A muscle biopsy method and device based on a multi-scale muscle biopsy topological network

Abstract

This invention provides a muscle biopsy method and apparatus based on a multi-scale muscle biopsy topology network. By automatically segmenting and identifying muscle fibers, cell nuclei, connective tissue, and microvessels in digital muscle biopsy slices, each muscle fiber instance is abstracted as a network node, and a topology network is constructed based on its spatial location. On this basis, the system not only quantifies the multi-dimensional attributes of individual muscle fibers, such as subtype, morphology, texture, and nuclear distribution, but also innovatively calculates network-level parameters reflecting the overall state of the tissue microenvironment, including muscle fiber type aggregation factors, fibrosis invasion gradient, and microvascular support index. This achieves a leap from isolated morphological analysis to comprehensive evaluation based on spatial topological networks. This invention transforms traditional pathological assessment, which relies on subjective experience, into a computable and repeatable quantitative analysis system, accurately quantifying key pathological patterns, improving the objectivity of muscle pathology diagnosis, and providing data-driven decision support for clinical practice and research.

Description

Technical Field

[0001] This invention relates to the field of muscle biopsy technology, and in particular to a muscle biopsy method and apparatus based on a multi-scale muscle biopsy topology network. Background Technology

[0002] Muscle biopsy is the gold standard for diagnosing neuromuscular diseases. The diagnostic process relies on pathologists observing and evaluating characteristics such as muscle fiber size, morphology, type distribution, nucleus location, and connective tissue proliferation in the biopsy sections. Traditional diagnostic methods are primarily based on manual interpretation, which suffers from high subjectivity, insufficient repeatability, and low diagnostic efficiency. Furthermore, the human eye struggles to accurately quantify complex information such as the spatial distribution patterns of muscle fibers and interactions within the pathological microenvironment, leading to the easy omission of key pathological features in the diagnosis of early, atypical, or complex myopathy.

[0003] With the development of digital pathology and computer-aided analysis technologies, some image processing-based muscle fiber analysis methods have emerged. However, existing technologies are mostly limited to measuring basic morphological parameters such as muscle fiber diameter and area, and have not yet achieved systematic mining and integration of multidimensional and multi-level pathological information in muscle biopsy images. For example, there is currently a lack of objective and quantitative assessment indicators for the diagnostically significant phenomenon of "muscle fiber type aggregation" in neurogenic damage; and the local effects of connective tissue hyperplasia on muscle fibers in diseases such as muscular dystrophy cannot be accurately characterized at the single-fiber level. In addition, existing methods usually ignore the spatial topological relationships between muscle fibers and their interactions with other biological structures such as microvessels and cell nuclei, thus making it difficult to comprehensively reflect the overall pathological state of the tissue microenvironment.

[0004] Therefore, there is an urgent need for a systematic method that can perform automated and networked quantitative analysis of muscle biopsy images from multiple scales and dimensions, which cannot meet the urgent clinical need for objective, accurate and repeatable assessment of muscle pathological status. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a muscle biopsy method and apparatus based on a multi-scale muscle biopsy topology network, in order to eliminate or improve one or more defects existing in the prior art, and solve the technical problem that existing muscle biopsy analysis methods cannot comprehensively and objectively quantify the spatial distribution of muscle fibers and their topology network characteristics in interaction with the pathological microenvironment.

[0006] One aspect of the present invention provides a muscle biopsy method based on a multi-scale muscle biopsy topology network, the method comprising the following steps:

[0007] Obtain digital slices of muscle biopsy of the sample to be analyzed after processing with the first staining scheme, and identify single muscle fibers, cell nuclei, connective tissue and microvessels and their masks in the digital slices of muscle biopsy based on a pre-trained image segmentation network;

[0008] The mask is aligned to the digital slice of muscle biopsy of the sample to be analyzed after processing with multiple second staining schemes through image registration; the optical feature values ​​of the individual muscle fibers are calculated and the muscle fiber subtype is determined based on a first preset rule; geometric morphological analysis is performed on each individual muscle fiber to calculate multiple preset morphological parameters; texture features are extracted for each individual muscle fiber using one or more texture analysis algorithms; the total number of cell nuclei and the proportion of central nuclei within the preset muscle fiber mask area are statistically analyzed.

[0009] Using the single muscle fiber as a node and determining its spatial representative point, the edges between the nodes are generated according to the coordinates of the spatial representative point using a preset spatial adjacency graph construction algorithm to obtain a muscle biopsy topology network; the muscle fiber subtype, morphological parameters, texture features, total number of cell nuclei, and central nucleus ratio of the single muscle fiber are added to the muscle biopsy topology network as node attributes to construct a graph data structure;

[0010] Based on the myofibril subtype of each node in the graph data structure, calculate the myofibril type aggregation factor for each myofibril subtype according to the second preset rule; calculate the fibrosis invasion gradient of the degree of influence of the surrounding fibrosis environment on the single myofibril corresponding to each node based on the third preset rule; take the microvessels as resource points, calculate the distance from each node to the surrounding resource points, and calculate the microvessel support index according to the fourth preset rule.

[0011] Based on the myofibril subtype, morphological parameters, texture features, total number of nuclei, central nucleus ratio, fibrosis invasion gradient, and microvascular support index corresponding to each individual myofibril, a quantitative analysis report containing a global statistical summary is generated and visualized, in conjunction with the myofibril type aggregation factor.

[0012] In some embodiments, the pre-training step of the image segmentation network includes:

[0013] Obtain a first training sample set containing multiple samples, each sample containing a muscle biopsy digital slice sample processed by a first staining protocol, and labeling the location of single muscle fibers, cell nuclei, connective tissue and microvessels as tags;

[0014] The initial image segmentation network is trained using the first training sample set. The initial image segmentation network takes the muscle biopsy digital slice sample as input and outputs the predicted location and predicted classification of single muscle fibers, cell nuclei, connective tissue and microvessels in the image. The initial image segmentation network is updated by constructing a loss based on the deviation between the predicted location, the predicted classification and the label to obtain the image segmentation network.

[0015] The initial image segmentation network includes a semantic segmentation network or an instance segmentation network based on a convolutional neural network architecture. The semantic segmentation network is a U-Net network or a DeepLab network, and the instance segmentation network is a Mask R-CNN network.

[0016] In some embodiments, after obtaining the muscle biopsy digital slices of the sample to be analyzed after processing with the first staining scheme, the Macenko method is used to perform color normalization on all patches.

[0017] The method further includes: using a mask-guided registration network MrRegNet to align the mask to the digital slices of muscle biopsy of the sample to be analyzed after processing with multiple second staining schemes;

[0018] The optical feature values ​​include average gray value and color channel intensity. The method determines the muscle fiber subtype based on a first preset rule, including: distinguishing the muscle fiber subtype of the single muscle fiber according to the preset value range of average gray value and color channel intensity.

[0019] In some embodiments, a geometric morphological analysis is performed on each of the individual muscle fibers to calculate a plurality of preset morphological parameters, including:

[0020] The total number of pixels within the mask is used as the cross-sectional area of ​​the single muscle fiber, and the total number of pixels at the mask boundary is used as the perimeter of the single muscle fiber.

[0021] The formula for calculating roundness is:

[0022] ;

[0023] Wherein, C represents the circularity, A represents the cross-sectional area, and P represents the perimeter;

[0024] The polygonal degree is calculated using the following formula:

[0025] ;

[0026] Where Poly represents the polygon degree, and A represents the cross-sectional area. This represents the convex hull area of ​​the mask corresponding to the single muscle fiber.

[0027] In some embodiments, one or more texture analysis algorithms are used to extract texture features for each of the individual muscle fibers, including:

[0028] Extract the grayscale image within the single muscle fiber mask and calculate the grayscale co-occurrence matrix;

[0029] The formula for calculating contrast texture features is:

[0030] ;

[0031] The energy texture features are calculated using the following formula:

[0032] ;

[0033] Where i and j represent the grayscale values ​​of pixel pairs with a preset spatial relationship. This represents the joint probability of a pixel pair with gray values ​​i and j appearing in the mask region.

[0034] The method further includes: identifying the probability of presence and severity level of a specified pathology based on a pre-trained pathology classifier according to the contrast texture features and the energy texture features;

[0035] The pre-training steps of the pathology classifier include: obtaining a second training sample set containing multiple samples, each sample containing a muscle biopsy digital slice sample processed by multiple second staining schemes, and labeling the sample contrast texture features and sample energy texture features of the grayscale image of the mask region corresponding to a single muscle fiber, and labeling the pathology type and severity level as labels; using the second training sample set to train a support vector machine to obtain the pathology classifier.

[0036] In some embodiments, the method further includes: extracting the geometric centroid of the single muscle fiber as the spatial representative point, and obtaining the coordinates of the geometric centroid;

[0037] The method uses the Delaunay triangulation algorithm, Voronoi graph algorithm, or radius-based nearest neighbor graph algorithm to generate the edges between the nodes by combining the coordinates of the spatial representative points, thus obtaining the muscle biopsy topology network.

[0038] In some embodiments, based on the myofiber subtype of each node in the graph data structure, a myofiber type aggregation factor for each myofiber subtype is calculated according to a second preset rule, including:

[0039] Based on the proportion of neighbors of the same type according to the muscle fiber subtype of the neighboring nodes, the mean of the proportion of neighbors of the same type is calculated for all nodes of the specified type, and a muscle fiber type aggregation factor is obtained to quantify the degree of aggregation of this type of fiber.

[0040] The fibrosis invasion gradient of the degree of influence of the surrounding fibrotic environment on the single muscle fiber corresponding to each node is calculated based on the third preset rule, including: for each pixel in the single muscle fiber mask region, the distance to the nearest connective tissue is calculated to construct a pressure field; the integral or average value of the pressure field values ​​of all pixels on the boundary contour of the single muscle fiber mask region is calculated to obtain the fibrosis invasion gradient.

[0041] Using the microvessels as resource points, the distance from each node to surrounding resource points is calculated, and the microvessel support index is calculated according to the fourth preset rule, including: counting the number of resource points within a set distance range from the node as the microvessel support index.

[0042] In some embodiments, the method further includes:

[0043] The percentage of fibers with a cross-sectional area less than 2 standard deviations from the normal mean was used as the shrinkage factor, and the percentage of fibers with a cross-sectional area greater than 2 standard deviations from the normal mean was used as the hypertrophy factor.

[0044] Based on a preset combination of multiple value ranges and judgment rules, the pathological pattern, disease type, activity, and severity are assessed by utilizing the muscle fiber subtype, morphological parameters, texture features, total number of cell nuclei, central nucleus ratio, fibrosis invasion gradient, and microvascular support index corresponding to each single muscle fiber, combined with the muscle fiber type aggregation factor, atrophy factor, and hypertrophy factor; and corresponding diagnosis and treatment strategies are extracted based on a preset diagnosis and treatment database.

[0045] On the other hand, the present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0046] On the other hand, the present invention also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0047] The muscle biopsy method and apparatus based on a multi-scale muscle biopsy topology network described in this invention accurately segment and identify muscle fibers, cell nuclei, connective tissue, and microvessels in digital muscle biopsy slices. Each muscle fiber is abstracted as a network node, constructing a topology network reflecting its spatial adjacency relationships. This quantifies the multi-dimensional intrinsic attributes of individual muscle fibers, such as subtype, morphology, texture, and nuclear distribution. Network-level parameters characterizing the overall pathological state of the tissue microenvironment are calculated, including muscle fiber type aggregation factors, fibrosis invasion gradient, and microvascular support index. This achieves a leap from isolated morphological feature analysis to comprehensive spatial topological network assessment. This invention transforms traditional pathological observation, which relies on subjective experience, into a computable and repeatable multi-scale quantitative analysis system. Its technical effect is reflected in its ability to objectively and accurately identify and quantify key pathological patterns such as fiber type aggregation in neurogenic damage, the degree of fibrosis invasion in muscular dystrophy, and local microcirculation status. This significantly improves the objectivity, consistency, and depth of muscle pathological diagnosis, providing structured, data-driven decision support for clinical diagnosis and research.

[0048] Additional advantages, objects, and features of the invention will be set forth in part in the description which follows, and will also become apparent in part to those skilled in the art upon studying the description, or may be learned by practice of the invention. The objects and other advantages of the invention can be realized and obtained by means of the structures specifically pointed out in the description and drawings.

[0049] Those skilled in the art will understand that the objectives and advantages achievable with the present invention are not limited to those specifically described above, and that the above and other objectives achievable with the present invention will become clearer from the following detailed description. Attached Figure Description

[0050] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, are not intended to limit the scope of the invention. In the drawings:

[0051] Figure 1 This is a schematic flowchart of a muscle biopsy method based on a multi-scale muscle biopsy topology network according to an embodiment of the present invention.

[0052] Figure 2 This is a flowchart illustrating the multi-scale muscle biopsy topology network analysis method according to an embodiment of the present invention.

[0053] Figure 3 This is a schematic flowchart of a muscle biopsy method based on a multi-scale muscle biopsy topology network of H&E and ATPase stained sections, as described in an embodiment of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.

[0055] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.

[0056] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0057] In the current field of muscle pathology diagnosis, conventional computer-aided analysis methods mostly focus on the automated measurement of basic morphological parameters of individual muscle fibers. While this improves efficiency to some extent, it remains at the level of isolated, static feature extraction, failing to quantitatively model the spatial organizational relationships between muscle fibers and their dynamic interactions with the surrounding microenvironment. This results in a lack of objective and calculable assessment indicators for pathological patterns crucial for the diagnosis of neuromuscular diseases. For example, the clustering distribution of muscle fiber types in neurogenic damage, the local invasive effect of fibrosis on muscle fibers in muscular dystrophy, and the microvascular support status in ischemic lesions all lack systematic detection methods. Therefore, current technology has not yet achieved a leap from single-fiber morphological measurement to systematic analysis of tissue spatial topology and microenvironment, making it difficult to comprehensively and accurately quantify the overall structure and functional state of complex muscle pathologies, thus limiting the objectivity, reproducibility, and revelation of deep pathological mechanisms in muscle biopsy diagnosis.

[0058] Therefore, this invention provides a muscle biopsy method based on a multi-scale muscle biopsy topology network, such as... Figure 1 As shown, the method includes the following steps S101~S105:

[0059] Step S101: Obtain digital slices of muscle biopsy of the sample to be analyzed after processing with the first staining scheme, and identify single muscle fibers, cell nuclei, connective tissue and microvessels and their masks in the digital slices of muscle biopsy based on a pre-trained image segmentation network.

[0060] Step S102: Align the mask to the digital slice of muscle biopsy of the sample to be analyzed after processing with multiple second staining schemes through image registration; calculate the optical feature value of a single muscle fiber and determine the muscle fiber subtype based on the first preset rule; perform geometric morphological analysis on each single muscle fiber and calculate multiple preset morphological parameters; extract texture features for each single muscle fiber using one or more texture analysis algorithms; and count the total number of cell nuclei and the proportion of the central nucleus within the preset muscle fiber mask area.

[0061] Step S103: Using a single muscle fiber as a node and determining its spatial representative point, the edges between the nodes are generated according to the coordinates of the spatial representative point using a preset spatial adjacency graph construction algorithm to obtain the muscle biopsy topology network; the muscle fiber subtype, morphological parameters, texture features, total number of cell nuclei, and central nucleus ratio of the single muscle fiber are added to the muscle biopsy topology network as node attributes to construct a graph data structure.

[0062] Step S104: Based on the myofibril subtype of each node in the graph data structure, calculate the myofibril type aggregation factor of each myofibril subtype according to the second preset rule; calculate the fibrosis invasion gradient of the degree of influence of the surrounding fibrosis environment on the single myofibril corresponding to each node based on the third preset rule; take microvessels as resource points, calculate the distance from each node to the surrounding resource points, and calculate the microvessel support index according to the fourth preset rule.

[0063] Step S105: Based on the myofiber subtype, morphological parameters, texture features, total number of nuclei, central nucleus ratio, fibrosis invasion gradient, and microvascular support index corresponding to each individual muscle fiber, a quantitative analysis report containing a global statistical summary is generated and visualized, combined with the myofiber type aggregation factor.

[0064] Specifically, step S101 aims to automatically identify four key biological targets from the raw digital pathology slides upon which subsequent analysis relies. In practice, firstly, whole-slice (WSI) digital images of muscle biopsies after a first staining protocol are acquired. For ease of processing, high-resolution WSIs are typically cut into a series of fixed-size, partially overlapping image blocks. To eliminate color differences between different slides caused by the staining process, color normalization algorithms such as the Macenko method can be used to preprocess all image blocks to achieve color consistency. Subsequently, the preprocessed image blocks are input into a pre-trained deep learning segmentation model. This model preferably employs advanced semantic segmentation or instance segmentation network architectures such as U-Net++ with attention mechanisms. Trained on a large amount of precisely labeled data by pathologists, it can output the probability of each pixel belonging to one of five categories: background, single muscle fiber, cell nucleus, connective tissue, and microvessels. By thresholding the probability map of muscle fiber categories and applying instance segmentation techniques such as the watershed algorithm, closely adjacent muscle fibers are separated into independent instances, thus obtaining a precise mask for each single muscle fiber. Similarly, binarized masks for cell nuclei, connective tissue regions, and microvessels can be obtained. The output of this step is four sets of instance-level or region-level masks spatially aligned with the original image.

[0065] In some embodiments, the pre-training step of the image segmentation network includes steps S1011-S1012:

[0066] Step S1011: Obtain a first training sample set containing multiple samples, each sample containing a muscle biopsy digital slice sample processed by a first staining protocol, and labeling the location of single muscle fibers, cell nuclei, connective tissue and microvessels as tags.

[0067] Step S1012: The initial image segmentation network is trained using the first training sample set. The initial image segmentation network takes the muscle biopsy digital slice sample as input and outputs the predicted location and predicted classification of single muscle fibers, cell nuclei, connective tissue and microvessels in the image. Based on the deviation between the predicted location, predicted classification and label, a loss is constructed to update the parameters of the initial image segmentation network to obtain the image segmentation network.

[0068] The initial image segmentation network includes a semantic segmentation network or an instance segmentation network based on a convolutional neural network architecture. The semantic segmentation network is either a U-Net network or a DeepLab network, and the instance segmentation network is a Mask R-CNN network.

[0069] In step S102, each independent myofiber instance segmented in step S101 is abstracted into a node, and its multidimensional features are comprehensively quantified. First, to determine the myofiber subtype (e.g., type I, type IIA), multimodal image registration is required. The mask on the slice processed by the first staining scheme is spatially aligned with digital slices of the same biopsy sample processed by multiple second staining schemes or other distinguishing staining methods using registration techniques such as affine transformation. It should be noted that the first staining scheme is mainly used to achieve image segmentation after staining. Its staining schemes include, but are not limited to, hematoxylin-eosin (H&E) staining. It should be considered that other staining schemes capable of achieving staining and structural differentiation can also be applied. After structural segmentation is completed in step S101, the obtained segmentation results are applied to step S102 through mask registration. The second staining scheme is mainly used for myofibril subtype classification and subsequent parameter calculation. The second staining scheme may include, but is not limited to, hematoxylin-eosin (HE) staining, modified Gomori staining, NADH-TR staining, succinate dehydrogenase (SDH) staining, cytochrome C oxidase (COX) staining, acid phosphatase (Acid) staining, glycogen (PAS) staining, Oil Red O staining, Sudan Black B staining, ATPase staining, and various immunohistochemical (IHC) stainings. In the process of classifying myofibril subtypes, different staining schemes can target specific antibody staining for myofibril type, connective tissue components, or vascular endothelial cells. Different staining schemes can be adaptively selected according to different subtype identification needs.

[0070] In some embodiments, the method employs a mask-guided registration network, MrRegNet, to align a mask to a digital slice of muscle biopsy of the sample to be analyzed, which has been processed with multiple second staining schemes.

[0071] Optical feature values ​​include average grayscale value and color channel intensity. The method determines the myofiber subtype based on a first preset rule, including: distinguishing the myofiber subtype of a single myofiber according to the preset range of average grayscale value and color channel intensity. Specifically, the average grayscale value of each myofiber mask in the corresponding area of ​​the aligned second staining scheme image can be calculated separately, and it can be classified into a specific subtype according to a preset grayscale threshold rule.

[0072] Secondly, a geometric morphological analysis is performed on each muscle fiber mask, calculating morphological parameters including cross-sectional area, perimeter, roundness, and major-minor axis ratio. The total number of pixels within the mask is taken as the cross-sectional area of ​​a single muscle fiber, and the total number of pixels at the mask boundary is taken as the perimeter of a single muscle fiber. Roundness is calculated using the following formula:

[0073] ;

[0074] Wherein, C represents the circularity, A represents the cross-sectional area, and P represents the perimeter;

[0075] The polygonal degree is calculated using the following formula:

[0076] ;

[0077] Where Poly represents the polygon degree, and A represents the cross-sectional area. This represents the convex hull area of ​​the mask corresponding to the single muscle fiber.

[0078] Furthermore, to quantify internal pathological changes, texture analysis algorithms, such as calculating the gray-level co-occurrence matrix (GLCM), are applied to the gray-level image region within the myofibril mask, and features such as contrast, energy, and homogeneity are extracted from it to objectively characterize the severity of lesions such as "worm-eaten" and "rimmed vacuoles".

[0079] Finally, the spatial distribution of cell nuclei was analyzed, and the total number of cell nuclei that completely fall within the myofibril mask was counted. The central region of the myofibril was defined by morphological erosion, and the number of cell nuclei located in this central region was counted. The difference between the two yielded the central nucleus ratio, a key pathological indicator.

[0080] In some embodiments, one or more texture analysis algorithms are used to extract texture features for each individual muscle fiber, including steps S1021 to S1023:

[0081] Step S1021: Extract the grayscale image within the mask of a single muscle fiber and calculate the grayscale co-occurrence matrix.

[0082] Step S1022: Calculate contrast texture features, the calculation formula is:

[0083] .

[0084] Step S1023: Calculate the energy texture features, the calculation formula is:

[0085] .

[0086] Where i and j represent the grayscale values ​​of pixel pairs with a preset spatial relationship. This represents the joint probability of a pixel pair with gray values ​​i and j appearing in the mask region.

[0087] The method also includes a pre-trained pathology classifier that identifies the probability of a specified pathology presence and its severity level based on contrast texture features and energy texture features.

[0088] The pre-training steps of the pathology classifier include: obtaining a second training sample set containing multiple samples, each sample containing a digital slice of muscle biopsy stained with a second staining scheme, and labeling the sample contrast texture features and sample energy texture features of the grayscale image of the mask region corresponding to a single muscle fiber, and labeling the pathology type and severity level as labels; training a support vector machine using the second training sample set to obtain the pathology classifier.

[0089] In step S103, discrete muscle fiber nodes are connected according to their spatial proximity relationships to construct a topological network describing the micro-spatial structure of the tissue. First, each muscle fiber node is spatially located by calculating the geometric centroid coordinates of its mask as the spatial representative point of that node. Then, based on the set of centroid coordinates of all nodes, a spatial adjacency graph construction algorithm is used to define the edges between nodes. The preferred algorithm is Delaunay triangulation, which generates a triangular mesh that naturally reflects the nearest neighbor relationship of points in space; that is, only spatially adjacent muscle fibers are connected by an edge. After construction, a graph data structure G=(V,E) is obtained, where the vertex set V is all muscle fiber nodes, and the edge set E is composed of edges generated by Delaunay triangulation. In other embodiments, the Voronoi graph algorithm or a radius-based proximity graph algorithm can be used in combination with the coordinates of the spatial representative point to generate the edges between the nodes, thus obtaining the muscle biopsy topological network. Finally, the attributes such as myofibril subtype, morphological parameters, texture feature values, total number of cell nuclei, and central nucleus ratio calculated for each node in step S102 are used as the attribute features of that node and added to the graph data structure to form an attribute graph containing rich attribute information.

[0090] In step S104, based on the constructed attribute topology network, three core organizational microenvironment quantification parameters are calculated.

[0091] First, calculate the muscle fiber type clustering factor: Based on the proportion of neighbors of the same muscle fiber subtype, calculate the average proportion of neighbors of the same type for all nodes of a specified type, thus obtaining the muscle fiber type clustering factor that quantifies the degree of clustering of that type of fiber. For any given muscle fiber subtype k in the network, traverse all nodes belonging to that type. For any node v... i Find all directly adjacent neighbor nodes of node v in the topology network, and calculate the proportion of nodes of type k among these neighbors, i.e., node v i The local aggregation degree is then calculated. The average local aggregation degree of all k-type nodes is then taken to obtain the myofibril type aggregation factor for the k-type fibers in the slice. A higher value indicates a more significant spatial aggregation of this type of fiber.

[0092] Second, calculate the fibrosis invasion gradient: For each pixel in the single muscle fiber mask region, calculate its distance to the nearest connective tissue to construct a pressure field; calculate the integral or average of the pressure field values ​​of all pixels on the boundary contour of the single muscle fiber mask region to obtain the fibrosis invasion gradient. The connective tissue mask obtained in step S101 is considered as the fibrosis source. Calculate the distance transformation map of the entire slice to obtain the Euclidean distance field D(p) from each pixel to the nearest connective tissue. Then, define a decaying pressure field. For each muscle fiber node, extract all pixels on its boundary contour and calculate the average value of the pressure field P(p) at these points. This average value is the fibrosis invasion gradient F experienced by the fiber. i The larger the value, the more severe the compression by connective tissue.

[0093] Third, calculate the microvessel support index: count the number of resource points within a set distance range from the node as the microvessel support index. The centroid of the microvessel instance identified in step S101 is considered as the resource point. For each myofiber node v i Calculate the distance d from its centroid to the nearest microvascular centroid. min,i Simultaneously, the number of microvessels n contained within a circular region of a preset radius R centered on its centroid is counted. R,i The microvascular support index of this node can then be defined as... This index takes into account both the proximity and density of blood vessels; a higher value indicates a better local blood supply environment.

[0094] Step S105 is responsible for integrating and displaying the analysis results at all levels. The system first summarizes global statistical information and generates a text summary, including but not limited to: the total number of muscle fibers, the number and proportion of each type of fiber, the average cross-sectional area and its distribution (which can calculate atrophy and hypertrophy factors), the average central nucleus proportion, and the aggregation factors of each type of fiber, the average fibrosis invasion gradient, and the average microvascular support index calculated in step S104. Secondly, data visualization is performed: histograms of the distribution of morphological parameters such as muscle fiber cross-sectional area and roundness are generated; more importantly, the fibrosis invasion gradient or microvascular support index value of each muscle fiber node is mapped to color and overlaid as a heatmap on the original tissue image background processed by the first staining scheme, thus intuitively and locally displaying hotspots of fibrosis invasion or areas with poor blood supply. Finally, all analytical data, including all attribute vectors of each muscle fiber node and their calculated microenvironment parameters, are exported in a structured format, providing a complete data foundation for subsequent in-depth statistical research or machine learning modeling. Through the above steps, this invention achieves the automated generation of multi-level, computable, and interpretable quantitative analysis reports from raw pathological images.

[0095] In some embodiments, the method further includes: calculating the proportion of fibers with a cross-sectional area less than two standard deviations of the normal mean as a shrinkage factor, and the proportion of fibers with a cross-sectional area greater than two standard deviations of the normal mean as a hypertrophy factor.

[0096] Based on multiple preset value ranges and judgment rules, the pathological pattern, disease type, activity, and severity are assessed by utilizing the muscle fiber subtype, morphological parameters, texture features, total number of cell nuclei, central nucleus ratio, fibrosis invasion gradient, and microvascular support index corresponding to each single muscle fiber, combined with muscle fiber type aggregation factor, atrophy factor, and hypertrophy factor; and corresponding diagnosis and treatment strategies are extracted based on a preset diagnosis and treatment database.

[0097] This embodiment introduces two quantitative indicators based on statistical distribution—atrophy factor and hypertrophy factor—to further transform morphological parameters into clinically significant assessment dimensions with clear pathological implications. Furthermore, by systematically integrating multi-scale extracted fiber properties, microenvironment parameters, and global statistical factors through pre-defined rules, a comprehensive evaluation system from micromorphology to macroscopic tissue patterns is constructed. This achieves automated and structured judgment of muscle pathological patterns, disease types, activity, and severity. Finally, by linking to a diagnostic and treatment database, the system can intelligently match or recommend corresponding diagnostic and treatment strategies based on the assessment results. This completes a closed-loop extension from quantitative analysis to auxiliary diagnostic and treatment decision support, significantly enhancing the clinical translational value and practicality of the method.

[0098] On the other hand, the present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0099] On the other hand, the present invention also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0100] The present invention will now be described in conjunction with specific embodiments:

[0101] Example 1: This example provides a multi-scale muscle biopsy topology network analysis method, such as... Figure 2 As shown, the steps are as follows: S1~S5:

[0102] Step S1: Multi-class target segmentation based on digital pathology images

[0103] The purpose of this step is to automatically identify and extract multiple target categories with pathological significance from the input digital pathological slide images of muscle biopsy.

[0104] First, the system acquires one or more digital pathological slide images of muscle biopsies as input. These images can be derived from various staining techniques, including but not limited to hematoxylin-eosin (HE) staining, modified Gomori staining, NADH-TR staining, succinate dehydrogenase (SDH) staining, cytochrome C oxidase (COX) staining, acid phosphatase (Acid) staining, glycogen (PAS) staining, Oil Red O staining, Sudan Black B staining, ATPase staining, and various immunohistochemical (IHC) stainings. Next, the system processes the input images using a multi-class image segmentation algorithm. Preferably, this algorithm is based on a deep learning segmentation model, such as semantic segmentation networks (e.g., U-Net, DeepLab series) or instance segmentation networks (e.g., Mask R-CNN) within a convolutional neural network (CNN) architecture. This model, pre-trained, can simultaneously identify and delineate the precise boundaries of the following target regions: (1) single muscle fibers; (2) cell nuclei; (3) connective tissue of the endomysium or perimysium; and (4) microvessels.

[0105] The output of the segmentation process is a set of label masks aligned with the pixels of the original image, where each pixel is assigned a category label, and sets of pixels belonging to the same muscle fiber or the same cell nucleus are identified as independent instances.

[0106] Step S2: Quantification of the multidimensional intrinsic properties of myofibril nodes

[0107] This step abstracts each individual myofibril instance segmented in step S1 into a node in the pathological microenvironment network, and performs comprehensive quantitative characterization on this node. For each myofibril node in the network, the system calculates and extracts a multi-dimensional attribute vector, which contains at least the following four types of information:

[0108] Myofibril subtype attributes: If the input contains a staining image that can distinguish fiber types (such as ATPase or IHC), the system will align the image with the image used for structural segmentation using image registration technology. Subsequently, by analyzing the optical properties (such as average gray value and color channel intensity) of the region corresponding to each myofibril node, it will automatically classify it into a specific physiological or pathological subtype.

[0109] Morphological parameters: The system performs geometric morphological analysis on the mask of each muscle fiber node, calculating a series of morphological descriptors. These include, but are not limited to: cross-sectional area (CSA), perimeter, minimum Freette diameter (minor axis), major-minor axis ratio, and circularity and polygonality used to describe shape regularity.

[0110] Internal texture pathological feature parameters: The system performs texture analysis on the pixel regions within each myofiber node to quantify pathological changes that are difficult to describe through macroscopic morphology. This can be achieved by applying one or more texture analysis algorithms, such as extracting statistical features like contrast, energy, and homogeneity based on the gray-level co-occurrence matrix (GLCM), or extracting frequency domain features using wavelet transform and Gabor filters. These features can objectively reflect the degree of pathological changes such as "worm-eaten" and "rimmed vacuolation."

[0111] Nuclear spatial distribution parameters: The system analyzes the spatial distribution of nuclear instances within each myofibril node region. By calculating the total number of nuclei falling within the myofibril mask and the number of nuclei located in the central region of the fiber, key pathological indicators such as "central nuclear percentage" are obtained.

[0112] Step S3: Construction of the pathological topology network

[0113] This step aims to establish spatial adjacency relationships between muscle fiber nodes, organizing discrete nodes into a network with a topological structure.

[0114] First, the system determines a spatial representative point for each muscle fiber node, typically its geometric centroid coordinates. Then, based on the set of representative points for all nodes, a spatial adjacency graph construction algorithm is used to generate the network edges. Preferred algorithms include, but are not limited to, Delaunay triangulation, Voronoi diagrams, or radius-based proximity graphs. Delaunay triangulation is preferred because it effectively reflects the "nearest neighbor" property in space.

[0115] After construction, a graph data structure G=(V, E) is obtained, where V is the set of all muscle fiber nodes and E is the set of edges representing the spatial adjacency relationships between nodes.

[0116] Step S4: Calculation of key parameters of the pathological microenvironment

[0117] This step, based on the constructed topological network, performs a higher-level analysis and calculates a series of quantitative parameters that can reveal tissue-level pathological patterns.

[0118] Fiber type clustering factor: For each node in the network, the system examines the fiber subtype attribute of all its directly connected neighboring nodes. By calculating the proportion of "same type neighbors" and taking a statistic (such as the mean) of that proportion for all nodes of a specific type, a global score quantifying the degree of fiber clustering for that type is obtained.

[0119] Fibrosis invasion gradient: The system models the connective tissue region segmented in step S1 as a spatial "pressure field". This field can be generated by calculating the distance transformation from each pixel in the image to the nearest connective tissue region. For each muscle fiber node, the system integrates or averages the value of this pressure field along its boundary contour to obtain a value that quantifies the degree to which the fiber is affected by the surrounding fibrotic environment.

[0120] Microvascular Support Index: The system treats the microvessels segmented in step S1 as "resource points" in space. For each muscle fiber node, the system calculates its spatial relationship with these resource points, such as the distance to the nearest microvessel and the number of microvessels within a preset radius. These measures can be integrated into a comprehensive index to assess the local blood supply microenvironment of the muscle fiber.

[0121] Step S5: Generation of Structured Quantitative Analysis Report

[0122] This step is the final output of the method, responsible for integrating and presenting all the analysis results.

[0123] The system summarizes the single fiber attribute statistics extracted in step S2 (e.g., average CSA distribution histogram of various fibers, atrophy factor, hypertrophy factor, central nucleus ratio, etc.) with the tissue-level microenvironment parameters calculated in step S4.

[0124] The final report is provided in one or more formats, including: 1) Text summary: containing numerical summaries of key indicators and automatically generated descriptive text. 2) Data visualization: including statistical charts of various parameters (such as histograms and box plots), and rendering microenvironmental parameter values ​​(such as fibrosis invasion gradients) as pseudo-color images and overlaying them onto the original pathological images to achieve intuitive visualization of pathological hotspots. 3) Machine-readable data: exporting detailed attributes of all nodes and calculated microenvironmental parameters in standard data formats (such as CSV or JSON) for subsequent scientific research and data mining.

[0125] Example 2: A muscle biopsy method based on a multi-scale muscle biopsy topology network using H&E and ATPase dual-staining sections is provided. This example uses a set of whole-slide images (WSI) of muscle biopsies containing standard H&E staining and ATPase staining as the processing object, such as... Figure 3 As shown, the implementation steps S1 to S5 of the method in this embodiment are described in detail.

[0126] Step S1: Multi-class image segmentation and recognition

[0127] The goal of this step is to accurately segment four types of targets from WSI: muscle fibers, cell nuclei, connective tissue, and microvessels.

[0128] 1. Image preprocessing:

[0129] WSI Reading and Tiling: Read high-resolution WSIs and divide them into multiple overlapping, fixed-size patches, such as 1024x1024 pixels, to fit the input requirements of deep learning models.

[0130] Color normalization: To eliminate color differences caused by different batches of coloring, the Macenko method was used to normalize the colors of all tiles to make them have a consistent color space.

[0131] 2. Training and Application of Deep Learning Segmentation Models:

[0132] Model selection: The U-Net++ network structure with attention mechanism was selected. This model performs well in medical image segmentation tasks and can effectively integrate multi-scale features.

[0133] Training data: Pathology experts manually labeled muscle fibers, cell nuclei, and connective tissue regions on a large number of maps as the "gold standard" (Ground Truth). Microvessel labeling can be aided by referring to the results of CD31 immunohistochemical staining.

[0134] Model training: The input model is trained using tile-label pairs. The loss function is a weighted sum of Dice loss and cross-entropy loss to balance the accuracy of region segmentation and the precision of pixel classification. ,in, Focus on the overlap of the segmented regions, and Focus on pixel-level classification accuracy. and This is a hyperparameter.

[0135] Segmentation reasoning: Input the image patch to be analyzed into the trained model, and the model outputs a multi-channel probability map, where each channel represents the probability of a pixel belonging to the corresponding category (background, muscle fiber, cell nucleus, connective tissue, microvessel).

[0136] 3. Instance segmentation and post-processing:

[0137] A binarized myofiber mask is obtained by applying a threshold (e.g., 0.5) to the myofiber probability map.

[0138] Since muscle fibers are closely connected, instance segmentation is required to distinguish individual fibers. The watershed algorithm is used to first calculate the distance transformation map of the muscle fiber mask, where the peak points can be used as seed points. Then, the watershed algorithm can be applied to separate individual muscle fiber instances.

[0139] Thresholding is also applied to the probability maps of other categories to obtain their respective masks.

[0140] Step S2: Quantification of muscle fiber node attributes

[0141] For each muscle fiber instance (i.e. node) segmented in S1, extract its multidimensional intrinsic properties.

[0142] 1. Myofibril subtype identification:

[0143] Align the H&E slices with the registered ATPase-stained slices. For each muscle fiber mask in the H&E slice, locate its corresponding region on the ATPase slice. Calculate the average pixel gray value within that region. Based on a preset gray value threshold, classify the muscle fibers into deeply stained (Type I), lightly stained (Type IIA), or moderately stained (Type IIB).

[0144] 2. Morphological parameter calculation:

[0145] Cross-sectional area (CSA): ,in, It is the total number of pixels within the mask. Represents the area of ​​a single pixel.

[0146] Perimeter (P): Calculates the total length of pixels at the mask boundary.

[0147] Circularity: The closer the value is to 1, the closer the shape is to a circle.

[0148] Polygonality: ,in, This is the convex hull area of ​​the fiber mask. This value reflects the sharpness of the fiber profile.

[0149] 3. Quantification of internal texture pathological feature parameters:

[0150] For each grayscale image region within the muscle fiber mask, calculate its gray-level co-occurrence matrix (GLCM). Extract texture features based on the GLCM, for example:

[0151] Contrast: It reflects the depth and clarity of the grooves in the image and can be used to quantify "worm-eaten" changes.

[0152] Energy (ASM): It reflects the uniformity of the grayscale distribution of an image.

[0153] By training a small classifier (such as SVM), these texture features can be used to identify and quantify the probability or severity of the presence of specific pathological changes such as "rimmed vacuoles" and "target-like fibers".

[0154] 4. Calculation of spatial distribution parameters of cell nuclei:

[0155] For each muscle fiber mask, count the number of centroids of the nuclear masks contained within it, denoted as . .

[0156] The central region of the muscle fiber mask is defined by performing morphological erosion. The number of cell nuclei whose centroids fall into the central region is counted and denoted as . Centralized core share: .

[0157] Step S3: Construction of Pathological Topology Network

[0158] Node definition: Each muscle fiber instance that has completed attribute quantization in S2 is treated as a node in the network. .

[0159] Node localization: Calculate each muscle fiber node Geometric centroid coordinates .

[0160] Topology network construction, based on the centroid coordinate set of all nodes As input, the Delaunay triangulation algorithm is applied. This algorithm constructs a triangular mesh such that no node lies inside the circumcircle of any triangle. The edges of this triangular mesh constitute the edges of the topology network of this invention. If a node... and In a triangular mesh, if two objects are connected by an edge, they are considered to be topologically adjacent.

[0161] Step S4: Calculation of pathological microenvironment parameters

[0162] 1. Myofiber type aggregation factor (Grouping Factor) ):

[0163] For any type in the network nodes Get the set of all its neighboring nodes. .calculate The medium type is also Number of nodes .node The local aggregation degree is .

[0164] Types in the entire slice fiber aggregation factor The average local clustering degree of all nodes of this type: ,in, Is all types A set of nodes. The higher the value, the more pronounced the aggregation of this type of muscle fiber.

[0165] 2. Fibrotic invasion gradient ):

[0166] The connective tissue mask obtained in S1 is considered as a source of fibrosis. The Euclidean distance transform (EDT) of the entire image is calculated to obtain a distance field. Each pixel The value is its distance to the nearest connective tissue pixel.

[0167] Define an invading pressure field The field value is 1 within connective tissue and decreases smoothly outwards. For each muscle fiber node... Obtain the set of its boundary pixels. .

[0168] node The fibrosis invasion gradient is the average value of the invasion pressure experienced at its boundary: . The higher the value, the more severe the compression of the muscle fiber by the surrounding connective tissue.

[0169] 3. Vascular Support Index ):

[0170] Obtain the set of centroid coordinates of all microvessel instances in S301. For each muscle fiber node Its center of mass is Calculate the distance to the nearest microvascular centroid: .

[0171] Calculate a preset radius For example, the number of microvessels contained within 100 micrometers .node The microvascular support index is defined as: This index takes into account both the number and distance of surrounding blood vessels; a higher value indicates a better local blood supply environment.

[0172] Step S5: Quantitative Analysis Report Generation

[0173] Generate a global statistical summary, outputting the total number of muscle fibers, the number and proportion of type I / II fibers, and the average CSA.

[0174] The percentage of fibers with a cross-sectional area less than two standard deviations from the normal mean was used as the shrinkage factor, and the percentage of fibers with a cross-sectional area greater than two standard deviations from the normal mean was used as the hypertrophy factor.

[0175] Output the global average central kernel percentage and the global average clustering factor. Invasion gradient and support index .

[0176] Perform data visualization, generating histogram distributions of various morphological parameters (such as CSA and roundness). Generate pseudo-color maps (Heatmaps): on the original tissue images, based on each myofibril node... or The values ​​are colored to visually display the areas most severely affected by fibrosis or areas with poor blood supply.

[0177] Perform data export, exporting all quantitative attributes of all nodes and calculated microenvironment parameters in tabular form (such as CSV file) to facilitate further statistical analysis and scientific research.

[0178] Through the above specific implementation methods, complex muscle biopsy images can be transformed into a structured data asset that can be deeply analyzed, thereby achieving an objective, accurate, and comprehensive quantitative assessment of muscle pathological conditions.

[0179] Corresponding to the above method, the present invention also provides an apparatus / system including a computer device, the computer device including a processor and a memory, the memory storing computer instructions, the processor executing the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the apparatus / system performs the steps of the method as described above.

[0180] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the aforementioned edge computing server deployment method. The computer-readable storage medium can be a tangible storage medium, such as random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, floppy disks, hard disks, removable storage disks, CD-ROMs, or any other form of storage medium known in the art.

[0181] In summary, the muscle biopsy method and apparatus based on a multi-scale muscle biopsy topology network described in this invention achieves multi-category precise segmentation and identification of muscle fibers, cell nuclei, connective tissue, and microvessels in digital muscle biopsy slices. Each muscle fiber is abstracted as a network node to construct a topology network reflecting its spatial adjacency relationships. This quantifies the multi-dimensional intrinsic attributes of individual muscle fibers, such as subtype, morphology, texture, and nuclear distribution. Network-level parameters characterizing the overall pathological state of the tissue microenvironment, including muscle fiber type aggregation factors, fibrosis invasion gradient, and microvascular support index, are calculated. This represents a leap from isolated morphological feature analysis to comprehensive spatial topological network assessment. This invention transforms traditional pathological observation, which relies on subjective experience, into a computable and repeatable multi-scale quantitative analysis system. Its technical effectiveness lies in its ability to objectively and accurately identify and quantify key pathological patterns such as fiber type aggregation in neurogenic damage, the degree of fibrosis invasion in muscular dystrophy, and local microcirculation status. This significantly improves the objectivity, consistency, and depth of muscle pathological diagnosis, providing structured, data-driven decision support for clinical diagnosis and research.

[0182] Those skilled in the art will understand that the exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention. When implemented in hardware, it can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the desired tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave.

[0183] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0184] In this invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or in place of features of other embodiments.

[0185] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations of the embodiments of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A muscle biopsy method based on a multi-scale muscle biopsy topological network, characterized by, The method includes the following steps: A digital slice of muscle biopsy is obtained from the sample to be analyzed after processing with a first staining scheme that facilitates image segmentation. Based on a pre-trained image segmentation network, single muscle fibers, cell nuclei, connective tissue, microvessels, and their masks in the digital slice of muscle biopsy are identified. Image registration is used to align the mask to the digital slice of muscle biopsy of the sample to be analyzed, which has been processed by a second staining scheme that facilitates feature recognition; optical feature values ​​of individual muscle fibers are calculated and muscle fiber subtypes are determined based on a first preset rule; geometric morphological analysis is performed on each individual muscle fiber to calculate a number of preset morphological parameters; texture features are extracted from each individual muscle fiber using one or more texture analysis algorithms; the total number of cell nuclei and the proportion of central nuclei within the preset muscle fiber mask area are statistically analyzed. Using the single muscle fiber as a node and determining its spatial representative point, the edges between the nodes are generated according to the coordinates of the spatial representative point using a preset spatial adjacency graph construction algorithm to obtain a muscle biopsy topology network; the muscle fiber subtype, morphological parameters, texture features, total number of cell nuclei, and central nucleus ratio of the single muscle fiber are added to the muscle biopsy topology network as node attributes to construct a graph data structure; Based on the myofibril subtype of each node in the graph data structure, calculate the myofibril type aggregation factor for each myofibril subtype according to the second preset rule; calculate the fibrosis invasion gradient of the degree of influence of the surrounding fibrosis environment on the single myofibril corresponding to each node based on the third preset rule; take the microvessels as resource points, calculate the distance from each node to the surrounding resource points, and calculate the microvessel support index according to the fourth preset rule. Based on the myofiber subtype, morphological parameters, texture features, total number of nuclei, central nucleus ratio, fibrosis invasion gradient, and microvascular support index corresponding to each individual myofiber, a quantitative analysis report containing a global statistical summary is generated and visualized, in conjunction with the myofiber type aggregation factor. Specifically, based on the myofiber subtype of each node in the graph data structure, a myofiber type aggregation factor for each myofiber subtype is calculated according to a second preset rule, including: Based on the proportion of neighbors of the same type according to the muscle fiber subtype of the neighboring nodes, the mean of the proportion of neighbors of the same type is calculated for all nodes of the specified type, and a muscle fiber type aggregation factor is obtained to quantify the degree of aggregation of this type of fiber. The fibrosis invasion gradient of the degree of influence of the surrounding fibrotic environment on the single muscle fiber corresponding to each node is calculated based on the third preset rule, including: for each pixel in the single muscle fiber mask region, the distance to the nearest connective tissue is calculated to construct a pressure field; the integral or average value of the pressure field values ​​of all pixels on the boundary contour of the single muscle fiber mask region is calculated to obtain the fibrosis invasion gradient. Using the microvessels as resource points, the distance from each node to surrounding resource points is calculated, and the microvessel support index is calculated according to the fourth preset rule, including: counting the number of resource points within a set distance range from the node as the microvessel support index.

2. The muscle biopsy method based on a multi-scale muscle biopsy topological network according to claim 1, characterized in that, The pre-training steps of the image segmentation network include: Obtain a first training sample set containing multiple samples, each sample containing a muscle biopsy digital slice sample processed by a first staining protocol, and labeling the location of single muscle fibers, cell nuclei, connective tissue and microvessels as tags; The initial image segmentation network is trained using the first training sample set. The initial image segmentation network takes the muscle biopsy digital slice sample as input and outputs the predicted location and predicted classification of single muscle fibers, cell nuclei, connective tissue and microvessels in the image. The initial image segmentation network is updated by constructing a loss based on the deviation between the predicted location, the predicted classification and the label to obtain the image segmentation network. The initial image segmentation network includes a semantic segmentation network or an instance segmentation network based on a convolutional neural network architecture. The semantic segmentation network is a U-Net network or a DeepLab network, and the instance segmentation network is a Mask R-CNN network.

3. The muscle biopsy method based on a multi-scale muscle biopsy topological network according to claim 1, characterized in that, After obtaining the muscle biopsy digital slices of the sample to be analyzed after processing with the first staining protocol, the Macenko method was used to normalize the color of all patches. The method further includes: using a mask-guided registration network MrRegNet to align the mask to the digital slices of muscle biopsy of the sample to be analyzed after processing with multiple second staining schemes; The optical feature values ​​include average gray value and color channel intensity. The method determines the muscle fiber subtype based on a first preset rule, including: distinguishing the muscle fiber subtype of the single muscle fiber according to the preset value range of average gray value and color channel intensity.

4. The muscle biopsy method based on a multi-scale muscle biopsy topology network according to claim 1, characterized in that, Geometric morphological analysis was performed on each of the individual muscle fibers to calculate a number of preset morphological parameters, including: The total number of pixels within the mask is used as the cross-sectional area of ​​the single muscle fiber, and the total number of pixels at the mask boundary is used as the perimeter of the single muscle fiber. The formula for calculating roundness is: ； Wherein, C represents the circularity, A represents the cross-sectional area, and P represents the perimeter; The polygonal degree is calculated using the following formula: ； Where Poly represents the polygon degree, and A represents the cross-sectional area. This represents the convex hull area of ​​the mask corresponding to the single muscle fiber.

5. The muscle biopsy method based on a multi-scale muscle biopsy topology network according to claim 1, characterized in that, One or more texture analysis algorithms are used to extract texture features for each of the individual muscle fibers, including: Extract the grayscale image within the single muscle fiber mask and calculate the grayscale co-occurrence matrix; The formula for calculating contrast texture features is: ； The energy texture features are calculated using the following formula: ； Where i and j represent the grayscale values ​​of pixel pairs with a preset spatial relationship. This represents the joint probability of a pixel pair with gray values ​​i and j appearing in the mask region. The method further includes: identifying the probability of presence and severity level of a specified pathology based on a pre-trained pathology classifier according to the contrast texture features and the energy texture features; The pre-training steps of the pathology classifier include: obtaining a second training sample set containing multiple samples, each sample containing a muscle biopsy digital slice sample processed by multiple second staining schemes, and labeling the sample contrast texture features and sample energy texture features of the grayscale image of the mask region corresponding to a single muscle fiber, and labeling the pathology type and severity level as labels; using the second training sample set to train a support vector machine to obtain the pathology classifier.

6. The method of claim 1, wherein the method further comprises: The method further includes: extracting the geometric centroid of the single muscle fiber as the spatial representative point, and obtaining the coordinates of the geometric centroid; The method uses the Delaunay triangulation algorithm, Voronoi graph algorithm, or radius-based nearest neighbor graph algorithm to generate the edges between the nodes by combining the coordinates of the spatial representative points, thus obtaining the muscle biopsy topology network.

7. The muscle biopsy method based on a multi-scale muscle biopsy topology network according to claim 1, characterized in that, The method further includes: The percentage of fibers with a cross-sectional area less than 2 standard deviations from the normal mean was used as the shrinkage factor, and the percentage of fibers with a cross-sectional area greater than 2 standard deviations from the normal mean was used as the hypertrophy factor. Based on a preset combination of multiple value ranges and judgment rules, the pathological pattern, disease type, activity, and severity are assessed by utilizing the muscle fiber subtype, morphological parameters, texture features, total number of cell nuclei, central nucleus ratio, fibrosis invasion gradient, and microvascular support index corresponding to each single muscle fiber, combined with the muscle fiber type aggregation factor, atrophy factor, and hypertrophy factor; and corresponding diagnosis and treatment strategies are extracted based on a preset diagnosis and treatment database.

8. A computer readable storage medium having stored thereon a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method as described in any one of claims 1 to 7.

9. A computer program product comprising computer programs or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 7.

Images