, the specific steps of non-local mean dynamic filtering denoising include:

[0011] Obtain skeletal muscle ultrasound images, use image windows to mark suspected skeletal muscle atrophy areas, and divide the skeletal muscle ultrasound images into suspected atrophy areas and non-suspected atrophy areas;

[0012] An initial search window and an initial neighborhood window are set for the skeletal muscle ultrasound image. When each pixel in the skeletal muscle ultrasound image is traversed one by one, if the current pixel is in a non-suspected atrophy area, the initial neighborhood window of the current pixel is used as a template to compare with the initial neighborhood windows of other pixels in the initial search window. The similarity weight between the current pixel and other pixels in the initial search window is obtained by calculating the difference in the pixel grayscale values ​​in the two initial neighborhood windows.

[0013] The grayscale values ​​of all pixels in the initial search window are weighted averaged according to their respective similarity weights to obtain the filtered grayscale value of the current pixel.

[0014] Specifically, the specific steps of non-local mean dynamic filtering denoising also include:

[0015] When the traversed pixel point is located in the suspected atrophy area, the initial search window and the initial neighborhood window are dynamically adjusted, and the initial search window and the initial neighborhood window are expanded to serve as the second search window and the second neighborhood window;

[0016] In the second search window, the second neighborhood window of the current pixel is used as a template and compared with the second neighborhood windows of other pixels, the difference in the grayscale values ​​of the pixels in the two second neighborhood windows is calculated, and the similarity of the texture features of the two second neighborhood windows is calculated; the difference in grayscale values ​​and the similarity of texture features are combined to obtain the similarity weight between the current pixel and other pixels in the second search window;

[0017] The grayscale values ​​of all pixels in the second search window are weighted averaged according to their respective similarity weights, and then weighted twice according to their spatial distance from the current pixel to obtain the filtered grayscale value of the current pixel;

[0018] All pixel points of the skeletal muscle ultrasound image are traversed to obtain the skeletal muscle ultrasound image after non-local mean dynamic filtering and denoising.

[0019] Specifically, the specific steps of nonlinear grayscale transformation enhancement include:

[0020] Obtain skeletal muscle ultrasound images after non-local mean dynamic filtering and denoising, and use logarithmic transformation function to perform nonlinear grayscale transformation;

[0021] Before nonlinear grayscale transformation, feature analysis was performed on the denoised skeletal muscle ultrasound images to obtain grayscale features, including grayscale value concentration range, grayscale value dispersion, and image contrast.

[0022] Setting an initial range of grayscale transformation parameters, selecting multiple representative parameter values ​​within the initial range of grayscale transformation parameters by an equidistance method, performing nonlinear grayscale transformation on the denoised skeletal muscle ultrasound image according to the representative parameter values, obtaining the grayscale features of the denoised skeletal muscle ultrasound image after the nonlinear grayscale transformation, narrowing the initial range of grayscale transformation parameters, and obtaining a fine-tuning range of grayscale transformation parameters;

[0023] The data interval is set, and within the grayscale transformation parameter fine-tuning range, a series of grayscale transformation fine-tuning parameter values ​​are generated according to the data interval. Each fine-tuning parameter value is used again to perform nonlinear grayscale transformation on the denoised skeletal muscle ultrasound image, and the structural similarity index and peak signal-to-noise ratio of the transformed skeletal muscle ultrasound image are calculated. The fine-tuning parameter value that achieves the optimal combination of the structural similarity index and peak signal-to-noise ratio is selected as the final logarithmic transformation function parameter value;

[0024] The target logarithmic transformation function is determined according to the parameter value of the logarithmic transformation function. The grayscale value of each pixel in the denoised skeletal muscle ultrasound image is substituted into the selected target logarithmic transformation function, and the transformed grayscale value is calculated to obtain the skeletal muscle ultrasound image with enhanced contrast.

[0025] Specifically, the specific steps for segmenting the skeletal muscle region include:

[0026] Obtain contrast-enhanced skeletal muscle ultrasound images and initialize skeletal muscle segmentation seed points;

[0027] Starting from the seed point, region growing is performed based on the spatial continuity and similarity features of skeletal muscle. The similarity features include texture similarity features and grayscale similarity features. The threshold range of the similarity features is set. If the similarity features of the adjacent pixels of the seed point are all within the threshold range, the adjacent pixels are added to the current region.

[0028] After the region growing is completed, for every two regions, the grayscale mean difference, texture feature distance and property feature difference are calculated to construct a multi-level feature. By calculating the multi-level feature similarity between every two regions and setting a similarity threshold, the regions with multi-level feature similarity greater than the similarity threshold are merged to preliminarily segment the skeletal muscle region.

[0029] For the skeletal muscle region after preliminary segmentation, morphological operations are used to refine and repair the muscle edges to obtain the final skeletal muscle region.

[0030] Specifically, the steps for calculating the fractal dimension of the skeletal muscle region include:

[0031] The segmented skeletal muscle ultrasound image is converted into a binary image to distinguish the skeletal muscle area from the background area, where white pixels represent the skeletal muscle area and black pixels represent the background area;

[0032] Set the initial mesh side length based on the side length of the smallest rectangular area where the skeletal muscle area is located;

[0033] Set the ending mesh length and create an empty list to store mesh lengths of different sizes. Starting from the initial mesh length, the mesh length is halved each time it is iterated until the iterative mesh length is less than the ending mesh length. Then, a list of mesh lengths of different sizes is obtained.

[0034] For each size of grid side length, a grid is slid on the binary image in the form of a grid, and two layers of nested loops are used to control the horizontal and vertical positions of the grid respectively;

[0035] For each grid position, determine whether the grid partially covers the skeletal muscle area, i.e., the white pixel area. If so, use the pixel counting method to calculate the number of skeletal muscle area pixels within the grid position and divide it by the total area of ​​the grid to obtain the coverage ratio;

[0036] Configure a ratio threshold. If the coverage ratio is greater than the ratio threshold, the grid is marked as covered. The total number of covered grids at each grid size is counted.

[0037] The total number of covered grids under different grid sizes was weighted using an exponential decay function. A scatter plot was drawn with the logarithm of the grid side length as the horizontal axis and the logarithm of the weighted total number of covered grids as the vertical axis. The scatter plot was then linearly fitted to obtain the slope of the fitting line. The absolute value of the slope of the fitting line was used as the fractal dimension of the skeletal muscle region.

[0038] Specifically, the specific steps for locating the area of ​​skeletal muscle atrophy include:

[0039] For the segmented skeletal muscle region, the gradient operator is used to calculate the gradient of the skeletal muscle region in the horizontal and vertical directions, and the gradient amplitude and gradient direction are obtained. The gray-level co-occurrence matrix is ​​used to obtain the texture direction of the skeletal muscle region.

[0040] Initialize the tensor and define a second-order tensor for each pixel in the skeletal muscle area. The tensor elements are initialized to zero, and the dimension of the tensor is consistent with the dimension of the skeletal muscle area.

[0041] Fill each tensor element according to the calculated gradient magnitude, gradient direction, and texture direction. For the horizontal gradient magnitude, normalize it and use it as a diagonal element of the tensor. Normalize the vertical gradient magnitude and use it as another diagonal element of the tensor.

[0042] Calculate the difference angle between the gradient direction and the texture direction, and use the trigonometric function relationship to determine the value of the cross term element of the tensor element;

[0043] After smoothing the constructed tensor, the sliding window technology is used for localized analysis. The sliding window is set, and for each set of tensors within the sliding window, its main eigenvalue and corresponding eigenvector are calculated.

[0044] Specifically, the specific steps of locating the skeletal muscle atrophy area also include:

[0045] Set the feature difference threshold and angle threshold. When the sliding window traverses the skeletal muscle area, for each sliding window, calculate the main eigenvalue difference between it and the adjacent sliding window and the angle between the corresponding eigenvectors. If the main eigenvalue difference is greater than the feature difference threshold and the angle between the eigenvectors is greater than the angle threshold, then mark the current sliding window as a potential atrophy area and record its location information.

[0046] The main eigenvalues ​​and eigenvectors extracted from all sliding windows marked as potential atrophy areas are combined into a set of eigenvectors, each eigenvector containing the main eigenvalue and eigenvector of the sliding window;

[0047] Set the number of cluster categories to represent normal areas and atrophic areas respectively, and randomly select the feature vectors in the feature vector set as the initial cluster center;

[0048] Iterate. In each iteration, for each feature vector in the feature vector set, calculate its Euclidean distance with the cluster center and assign it to the cluster with the nearest cluster center.

[0049] Recalculate the cluster center of each cluster, that is, the mean of all eigenvectors in the cluster, and repeat the iterative steps until the cluster center position of each cluster converges;

[0050] After clustering is completed, the eigenvalue ranges and eigenvector directions of the two clusters are obtained respectively to locate the skeletal muscle atrophy region clusters.

[0051] Specifically, the steps for calculating the main eigenvalue and the corresponding eigenvector include:

[0052] For each set of tensors in the sliding window, randomly select an initial vector and make its dimension consistent with the dimension of the tensor;

[0053] Iterate, and in the ηth iteration, calculate u η =Tv η-1 , where T is the matrix form corresponding to the tensor set in the sliding window, that is, the tensor matrix, v η-1 is the principal eigenvector of the η-1th iteration, u η is the intermediate eigenvector of the ηth iteration, for u η Normalize and get the main eigenvector v of the ηth iteration η ,Right now

[0054] Set the convergence threshold to determine whether the iteration has converged. When the difference between the eigenvectors in two adjacent iterations is less than the convergence threshold, the iteration has converged, stop the iteration, and obtain the main eigenvector v after convergence. o , otherwise continue to iterate; when the main eigenvector v after convergence is obtained o Then, calculate the main eigenvalue λ1 of the main eigenvector, the formula is in, is the principal eigenvector v after convergence o The transposed vector of .

[0055] A method for processing skeletal muscle ultrasound images for ultrasound detection comprises the following steps:

[0056] Step S1: After denoising the skeletal muscle ultrasound image using non-local mean dynamic filtering, the contrast of the denoised skeletal muscle ultrasound image is enhanced by nonlinear grayscale transformation, and the skeletal muscle region in the skeletal muscle ultrasound image is segmented using the spatial continuity and similarity of the skeletal muscle;

[0057] Step S2: Based on the segmented skeletal muscle region, the skeletal muscle ultrasound image is binarized, and the fractal dimension is calculated using grid sliding statistics and weighted processing to quantify the complexity and irregularity of the skeletal muscle texture. The gradient amplitude, gradient direction, and texture direction of the skeletal muscle region are calculated, a tensor model is constructed and smoothed, and the main eigenvalues ​​and eigenvectors are calculated. The skeletal muscle atrophy region is located through threshold judgment and cluster analysis.

[0058] Step S3: perform post-annotation processing on the skeletal muscle ultrasound image, highlight the skeletal muscle area in the skeletal muscle ultrasound image, mark the fractal dimension of the skeletal muscle area, and locally magnify the skeletal muscle atrophy area to generate a skeletal muscle ultrasound image annotation report.

[0059] Beneficial effects of the present invention:

[0060] A skeletal muscle ultrasound image processing system and method for ultrasound testing uses non-local mean dynamic filtering denoising to accurately remove noise, retain subtle features, enhance contrast through nonlinear grayscale transformation, and use spatial continuity and similarity to accurately segment skeletal muscle areas, laying a solid foundation for subsequent analysis; in the quantitative analysis link, fractal dimensions are calculated to quantify texture features, and a tensor model is constructed and optimized to locate atrophic areas, comprehensively displaying muscle changes in different dimensions; an intelligent labeling module highlights skeletal muscle areas, annotates fractal dimensions, and generates reports by locally zooming in on atrophic areas, making key information clear at a glance; the various modules of the system work together and are highly integrated, significantly improving the efficiency and accuracy of skeletal muscle ultrasound image analysis, and providing comprehensive and powerful support for doctors to accurately diagnose muscle atrophy and formulate treatment plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 This is a schematic structural diagram of a skeletal muscle ultrasonic image processing system for ultrasonic testing according to the present invention;

[0062] Figure 2 This is a flowchart of the specific steps of the non-local mean dynamic filtering denoising method of the present invention;

[0063] Figure 3 Flowchart of the step of selecting the nonlinear grayscale transformation enhancement parameters of the present invention;

[0064] Figure 4 A schematic diagram of the nonlinear grayscale transformation enhancement parameter selection of the present invention;

[0065] Figure 5 Flowchart of the specific steps of skeletal muscle region segmentation of the present invention;

[0066] Figure 6 A flowchart of the specific steps for locating skeletal muscle atrophy areas according to the present invention;

[0067] Figure 7 The present invention is a flowchart of the steps of a skeletal muscle ultrasonic image processing method for ultrasonic testing. DETAILED DESCRIPTION

[0068] Example 1

[0069] See also Figure 1 ,This embodiment introduces a skeletal muscle ultrasound image processing system for ultrasonic detection, including an image preprocessing module, a quantitative analysis module and an intelligent labeling module;

[0070] The image preprocessing module is used to denoise the skeletal muscle ultrasound image using non-local mean dynamic filtering, enhance the contrast of the denoised skeletal muscle ultrasound image through nonlinear grayscale transformation, and segment the skeletal muscle area in the skeletal muscle ultrasound image using the spatial continuity and similarity of the skeletal muscle;

[0071] In this embodiment, for skeletal muscle ultrasound images, the image preprocessing module first uses a non-local mean filtering algorithm to deeply mine the similar structural information that is widely present in the image, achieving efficient noise suppression. Compared with traditional filtering methods, non-local mean filtering can effectively remove various types of noise interference while preserving the subtle texture and edge features of skeletal muscle in the image to the greatest extent, providing a clear and accurate image basis for subsequent analysis. After denoising, the grayscale value of the image is reshaped with the help of nonlinear grayscale transformation technology. In this way, the contrast of the denoised skeletal muscle ultrasound image is significantly enhanced, not only allowing key information such as the morphology and structure of skeletal muscle to be more clearly displayed, but also further highlighting the subtle differences caused by pathological changes such as skeletal muscle atrophy, greatly improving the recognizability of image features, and providing strong support for doctors or subsequent analysis algorithms to accurately identify and judge muscle status. Finally, based on the spatial continuity and similarity characteristics of skeletal muscle, the module uses an advanced region-based segmentation algorithm to accurately segment the skeletal muscle region in the ultrasound image. It can fully utilize the coherence of muscle tissue in spatial distribution and the similarity of internal pixel features to effectively distinguish skeletal muscle from surrounding fat, bone, blood vessels and other tissues, and completely and accurately separate the skeletal muscle area. Accurate segmentation of skeletal muscle area has potential significance for evaluating changes in bone density. Because skeletal muscle is closely connected to bones, changes in the state of skeletal muscle will directly affect the stress on the bones. By accurately segmenting the skeletal muscle area, other technical means can be further combined, including the use of grayscale information in ultrasound images, the potential relationship model between texture features and bone density, to preliminarily explore the acquisition of indirect information related to bone density from ultrasound images, providing more basis for the comprehensive assessment of patients' muscle-bone health status.

[0072] See also Figure 2 Preferably, the specific steps of non-local mean dynamic filtering denoising include:

[0073] Acquire skeletal muscle ultrasound images and use the image window to mark suspected areas of skeletal muscle atrophy, thereby dividing the skeletal muscle ultrasound image into suspected atrophy areas and non-suspected atrophy areas. Use rectangle or polygon drawing functions in the image processing library, either manually or based on preset grayscale thresholds, texture features, and other algorithms to mark suspected skeletal muscle atrophy areas on the image. Once marked, the image is divided into suspected atrophy areas and non-suspected atrophy areas, and the pixel coordinate range of each area is stored to prepare for subsequent differentiation processing.

[0074] An initial search window and an initial neighborhood window are set for the skeletal muscle ultrasound image. When traversing each pixel point in the skeletal muscle ultrasound image one by one, if the current pixel point is in a non-suspected atrophy area, the initial neighborhood window of the current pixel point is used as a template and compared with the initial neighborhood windows of other pixel points in the initial search window. The difference in the grayscale values ​​of the pixels in the two initial neighborhood windows can be calculated, for example, by using the Euclidean distance formula to calculate the difference in the grayscale values ​​of the corresponding pixels in the two neighborhood windows, to obtain the similarity weight between the current pixel point and the other pixel points in the initial search window.

[0075] The grayscale values ​​of all pixels in the initial search window are weighted averaged according to their respective similarity weights to obtain the filtered grayscale value of the current pixel;

[0076] When the traversed pixel point is located in a suspected atrophy area, the initial search window and the initial neighborhood window are dynamically adjusted to expand the initial search window and the initial neighborhood window as the second search window and the second neighborhood window; for example, the side length of the initial search window is increased by 5-10 pixels to obtain more possible similar structural information, because muscle atrophy may cause features to be dispersed over a larger range. At the same time, the size of the initial neighborhood window is fine-tuned according to the image resolution and the complexity of the muscle texture in the area. Generally, the side length of the initial neighborhood window can be increased by 1-2 pixels to ensure that local feature changes can be accurately captured. The adjusted windows are used as the second search window and the second neighborhood window, respectively.

[0077] In the second search window, the second neighborhood window of the current pixel point is used as a template and compared with the second neighborhood windows of other pixel points. The difference in the grayscale values ​​of the pixels in the two second neighborhood windows is calculated, and the grayscale co-occurrence matrix is ​​used to calculate the texture feature similarity of the two second neighborhood windows. The texture features include energy features, entropy features, and contrast features. The grayscale value difference and texture feature similarity are combined in a linear weighted manner to obtain the similarity weight between the current pixel point and other pixels in the second search window. This highlights the importance of texture features in the diagnosis of muscle atrophy. Even if the grayscale value does not change significantly, similar areas can be accurately identified when the texture changes.

[0078] The grayscale values ​​of all pixels in the second search window are weighted averaged according to their respective similarity weights, and then quadratically weighted using the distance attenuation function based on their spatial distance from the current pixel. The closer the pixel is to the current pixel, the higher the weight is assigned to it, and the filtered grayscale value of the current pixel is obtained. This enables the denoised image to more accurately reflect the true state of skeletal muscle, providing a high-quality image foundation for subsequent analysis and diagnosis of muscle atrophy.

[0079] All pixel points of the skeletal muscle ultrasound image are traversed to obtain the skeletal muscle ultrasound image after non-local mean dynamic filtering and denoising.

[0080] Preferably, the specific steps of nonlinear grayscale transformation enhancement include:

[0081] Obtain skeletal muscle ultrasound images after non-local mean dynamic filtering and denoising. Skeletal muscle atrophy will cause the image grayscale range to narrow and the details to blur. A logarithmic transformation function that can effectively stretch the low-grayscale area is used for nonlinear grayscale transformation, namely:

[0082] y=c×log(1+x)

[0083] Where x is the grayscale value of the original image pixel, y is the grayscale value of the transformed pixel, and c is a constant. For subtle grayscale changes caused by muscle atrophy, logarithmic transformation can amplify them, highlighting the subtle structure and pathological characteristics of muscle tissue, making it easier to observe signs of muscle atrophy.

[0084] See also Figure 3 、 Figure 4 The traditional grayscale transformation parameters are fixed and difficult to adapt to the differences between different individuals and the degree of muscle atrophy. In this embodiment, before the nonlinear grayscale transformation, the denoised skeletal muscle ultrasound image is subjected to feature analysis to obtain grayscale features, including calculating the grayscale histogram of the denoised skeletal muscle ultrasound image to obtain the concentration range of pixel grayscale values ​​and the degree of grayscale dispersion in the denoised skeletal muscle ultrasound image; the image contrast of the denoised skeletal muscle ultrasound image is obtained by calculating the difference between the maximum grayscale value and the minimum grayscale value in the denoised skeletal muscle ultrasound image and dividing it by the average grayscale value of the denoised skeletal muscle ultrasound image;

[0085] The initial range of grayscale transformation parameters was set. Based on experimental data and experience, the initial range of parameter c for the logarithmic transformation function was determined. When processing denoised skeletal muscle ultrasound images, testing samples with varying degrees of muscle atrophy revealed that c values ​​between 1.5 and 3.5 were able to achieve a certain degree of effective grayscale transformation. This range was then used as the initial grayscale transformation parameter range, providing a basis for subsequent parameter adjustments.

[0086] Within the initial range of grayscale transformation parameters, several representative parameter values ​​are selected using the equidistance method. Nonlinear grayscale transformation is then performed on the denoised skeletal muscle ultrasound images based on these representative parameter values. The grayscale features of the denoised skeletal muscle ultrasound images after the nonlinear grayscale transformation are then obtained. This narrows the initial range of grayscale transformation parameters and determines the fine-tuning range of the grayscale transformation parameters. For example, within the initial range of 1.5-3.5, several representative parameter values ​​are selected, including c1 = 1.5, c2 = 2.0, c3 = 2.5, c4 = 3.0, and c5 = 3.5. These parameter values ​​are then used to perform grayscale transformation on the denoised skeletal muscle ultrasound images, yielding multiple transformed skeletal muscle ultrasound images. The grayscale histogram, contrast, and other features of each transformed image are then recalculated and compared with the grayscale features of the original denoised skeletal muscle ultrasound images. Based on the comparison results, a parameter value range is initially selected that significantly improves the contrast of the denoised skeletal muscle ultrasound images and achieves a more reasonable grayscale distribution. For example, after comparative analysis, when the value of c is between 2.0 and 3.0, the skeletal muscle ultrasound imaging effect is better, and this range is used as the parameter range for the next fine-tuning.

[0087] A data interval was set, and within the grayscale transformation parameter fine-tuning range, a series of grayscale transformation fine-tuning parameter values ​​were generated based on the data interval. Each fine-tuning parameter value was then used to perform a nonlinear grayscale transformation on the denoised skeletal muscle ultrasound images. The structural similarity index and peak signal-to-noise ratio (PSNR) of the transformed skeletal muscle ultrasound images were calculated. The structural similarity index measures the degree of similarity between the transformed image and the original image in terms of structure, brightness, and contrast, while the PSNR reflects the image noise level and signal quality. The fine-tuning parameter value that achieves the optimal combination of the structural similarity index and PSNR was selected as the final logarithmic transformation function parameter value. For example, a comparative analysis showed that a c value of 2.375 achieved the optimal combination of the structural similarity index and PSNR, and therefore the logarithmic transformation function parameter value was chosen as 2.375.

[0088] The target logarithmic transformation function is determined according to the parameter value of the logarithmic transformation function. The grayscale value of each pixel in the denoised skeletal muscle ultrasound image is substituted into the selected target logarithmic transformation function, and the transformed grayscale value is calculated to obtain the skeletal muscle ultrasound image with enhanced contrast.

[0089] See also Figure 5 Preferably, the specific steps of segmenting the skeletal muscle region include:

[0090] Acquire contrast-enhanced ultrasound images of skeletal muscle and initialize skeletal muscle segmentation seed points. Based on characteristics such as texture and grayscale changes associated with muscle atrophy, manually or automatically select seed points in suspected atrophic and normal areas to ensure that the seed points fully reflect the true state of the muscle. Furthermore, the density of seed points is increased in suspected atrophic areas to more accurately capture the boundaries and internal structural changes of the atrophic areas.

[0091] Starting from the seed point, region growing is performed based on the spatial continuity and similarity features of skeletal muscle. The similarity features include texture similarity features and grayscale similarity features. The threshold range of the similarity features is set. If the similarity features of the adjacent pixels of the seed point are all within the threshold range, the adjacent pixel is added to the current region.

[0092] After the region growing is completed, for every two regions, the grayscale mean difference, texture feature distance and property feature difference are calculated to construct a multi-level feature. By calculating the multi-level feature similarity between every two regions and setting a similarity threshold, the regions with multi-level feature similarity greater than the similarity threshold are merged to preliminarily segment the skeletal muscle region.

[0093] After the initial segmentation of the skeletal muscle region, morphological operations are used to refine and repair the muscle edges to obtain the final skeletal muscle region. For skeletal muscle atrophy images, in addition to the traditional removal of isolated small noise areas and filling of holes within the region, muscle edge refinement and repair are also added. Because muscle atrophy can cause blurred or discontinuous muscle edges, morphological operations such as corrosion, dilation, opening, and closing are used to refine the muscle edges and remove burrs and irregularities. Small gaps and discontinuities at the edges are repaired using image interpolation methods, making the edges of the segmented skeletal muscle region more accurate and complete.

[0094] The quantitative analysis module uses the segmented skeletal muscle regions as a basis. By binarizing the skeletal muscle ultrasound images and calculating the fractal dimension using grid sliding statistics and weighted processing, the module quantifies the complexity and irregularity of skeletal muscle texture, providing a quantitative basis for determining the degree of muscle atrophy. The module also calculates the gradient amplitude, gradient direction, and texture direction of the skeletal muscle region, constructs and smoothes the tensor model, and locates areas of skeletal muscle atrophy by calculating the main eigenvalues ​​and eigenvectors, followed by threshold judgment and cluster analysis. This module obtains trend characteristics of muscle in different directions, providing key data support for in-depth analysis of the manifestations of muscle atrophy in different dimensions.

[0095] In this embodiment, an in-depth analysis is conducted based on the skeletal muscle regions segmented by the image preprocessing module. Based on the distribution patterns of pixel grayscale values ​​within the skeletal muscle region, the self-similarity characteristics of skeletal muscle structure and texture are accurately identified and analyzed. Fractal geometry theory is used to calculate the key parameter, the fractal dimension. As a quantitative indicator of the complexity and irregularity of muscle texture, changes in its value during muscle atrophy can intuitively reflect changes in muscle texture characteristics. The quantitative analysis module, by keenly capturing changes in the fractal dimension, provides doctors with a powerful tool for quantitatively determining the extent of muscle atrophy. Using information such as the grayscale gradients and texture orientation of the skeletal muscle region in different directions, the quantitative analysis module constructs a tensor model of the skeletal muscle region. This model can comprehensively and meticulously describe the characteristics of muscle tissue in various dimensions. In muscle atrophy analysis, the tensor model can clearly demonstrate the changing trends of muscle in different directions, such as changes in the arrangement of muscle fibers in different directions and changes in muscle contraction characteristics in different dimensions. This rich information provides key data support for a deeper understanding of the manifestations of muscle atrophy in different dimensions. Furthermore, when the fractal dimension of skeletal muscle texture changes significantly due to muscle atrophy, the bone density of the associated skeletal parts may also fluctuate accordingly. By analyzing ultrasound imaging data from a large number of clinical cases, a correlation model between fractal dimension and bone density changes can be established, providing an indirect method for assessing bone density.

[0096] Preferably, the step of calculating the fractal dimension of the skeletal muscle region includes:

[0097] The segmented skeletal muscle ultrasound image is converted into a binary image to distinguish the skeletal muscle area from the background area, where white pixels represent the skeletal muscle area and black pixels represent the background area;

[0098] According to the side length of the minimum rectangular area where the skeletal muscle area is located, the initial mesh side length is set, that is:

[0099]

[0100] Where K0 is the initial grid side length, α is a constant greater than 2 to ensure that the initial grid can cover a certain range of skeletal muscle area, and a and b are the two side lengths of the minimum rectangular area where the skeletal muscle area is located.

[0101] Set the ending grid length, create an empty list to store grid lengths of different sizes, fill the empty list, start from the initial grid length, and reduce the grid length by half each time until the iterative grid length is less than the ending grid length, thus obtaining a list of grid lengths of different sizes.

[0102] For each size of grid side length, a grid is slid on the binary image in the form of a grid, and two layers of nested loops are used to control the horizontal and vertical positions of the grid respectively;

[0103] For each grid position, determine whether the grid partially covers the skeletal muscle area, i.e., the white pixel area. If so, use the pixel counting method to calculate the number of skeletal muscle area pixels within the grid position and divide it by the total area of ​​the grid to obtain the coverage ratio;

[0104] Configure a ratio threshold. If the coverage ratio is greater than the ratio threshold, the grid is marked as covered. The total number of covered grids at each grid size is counted and stored in a list corresponding to the grid edge length. This fully considers the grid coverage of the edges of the skeletal muscle area and solves the problem that muscle atrophy causes irregular edges that affect the accuracy of fractal dimension calculations.

[0105] Grids of different sizes contribute differently to the calculation of fractal dimension. The total number of covered grids under grids of different sizes is weighted using an exponential decay function. Smaller grids reflect detailed information, which is more critical in the diagnosis of muscle atrophy and is therefore given a higher weight. Larger grids reflect overall characteristics and are given a relatively lower weight.

[0106] A scatter plot was drawn with the logarithm of the grid side length as the abscissa and the logarithm of the total number of weighted covered grids as the ordinate. A linear fit was performed on the scatter plot to obtain the slope of the fitted line. The absolute value of the slope of the fitted line was used as the fractal dimension of the skeletal muscle region, that is:

[0107]

[0108] Where D is the fractal dimension of the skeletal muscle region, β is the grid side length, N(β) is the total number of grids covered by grid side length β, and k is the attenuation weight determined by experiments. The larger the fractal dimension, the higher the complexity and irregularity of the skeletal muscle texture. During muscle atrophy, changes in the fractal dimension can reflect changes in muscle texture characteristics.

[0109] See also Figure 6 Preferably, the specific steps of locating the skeletal muscle atrophy area include:

[0110] For the segmented skeletal muscle region, the gradient operator is used to calculate the gradient of the skeletal muscle region in the horizontal and vertical directions to obtain the gradient amplitude and gradient direction, and the gray-level co-occurrence matrix is ​​used to obtain the texture direction of the skeletal muscle region, reflecting the direction of the muscle fibers.

[0111] Initialize the tensor and define a second-order tensor for each pixel in the skeletal muscle area. The tensor elements are initialized to zero, and the dimension of the tensor is consistent with the dimension of the skeletal muscle area. It is used to describe the characteristics of each pixel in different directions.

[0112] Based on the calculated gradient amplitude, gradient direction, and texture direction, each tensor element is filled. The horizontal gradient amplitude reflects the activity and degree of change of the muscle fibers in the horizontal direction. It is normalized to the interval [0,1] and used as a diagonal element of the tensor. The vertical gradient amplitude is normalized to the interval [0,1] and used as another diagonal element of the tensor. This allows the tensor diagonal elements to directly reflect the intensity of the change in the muscle fibers in the horizontal and vertical directions. The larger the value, the more intense the change in that direction. The difference angle between the gradient direction and the texture direction is calculated, and the value of the cross-term element of the tensor element is determined using the trigonometric function relationship, that is:

[0113]

[0114] Among them, γ is the value of the cross term element of the tensor, θ is the gradient direction, is the texture direction; when the gradient direction and texture direction are aligned, the cross-term element is 1, indicating strong synergy between the two directions; when the two directions are perpendicular, the cross-term element is 0, indicating that the two directions are independent and have no obvious correlation. This filling method can intuitively reflect the relationship between the gradient direction and texture direction, thereby more comprehensively describing the characteristics of muscle fibers in different directions.

[0115] The constructed tensor can be smoothed by using methods such as Gaussian filtering to perform weighted averaging on the tensor of each pixel so that it can better reflect the overall characteristics of the surrounding area.

[0116] For the smoothed tensor, the sliding window technique is used to perform localized analysis. The sliding window is set, and for each set of tensors within the sliding window, its main eigenvalue and corresponding eigenvector are calculated.

[0117] For each set of tensors within the sliding window, an initial vector is randomly selected, whose dimensions match those of the tensor. To ensure computational stability, the initial vector is typically normalized to a modulus of 1. This ensures the stability of the iterative calculation of the principal eigenvalues ​​and eigenvectors, preventing interference with the results due to differences in the size of the initial vector, and enabling the iterative process to more reliably converge to the principal eigenvector.

[0118] Perform multiple iterations, and in the nth iteration, calculate u η =Tv η-1 , where T is the matrix form corresponding to the tensor set in the sliding window, that is, the tensor matrix, v η-1 is the principal eigenvector of the η-1th iteration, u η is the intermediate eigenvector of the ηth iteration, and then calculate That is, u ηNormalize to get the main eigenvector v of the ηth iteration η ,With the increase of the number of iterations, it gradually converges to the main eigenvector; by iteratively approaching the main eigenvector, the eigenvector of the tensor set in each sliding window in the main change direction is accurately obtained, which provides accurate data support for subsequent analysis based on the eigenvector, and can more accurately describe the characteristics of muscle tissue in the main direction.

[0119] Set the convergence threshold to determine whether the iteration has converged. When the difference between the eigenvectors in two adjacent iterations is less than the convergence threshold, the iteration has converged, stop the iteration, and obtain the main eigenvector v after convergence. o , otherwise continue to iterate;

[0120] When the main eigenvector v after convergence is obtained o Then, calculate the main eigenvalue λ1 of the main eigenvector, the formula is in, is the principal eigenvector v after convergence o The transposed vector of is . The change intensity of the tensor set in each sliding window in the direction of the principal eigenvector is accurately calculated. Combined with the principal eigenvector, this method comprehensively describes the main characteristics of the muscle tissue within the sliding window, providing a quantitative basis for subsequent identification of potential atrophy areas.

[0121] The feature difference threshold and angle threshold are set. When the sliding window traverses the skeletal muscle area, for each sliding window, the main eigenvalue difference and the angle of the corresponding eigenvector between it and the adjacent sliding window are calculated. If the main eigenvalue difference is greater than the feature difference threshold, and the angle of the eigenvector is greater than the angle threshold, the current sliding window is marked as a potential atrophy area and its location information is recorded. Otherwise, no processing is performed. By comparing the key feature differences of adjacent sliding windows, areas where muscle atrophy may exist are preliminarily screened out, which narrows the scope of subsequent analysis, improves analysis efficiency, and provides candidate areas for accurately dividing atrophy areas.

[0122] The main eigenvalues ​​and eigenvectors extracted from all sliding windows marked as potential atrophy areas are combined into a set of eigenvectors, each eigenvector containing the main eigenvalue and eigenvector of the sliding window;

[0123] The number of cluster categories is set to 2, representing normal areas and atrophic areas respectively, and the eigenvectors in the eigenvector set are randomly selected as the initial cluster centers;

[0124] The algorithm iterates, calculating the Euclidean distance between each eigenvector and the cluster center and assigning it to the cluster with the closest cluster center. The cluster center of each cluster is then recalculated, representing the mean of all eigenvectors within the cluster. This process is repeated until the cluster center converges. Through continuous iteration, eigenvectors with similar characteristics are clustered together, gradually classifying potential atrophic areas into normal and atrophic regions. This allows for accurate muscle classification and provides an effective method for accurately identifying atrophic areas.

[0125] After clustering is completed, the eigenvalue ranges and eigenvector directions of the two clusters are obtained respectively to locate the skeletal muscle atrophy region clusters.

[0126] The intelligent annotation module is used to perform annotation and post-processing on skeletal muscle ultrasound images, highlight the skeletal muscle area in the skeletal muscle ultrasound image, mark the fractal dimension of the skeletal muscle area, and locally magnify the skeletal muscle atrophy area to generate a skeletal muscle ultrasound image annotation report.

[0127] Preferably, the specific steps of generating a skeletal muscle ultrasound image annotation report include:

[0128] The system receives the skeletal muscle region segmented by the image preprocessing module and uses image rendering technology to prominently mark the skeletal muscle region in the skeletal muscle ultrasound image. Using color mapping and edge thickening, the skeletal muscle region is clearly distinguished from the surrounding background, allowing doctors or analysts to quickly focus on the region of interest. Furthermore, by adjusting image parameters such as brightness, contrast, and saturation, the visual recognition of the skeletal muscle region is further enhanced, ensuring intuitive identification of the skeletal muscle region in complex ultrasound images.

[0129] The quantitative analysis module obtains fractal dimension data for the skeletal muscle region and clearly and prominently annotates the specific fractal dimension values ​​at specific locations on the ultrasound image, including image corners and adjacent to the skeletal muscle region. Furthermore, the system provides interpretation of the fractal dimension, comparing the current fractal dimension to a standard range and displaying the complexity and irregularity of the skeletal muscle texture as reflected by the fractal dimension in the form of charts or text annotations. This helps even non-specialists understand the clinical significance of fractal dimension and provides a quantitative and intuitive display for diagnosing muscle atrophy.

[0130] Based on the skeletal muscle atrophy area located by the quantitative analysis module, a local magnification operation is performed on the skeletal muscle atrophy area. The image interpolation algorithm is used to maintain the clarity and detail integrity of the image during the magnification process to avoid blurring or distortion. At the same time, an indicator frame and a zoom ratio mark are added around the magnified area to facilitate users to understand the position and zoom level of the magnified area in the original image. In addition, feature annotation is performed on the atrophy area, including marking the abnormal range of eigenvalues ​​and the main direction of the eigenvector, to further highlight the characteristics of the atrophy area and provide doctors with more detailed information for in-depth analysis of the specific situation of muscle atrophy.

[0131] The highlighted skeletal muscle area, fractal dimension annotation, and magnification and feature annotation of the atrophy area are integrated to generate a complete annotation report. The report is presented in the form of pictures and text, including the original ultrasound image, the annotated image, and a detailed text description. The text description covers the basic information of the skeletal muscle area, the analysis results of the fractal dimension, and the feature description of the atrophy area. Information related to bone density can also be added. The bone density data obtained from clinical studies is correlated with the current ultrasound image processing results to show the possible correlation between skeletal muscle area characteristics and bone density in the report. For example, when annotating the fractal dimension of the skeletal muscle area, the potential relationship between the change in the fractal dimension and the change in bone density is also explained. For the skeletal muscle atrophy area, in addition to local amplification and feature annotation, the bone density data can also be combined to annotate the possible bone density changes in the corresponding bone part of the area, providing doctors with more comprehensive information to assist them in making a comprehensive diagnosis.

[0132] Example 2

[0133] See also Figure 7 This embodiment introduces a skeletal muscle ultrasonic image processing method for ultrasonic detection, comprising the following steps:

[0134] Step S1: After denoising the skeletal muscle ultrasound image using non-local mean dynamic filtering, the contrast of the denoised skeletal muscle ultrasound image is enhanced by nonlinear grayscale transformation, and the skeletal muscle region in the skeletal muscle ultrasound image is segmented using the spatial continuity and similarity of the skeletal muscle;

[0135] Step S2: Based on the segmented skeletal muscle region, the skeletal muscle ultrasound image is binarized, and the fractal dimension is calculated using grid sliding statistics and weighted processing to quantify the complexity and irregularity of the skeletal muscle texture. The gradient amplitude, gradient direction, and texture direction of the skeletal muscle region are calculated, a tensor model is constructed and smoothed, and the main eigenvalues ​​and eigenvectors are calculated. The skeletal muscle atrophy region is located through threshold judgment and cluster analysis.

[0136] Step S3: perform post-annotation processing on the skeletal muscle ultrasound image, highlight the skeletal muscle area in the skeletal muscle ultrasound image, mark the fractal dimension of the skeletal muscle area, and locally magnify the skeletal muscle atrophy area to generate a skeletal muscle ultrasound image annotation report.

[0137] Preferably, the specific steps of non-local mean dynamic filtering denoising include:

[0138] Obtain skeletal muscle ultrasound images, use image windows to mark suspected skeletal muscle atrophy areas, and divide the skeletal muscle ultrasound images into suspected atrophy areas and non-suspected atrophy areas;

[0139] An initial search window and an initial neighborhood window are set for the skeletal muscle ultrasound image. When each pixel in the skeletal muscle ultrasound image is traversed one by one, if the current pixel is in a non-suspected atrophy area, the initial neighborhood window of the current pixel is used as a template to compare with the initial neighborhood windows of other pixels in the initial search window. The similarity weight between the current pixel and other pixels in the initial search window is obtained by calculating the difference in the pixel grayscale values ​​in the two initial neighborhood windows.

[0140] The grayscale values ​​of all pixels in the initial search window are weighted averaged according to their respective similarity weights to obtain the filtered grayscale value of the current pixel;

[0141] When the traversed pixel point is located in the suspected atrophy area, the initial search window and the initial neighborhood window are dynamically adjusted, and the initial search window and the initial neighborhood window are expanded to serve as the second search window and the second neighborhood window;

[0142] In the second search window, the second neighborhood window of the current pixel is used as a template and compared with the second neighborhood windows of other pixels, the difference in the grayscale values ​​of the pixels in the two second neighborhood windows is calculated, and the similarity of the texture features of the two second neighborhood windows is calculated; the difference in grayscale values ​​and the similarity of texture features are combined to obtain the similarity weight between the current pixel and other pixels in the second search window;

[0143] The grayscale values ​​of all pixels in the second search window are weighted averaged according to their respective similarity weights, and then weighted twice according to their spatial distance from the current pixel to obtain the filtered grayscale value of the current pixel;

[0144] All pixel points of the skeletal muscle ultrasound image are traversed to obtain the skeletal muscle ultrasound image after non-local mean dynamic filtering and denoising.

[0145] Preferably, the specific steps of segmenting the skeletal muscle region include:

[0146] Obtain contrast-enhanced skeletal muscle ultrasound images and initialize skeletal muscle segmentation seed points;

[0147] Starting from the seed point, region growing is performed based on the spatial continuity and similarity features of skeletal muscle. The similarity features include texture similarity features and grayscale similarity features. The threshold range of the similarity features is set. If the similarity features of the adjacent pixels of the seed point are all within the threshold range, the adjacent pixels are added to the current region.

[0148] After the region growing is completed, for every two regions, the grayscale mean difference, texture feature distance and property feature difference are calculated to construct a multi-level feature. By calculating the multi-level feature similarity between every two regions and setting a similarity threshold, the regions with multi-level feature similarity greater than the similarity threshold are merged to preliminarily segment the skeletal muscle region.

[0149] For the skeletal muscle region after preliminary segmentation, morphological operations are used to refine and repair the muscle edges to obtain the final skeletal muscle region.

[0150] Preferably, the step of calculating the fractal dimension of the skeletal muscle region includes:

[0151] The segmented skeletal muscle ultrasound image is converted into a binary image to distinguish the skeletal muscle area from the background area, where white pixels represent the skeletal muscle area and black pixels represent the background area;

[0152] Set the initial mesh side length based on the side length of the smallest rectangular area where the skeletal muscle area is located;

[0153] Set the ending mesh length and create an empty list to store mesh lengths of different sizes. Starting from the initial mesh length, the mesh length is halved each time it is iterated until the iterative mesh length is less than the ending mesh length. Then, a list of mesh lengths of different sizes is obtained.

[0154] For each size of grid side length, a grid is slid on the binary image in the form of a grid, and two layers of nested loops are used to control the horizontal and vertical positions of the grid respectively;

[0155] For each grid position, determine whether the grid partially covers the skeletal muscle area, i.e., the white pixel area. If so, use the pixel counting method to calculate the number of skeletal muscle area pixels within the grid position and divide it by the total area of ​​the grid to obtain the coverage ratio;

[0156] Configure a ratio threshold. If the coverage ratio is greater than the ratio threshold, the grid is marked as covered. The total number of covered grids at each grid size is counted.

[0157] The total number of covered grids under different grid sizes was weighted using an exponential decay function. A scatter plot was drawn with the logarithm of the grid side length as the horizontal axis and the logarithm of the weighted total number of covered grids as the vertical axis. The scatter plot was then linearly fitted to obtain the slope of the fitting line. The absolute value of the slope of the fitting line was used as the fractal dimension of the skeletal muscle region.

[0158] Preferably, the specific steps of locating the skeletal muscle atrophy area include:

[0159] For the segmented skeletal muscle region, the gradient operator is used to calculate the gradient of the skeletal muscle region in the horizontal and vertical directions, and the gradient amplitude and gradient direction are obtained. The gray-level co-occurrence matrix is ​​used to obtain the texture direction of the skeletal muscle region.

[0160] Initialize the tensor and define a second-order tensor for each pixel in the skeletal muscle area. The tensor elements are initialized to zero, and the dimension of the tensor is consistent with the dimension of the skeletal muscle area.

[0161] Each tensor element is filled according to the calculated gradient amplitude, gradient direction, and texture direction. The horizontal gradient amplitude reflects the activity and change degree of muscle fibers in the horizontal direction. It is normalized and used as a diagonal element of the tensor. The vertical gradient amplitude is normalized and used as another diagonal element of the tensor.

[0162] Calculate the difference angle between the gradient direction and the texture direction, and use the trigonometric function relationship to determine the value of the cross term element of the tensor element;

[0163] After smoothing the constructed tensor, the sliding window technique is used for localized analysis. The sliding window is set, and for each set of tensors within the sliding window, its main eigenvalue and corresponding eigenvector are calculated.

[0164] Set the feature difference threshold and angle threshold. When the sliding window traverses the skeletal muscle area, for each sliding window, calculate the main eigenvalue difference between it and the adjacent sliding window and the angle between the corresponding eigenvectors. If the main eigenvalue difference is greater than the feature difference threshold and the angle between the eigenvectors is greater than the angle threshold, then mark the current sliding window as a potential atrophy area and record its location information.

[0165] The main eigenvalues ​​and eigenvectors extracted from all sliding windows marked as potential atrophy areas are combined into a set of eigenvectors, each eigenvector containing the main eigenvalue and eigenvector of the sliding window;

[0166] Set the number of cluster categories to represent normal areas and atrophic areas respectively, and randomly select the feature vectors in the feature vector set as the initial cluster center;

[0167] Iterate. In each iteration, for each feature vector in the feature vector set, calculate its Euclidean distance with the cluster center and assign it to the cluster with the nearest cluster center.

[0168] Recalculate the cluster center of each cluster, that is, the mean of all eigenvectors in the cluster, and repeat the iterative steps until the cluster center position of each cluster converges;

[0169] After clustering is completed, the eigenvalue ranges and eigenvector directions of the two clusters are obtained respectively to locate the skeletal muscle atrophy region clusters.

[0170] Working principle and its effect:

[0171] After acquiring skeletal muscle ultrasound images, the image window is used to delineate suspected and non-suspected atrophy regions. Non-local mean dynamic filtering denoising is then performed. This method, while potentially dispersive, suppresses noise more precisely, preserving subtle texture and edge features of skeletal muscle to the greatest extent possible, providing a clear and accurate image foundation for subsequent analysis. Nonlinear grayscale enhancement employs a logarithmic transformation function. By analyzing the grayscale features of the denoised image, the algorithm selects optimal parameters based on the structural similarity index and peak signal-to-noise ratio, moving from an initial range screening to a fine-tuning range. This significantly enhances image contrast, highlighting subtle differences in skeletal muscle morphology, structure, and atrophy, improving image feature discernibility and facilitating accurate muscle identification and assessment. A region segmentation method based on spatial continuity and similarity of skeletal muscle begins with an initial seed point, then performs region growing based on texture and grayscale similarity. Similar regions are merged by constructing multi-level features, and finally, morphological operations are used to refine and restore muscle edges. This method effectively distinguishes skeletal muscle from surrounding tissue, completely and accurately isolating skeletal muscle regions. This provides reliable data support for quantitative assessment of muscle atrophy severity, improving the accuracy and efficiency of subsequent analysis and diagnosis.

[0172] By binarizing skeletal muscle ultrasound images and calculating the fractal dimension using grid sliding statistics and weighted processing, the complexity and irregularity of skeletal muscle texture can be accurately quantified. It can keenly capture the changes in texture features during muscle atrophy, providing doctors with a powerful tool to determine the degree of muscle atrophy from a quantitative perspective. A tensor model is constructed by calculating the gradient amplitude, gradient direction, and texture direction of the skeletal muscle area and then smoothing it. The sliding window technology is combined with multiple iterations to calculate the main eigenvalues ​​and eigenvectors to accurately obtain the changing trends of muscles in different directions, and comprehensively display the changes in muscle fiber arrangement, contraction characteristics, etc. in different dimensions. It provides key data support for in-depth understanding of the manifestations of muscle atrophy in different dimensions, which helps doctors fully grasp the condition.

[0173] The intelligent annotation module performs post-processing on skeletal muscle ultrasound images, highlighting skeletal muscle areas and enabling doctors to quickly focus on areas of concern and assist with diagnosis. It also zooms in on areas of skeletal muscle atrophy, using image interpolation algorithms to maintain clarity. It also adds feature annotations, such as the range of abnormal eigenvalues ​​and the direction of eigenvectors, to facilitate in-depth analysis of atrophy. Finally, it generates a graphically illustrated annotated report, integrating various information and clinical recommendations, improving the system's practicality and clinical value, and facilitating medical record keeping and research analysis.

[0174] From image optimization to feature analysis and result presentation, the system provides a comprehensive, systematic solution for the diagnosis of skeletal muscle diseases, particularly muscle atrophy. Through a series of advanced image processing and analysis technologies, it can quickly and accurately extract relevant skeletal muscle features, locate atrophic areas, and present them to doctors in an intuitive manner. This significantly reduces the time and workload of doctors during diagnosis, while improving the accuracy and reliability of diagnosis and facilitating the development of more precise treatment plans.

[0175] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the concept of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A skeletal muscle ultrasonic image processing system for ultrasonic detection, characterized in that: Includes image preprocessing module, quantitative analysis module and intelligent labeling module; The image preprocessing module is used to perform denoising on the skeletal muscle ultrasound image using non-local mean dynamic filtering, enhance the contrast of the denoised skeletal muscle ultrasound image through nonlinear grayscale transformation, and segment the skeletal muscle region in the skeletal muscle ultrasound image using the spatial continuity and similarity of the skeletal muscle; The quantitative analysis module is used to quantify the complexity and irregularity of skeletal muscle texture based on the segmented skeletal muscle area by binarizing the skeletal muscle ultrasound image and calculating the fractal dimension using grid sliding statistics and weighted processing; The gradient amplitude, gradient direction, and texture direction of the skeletal muscle area are calculated, a tensor model is constructed and smoothed, the main eigenvalues ​​and eigenvectors are calculated, and the skeletal muscle atrophy area is located through threshold judgment and cluster analysis; The intelligent annotation module is used to perform annotation post-processing on the skeletal muscle ultrasound image, highlight the skeletal muscle area in the skeletal muscle ultrasound image, mark the fractal dimension of the skeletal muscle area, and locally magnify the skeletal muscle atrophy area to generate a skeletal muscle ultrasound image annotation report; The specific steps of locating the skeletal muscle atrophy area include: For the segmented skeletal muscle region, the gradient operator is used to calculate the gradient of the skeletal muscle region in the horizontal and vertical directions, and the gradient amplitude and gradient direction are obtained. The gray-level co-occurrence matrix is ​​used to obtain the texture direction of the skeletal muscle region. Initialize the tensor and define a second-order tensor for each pixel in the skeletal muscle area. The tensor elements are initialized to zero, and the dimension of the tensor is consistent with the dimension of the skeletal muscle area. Fill each tensor element according to the calculated gradient magnitude, gradient direction, and texture direction. For the horizontal gradient magnitude, normalize it and use it as a diagonal element of the tensor. Normalize the vertical gradient magnitude and use it as another diagonal element of the tensor. Calculate the difference angle between the gradient direction and the texture direction, and use the trigonometric function relationship to determine the value of the cross term element of the tensor element; After smoothing the constructed tensor, the sliding window technology is used for localized analysis. The sliding window is set, and for each set of tensors within the sliding window, its main eigenvalue and corresponding eigenvector are calculated.

2. The skeletal muscle ultrasonic image processing system for ultrasonic testing according to claim 1, characterized in that: The specific steps of the non-local mean dynamic filtering denoising include: Obtain skeletal muscle ultrasound images, use image windows to mark suspected skeletal muscle atrophy areas, and divide the skeletal muscle ultrasound images into suspected atrophy areas and non-suspected atrophy areas; An initial search window and an initial neighborhood window are set for the skeletal muscle ultrasound image. When each pixel in the skeletal muscle ultrasound image is traversed one by one, if the current pixel is in a non-suspected atrophy area, the initial neighborhood window of the current pixel is used as a template to compare with the initial neighborhood windows of other pixels in the initial search window. The similarity weight between the current pixel and other pixels in the initial search window is obtained by calculating the difference in the pixel grayscale values ​​in the two initial neighborhood windows. The grayscale values ​​of all pixels in the initial search window are weighted averaged according to their respective similarity weights to obtain the filtered grayscale value of the current pixel.

3. The skeletal muscle ultrasonic image processing system for ultrasonic testing according to claim 2, characterized in that: The specific steps of the non-local mean dynamic filtering denoising further include: When the traversed pixel point is located in the suspected atrophy area, the initial search window and the initial neighborhood window are dynamically adjusted, and the initial search window and the initial neighborhood window are expanded to serve as the second search window and the second neighborhood window; In the second search window, the second neighborhood window of the current pixel is used as a template and compared with the second neighborhood windows of other pixels, the difference in the grayscale values ​​of the pixels in the two second neighborhood windows is calculated, and the similarity of the texture features of the two second neighborhood windows is calculated; the difference in grayscale values ​​and the similarity of texture features are combined to obtain the similarity weight between the current pixel and other pixels in the second search window; The grayscale values ​​of all pixels in the second search window are weighted averaged according to their respective similarity weights, and then weighted twice according to their distance from the current pixel in space to obtain the filtered grayscale value of the current pixel; All pixel points of the skeletal muscle ultrasound image are traversed to obtain the skeletal muscle ultrasound image after non-local mean dynamic filtering and denoising.

4. The skeletal muscle ultrasonic image processing system for ultrasonic testing according to claim 3, characterized in that: The specific steps of the nonlinear grayscale transformation enhancement include: Obtain skeletal muscle ultrasound images after non-local mean dynamic filtering and denoising, and use logarithmic transformation function to perform nonlinear grayscale transformation; Before nonlinear grayscale transformation, feature analysis was performed on the denoised skeletal muscle ultrasound images to obtain grayscale features, including grayscale value concentration range, grayscale value dispersion, and image contrast. Setting an initial range of grayscale transformation parameters, selecting multiple representative parameter values ​​within the initial range of grayscale transformation parameters by an equidistance method, performing nonlinear grayscale transformation on the denoised skeletal muscle ultrasound image according to the representative parameter values, obtaining the grayscale features of the denoised skeletal muscle ultrasound image after the nonlinear grayscale transformation, narrowing the initial range of grayscale transformation parameters, and obtaining a fine-tuning range of grayscale transformation parameters; The data interval is set, and within the grayscale transformation parameter fine-tuning range, a series of grayscale transformation fine-tuning parameter values ​​are generated according to the data interval. Each fine-tuning parameter value is used again to perform nonlinear grayscale transformation on the denoised skeletal muscle ultrasound image, and the structural similarity index and peak signal-to-noise ratio of the transformed skeletal muscle ultrasound image are calculated. The fine-tuning parameter value that achieves the optimal combination of the structural similarity index and peak signal-to-noise ratio is selected as the final logarithmic transformation function parameter value; The target logarithmic transformation function is determined according to the parameter value of the logarithmic transformation function. The grayscale value of each pixel in the denoised skeletal muscle ultrasound image is substituted into the selected target logarithmic transformation function, and the transformed grayscale value is calculated to obtain the skeletal muscle ultrasound image with enhanced contrast.

5. The skeletal muscle ultrasonic image processing system for ultrasonic testing according to claim 4, characterized in that: The specific steps of segmenting the skeletal muscle region include: Obtain contrast-enhanced skeletal muscle ultrasound images and initialize skeletal muscle segmentation seed points; Starting from the seed point, region growing is performed based on the spatial continuity and similarity features of skeletal muscle. The similarity features include texture similarity features and grayscale similarity features. The threshold range of the similarity features is set. If the similarity features of the adjacent pixels of the seed point are all within the threshold range, the adjacent pixels are added to the current region. After the region growing is completed, for every two regions, the grayscale mean difference, texture feature distance and property feature difference are calculated to construct a multi-level feature. By calculating the multi-level feature similarity between every two regions and setting a similarity threshold, the regions with multi-level feature similarity greater than the similarity threshold are merged to preliminarily segment the skeletal muscle region. For the skeletal muscle region after preliminary segmentation, morphological operations are used to refine and repair the muscle edges to obtain the final skeletal muscle region.

6. The skeletal muscle ultrasonic image processing system for ultrasonic testing according to claim 1, characterized in that: The steps of calculating the fractal dimension of the skeletal muscle region include: The segmented skeletal muscle ultrasound image is converted into a binary image to distinguish the skeletal muscle area from the background area, where white pixels represent the skeletal muscle area and black pixels represent the background area; Set the initial mesh side length based on the side length of the smallest rectangular area where the skeletal muscle area is located; Set the ending mesh length and create an empty list to store mesh lengths of different sizes. Starting from the initial mesh length, the mesh length is halved each time it is iterated until the iterative mesh length is less than the ending mesh length. Then, a list of mesh lengths of different sizes is obtained. For each size of grid side length, a grid is slid on the binary image in the form of a grid, and two layers of nested loops are used to control the horizontal and vertical positions of the grid respectively; For each grid position, determine whether the grid partially covers the skeletal muscle area, i.e., the white pixel area. If so, use the pixel counting method to calculate the number of skeletal muscle area pixels within the grid position and divide it by the total area of ​​the grid to obtain the coverage ratio; Configure a ratio threshold. If the coverage ratio is greater than the ratio threshold, the grid is marked as covered. The total number of covered grids at each grid size is counted. The total number of covered grids under different grid sizes was weighted using an exponential decay function. A scatter plot was drawn with the logarithm of the grid side length as the horizontal axis and the logarithm of the weighted total number of covered grids as the vertical axis. The scatter plot was then linearly fitted to obtain the slope of the fitting line. The absolute value of the slope of the fitting line was used as the fractal dimension of the skeletal muscle region.

7. The skeletal muscle ultrasonic image processing system for ultrasonic testing according to claim 1, characterized in that: The specific step of locating the skeletal muscle atrophy area also includes: Set the feature difference threshold and angle threshold. When the sliding window traverses the skeletal muscle area, for each sliding window, calculate the main eigenvalue difference between it and the adjacent sliding window and the angle between the corresponding eigenvectors. If the main eigenvalue difference is greater than the feature difference threshold and the angle between the eigenvectors is greater than the angle threshold, then mark the current sliding window as a potential atrophy area and record its location information. The main eigenvalues ​​and eigenvectors extracted from all sliding windows marked as potential atrophy areas are combined into a set of eigenvectors, each eigenvector containing the main eigenvalue and eigenvector of the sliding window; Set the number of cluster categories to represent normal areas and atrophic areas respectively, and randomly select the feature vectors in the feature vector set as the initial cluster center; Iterate. In each iteration, for each feature vector in the feature vector set, calculate its Euclidean distance with the cluster center and assign it to the cluster with the nearest cluster center. Recalculate the cluster center of each cluster, that is, the mean of all eigenvectors in the cluster, and repeat the iterative steps until the cluster center position of each cluster converges; After clustering is completed, the eigenvalue ranges and eigenvector directions of the two clusters are obtained respectively to locate the skeletal muscle atrophy region clusters.

8. The skeletal muscle ultrasonic image processing system for ultrasonic testing according to claim 1, characterized in that: The steps of calculating the main eigenvalue and the corresponding eigenvector include: For each set of tensors in the sliding window, randomly select an initial vector and make its dimension consistent with the dimension of the tensor; Iterate, and in the ηth iteration, calculate u η =Tv η-1 , where T is the matrix form corresponding to the tensor set in the sliding window, that is, the tensor matrix, v η-1 is the principal eigenvector of the η-1th iteration, u η is the intermediate eigenvector of the ηth iteration, for u η Normalize to get the main eigenvector v of the ηth iteration η ,Right now Set the convergence threshold to determine whether the iteration has converged. When the difference between the eigenvectors in two adjacent iterations is less than the convergence threshold, the iteration has converged, stop the iteration, and obtain the main eigenvector v after convergence. ° , otherwise continue to iterate; When the main eigenvector v after convergence is obtained ° Then, calculate the main eigenvalue λ1 of the main eigenvector, the formula is in, is the principal eigenvector v after convergence ° The transposed vector of .

9. A skeletal muscle ultrasonic image processing method for ultrasonic testing, which is implemented based on a skeletal muscle ultrasonic image processing system for ultrasonic testing according to any one of claims 1 to 8, characterized in that: The following steps are involved: Step S1: After denoising the skeletal muscle ultrasound image using non-local mean dynamic filtering, the contrast of the denoised skeletal muscle ultrasound image is enhanced by nonlinear grayscale transformation, and the skeletal muscle region in the skeletal muscle ultrasound image is segmented using the spatial continuity and similarity of the skeletal muscle; Step S2: Based on the segmented skeletal muscle region, the skeletal muscle ultrasound image is binarized, and the fractal dimension is calculated using grid sliding statistics and weighted processing to quantify the complexity and irregularity of the skeletal muscle texture; The gradient amplitude, gradient direction, and texture direction of the skeletal muscle area are calculated, a tensor model is constructed and smoothed, the main eigenvalues ​​and eigenvectors are calculated, and the skeletal muscle atrophy area is located through threshold judgment and cluster analysis; Step S3: perform post-annotation processing on the skeletal muscle ultrasound image, highlight the skeletal muscle area in the skeletal muscle ultrasound image, mark the fractal dimension of the skeletal muscle area, and locally magnify the skeletal muscle atrophy area to generate a skeletal muscle ultrasound image annotation report.

Images