Bone microstructure parameter extraction method, computing device, computer readable storage medium and computer program product

Abstract

The embodiment of the invention provides a bone microstructure parameter extraction method, computing equipment, a computer readable storage medium and a computer program product. The bone microstructure parameter extraction method comprises the following steps: firstly, carrying out super-resolution reconstruction on a CT (Computed Tomography) image to enhance the image quality; then, a first bone mask is generated through a global threshold value, a local adaptive threshold value is constructed in combination with local gray statistical characteristics, and a mask edge and a low-contrast region are subjected to reclassification to obtain a more accurate second bone mask; and then, noise connected domain elimination under volume constraint is performed on the second bone mask, and a fracture or a pseudo joint of the bone trabecula network is repaired based on morphological characteristics to obtain a high-fidelity bone tissue structure segmentation result. And finally, cortical bones are removed from the result to separate out cancellous bones, and bone microstructure parameters are extracted from the cancellous bones. According to the technical scheme provided by the embodiment of the invention, the accuracy of bone microstructure parameter extraction is improved.

Description

Technical Field

[0001] This application relates to the field of image segmentation technology, and in particular to a method for extracting bone microstructure parameters, a computing device, a computer-readable storage medium, and a computer program product. Background Technology

[0002] Skeletal system diseases (such as osteoarthritis and osteoporosis) are a major global public health problem. Most serious consequences, such as fragility fractures, are rooted in the degeneration of bone tissue microstructure, such as the reduction or disruption of the number (Tb.N), thickness (Tb.Th), spacing (Tb.Sp), and connectivity (Conn.D) of trabeculae in cancellous bone.

[0003] However, existing clinical gold standards, such as bone mineral density (BMD) measurements (e.g., dual-energy X-ray absorptiometry, DEXA) and commonly used risk assessment models, mainly reflect the mineral content of bone tissue. They are difficult to fully extract bone as a structural parameter to reflect the complex changes in the bone microstructure network, resulting in a significant blind spot in fracture prediction in people with normal BMD or reduced bone mass.

[0004] Therefore, how to accurately extract bone microstructure parameters has become an urgent technical problem to be solved. Summary of the Invention

[0005] This application provides a method for extracting bone microstructure parameters, a computing device, a computer-readable storage medium, and a computer program product.

[0006] In a first aspect, embodiments of this application provide a method for extracting bone microstructure parameters, including: Acquire ultra-high resolution CT images of the bone tissue to be examined; Super-resolution reconstruction is performed on the ultra-high resolution CT image to obtain a reconstructed ultra-high resolution CT image; The global grayscale distribution features of the reconstructed ultra-high resolution CT image are calculated to determine the global segmentation threshold, and a first bone mask is generated on the reconstructed ultra-high resolution CT image based on the global segmentation threshold to obtain a bone mask ultra-high resolution CT image. Local gray-level statistical features are calculated for voxels in the ultra-high resolution CT image of the bone mask, and a local adaptive threshold is constructed based on the local gray-level statistical features. The edge regions or low-contrast regions in the first bone mask are reclassified using the local adaptive threshold to obtain the second bone mask. Based on the preset volume constraints, isolated noise connected domains are removed from the second bone mask, and the fractures or pseudo-connections of the trabecular network are repaired based on the morphological characteristics of the bone tissue to obtain the bone tissue structure segmentation results. The cortical bone is removed from the bone tissue segmentation results to obtain the cancellous bone segmentation results; Bone microstructure parameters were extracted from the cancellous bone segmentation results.

[0007] Secondly, embodiments of this application provide a device for extracting bone microstructure parameters, comprising: The first acquisition module is used to acquire ultra-high resolution CT images of the bone tissue to be detected. The reconstruction module is used to perform super-resolution reconstruction on the ultra-high resolution CT image to obtain a reconstructed ultra-high resolution CT image. The first mask extraction module is used to calculate the global grayscale distribution features of the reconstructed ultra-high resolution CT image to determine the global segmentation threshold, and generate a first bone mask on the reconstructed ultra-high resolution CT image based on the global segmentation threshold to obtain a bone mask ultra-high resolution CT image. The second mask extraction module is used to calculate local gray-level statistical features of voxels in the ultra-high resolution CT image of the bone mask, and to construct a local adaptive threshold based on the local gray-level statistical features. The third mask extraction module is used to reclassify the edge region or low contrast region in the first bone mask using the local adaptive threshold to obtain the second bone mask. The first result generation module is used to remove isolated noise connected domains from the second bone mask based on preset volume constraints, and repair the fractures or pseudo-connections of the trabecular network based on the morphological characteristics of bone tissue to obtain the bone tissue structure segmentation result. The removal module is used to remove cortical bone from the bone tissue segmentation results to obtain cancellous bone segmentation results; The parameter extraction module is used to extract bone microstructure parameters from the cancellous bone segmentation results.

[0008] Thirdly, this application provides a computing device, including a processing component and a storage component; The storage component stores a computer program; the computer program is invoked and executed by the processing component to implement the bone microstructure parameter extraction method provided in this application embodiment.

[0009] Fourthly, this application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processing component, implements the method for extracting bone microstructure parameters provided in this application.

[0010] Fifthly, this application provides a computer program product, including a computer program or instructions, which, when executed by a processing component, implements the bone microstructure parameter extraction method provided in this application.

[0011] This application employs a super-resolution reconstruction step to enhance the original CT images with high fidelity, effectively suppressing noise, blurring, and artifacts introduced by low-dose scanning. This significantly improves the sharpness and detail of trabecular and cortical bone boundaries. Secondly, a gray-scale-spatial consistency fusion strategy is used in the segmentation stage. Efficient initial separation is achieved through a global threshold, followed by accurate reclassification of edges and low-contrast regions using local gray-scale statistical features. This ensures accurate capture of subtle trabecular bone structures. Furthermore, isolated noise is eliminated through three-dimensional connected component analysis, and topological repair is performed based on the morphological characteristics of bone tissue. This effectively corrects trabecular network breaks or pseudo-connections caused by imaging and segmentation errors, significantly improving the accuracy and reliability of topological parameters such as connectivity density. Finally, by removing cortical bone and selecting cancellous bone from standardized regions for analysis, the extracted bone microstructure parameters are ensured to be based on clinically meaningful and pure microstructures, exhibiting good repeatability and comparability. This achieves accurate extraction of bone microstructure parameters under clinically acceptable conditions, strongly supporting the early diagnosis and risk prediction of diseases such as osteoporosis.

[0012] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0013] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart of an embodiment of a method for extracting bone microstructure parameters provided in this application; Figure 2 A block diagram of a device for extracting bone microstructure parameters provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of one embodiment of a computing device provided in this application. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] It should be noted that, in the cases involving user information in the embodiments of this application, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse. In addition, the various models involved in this application (including but not limited to language models or large models) comply with relevant laws and standards.

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] Skeletal system diseases (such as osteoarthritis and osteoporosis) are a major global public health problem. Most serious consequences, such as fragility fractures, are rooted in the degeneration of bone tissue microstructure, such as the reduction or disruption of the number (Tb.N), thickness (Tb.Th), spacing (Tb.Sp), and connectivity (Conn.D) of trabeculae in cancellous bone.

[0018] However, existing clinical gold standards, such as bone mineral density (BMD) measurements (e.g., dual-energy X-ray absorptiometry, DEXA) and commonly used risk assessment models, mainly reflect the mineral content of bone tissue and are difficult to fully extract bone microstructure parameters to reflect the complex changes in the bone microstructure network. This results in a significant blind spot in fracture prediction in people with normal BMD or reduced bone mass.

[0019] Among related technologies, high-resolution peripheral quantitative CT (HR-pQCT) technology has been developed to achieve in vivo bone microstructure imaging, which can accurately measure the morphology and connectivity parameters of bone trabeculae. However, due to the high cost of HR-pQCT equipment, relatively high radiation dose, limitation to scanning of the distal extremities, and long examination time, it is currently mainly used for scientific research and is difficult to popularize in routine clinical screening.

[0020] Currently, the only equipment capable of visualizing and quantitatively assessing bone microstructure in living organisms in routine clinical practice is ultra-high resolution CT (U-HRCT). U-HRCT can achieve a stereoscopic resolution of approximately 50 μm in specific scanning modes, enabling precise visualization of the trabecular bone network structure and measurement of morphological parameters in vivo. However, during the development of this invention, the inventors discovered that the ultra-high resolution CT images acquired by U-HRCT are of poor quality, exhibiting issues such as high image noise, low sharpness of trabecular bone edges, and low contrast between trabecular gaps, thus making it impossible to accurately extract bone microstructure parameters.

[0021] Therefore, how to accurately extract bone microstructure parameters has become an urgent technical problem to be solved.

[0022] To address the aforementioned technical issues, this application provides a method for extracting bone microstructure parameters. It employs a super-resolution reconstruction step to enhance the original CT image with high fidelity, effectively suppressing noise, blurring, and artifacts introduced by low-dose scanning. This significantly improves the sharpness and detail of trabecular and cortical bone boundaries. Secondly, a gray-scale-spatial consistency fusion strategy is used in the segmentation stage. Efficient initial separation is achieved through a global threshold, followed by the construction of a local adaptive threshold using local gray-scale statistical features to accurately reclassify edges and low-contrast regions, ensuring accurate capture of fine trabecular structures. Furthermore, isolated noise is eliminated through three-dimensional connected component analysis, and topological repair is performed based on the morphological characteristics of bone tissue. This effectively corrects trabecular network breaks or pseudo-connections caused by imaging and segmentation errors, significantly improving the accuracy and reliability of topological parameters such as connectivity density. Finally, by removing cortical bone and selecting cancellous bone from standardized regions as the analysis object, the extracted bone microstructure parameters were ensured to be based on clinically significant and pure microstructures, with good reproducibility and comparability. This enabled the accurate extraction of bone microstructure parameters under clinically acceptable conditions, providing strong support for the early diagnosis and risk prediction of diseases such as osteoporosis.

[0023] Figure 1 A flowchart of an embodiment of the method for extracting bone microstructure parameters provided in this application is shown below. Figure 1 As shown, the method may include the following steps: 101: Acquire ultra-high resolution CT images of the bone tissue to be examined.

[0024] Ultra-high resolution CT images are obtained by scanning specific anatomical sites of a subject, such as the distal radius, proximal tibia, or vertebral body, using clinical ultra-high resolution CT (U-HRCT) equipment. Considering clinical ethics and safety standards, this scanning process typically employs a low-dose radiation protocol. This may result in the acquired original ultra-high resolution CT images, while offering improved spatial resolution, inevitably carrying significant noise introduced by the low dose, blurring artifacts, and systematic artifacts due to effects such as beam hardening.

[0025] 102: Perform super-resolution reconstruction on ultra-high resolution CT images to obtain reconstructed ultra-high resolution CT images.

[0026] In the embodiments of this application, by performing super-resolution reconstruction on ultra-high resolution CT images, image quality can be improved to suppress high-frequency noise and artifacts in the original ultra-high resolution CT images. At the same time, the sharpness and texture details of the trabecular bone edges can be restored and enhanced, so that the obtained reconstructed ultra-high resolution CT images have higher clarity, lower noise levels and richer microstructural details than the original ultra-high resolution CT images, providing a high-quality data foundation for subsequent accurate segmentation and quantitative analysis.

[0027] The process of super-resolution reconstruction of ultra-high resolution CT images will be described in the following embodiments.

[0028] 103: Calculate the global grayscale distribution features of the reconstructed ultra-high resolution CT image to determine the global segmentation threshold, and generate the first bone mask on the reconstructed ultra-high resolution CT image based on the global segmentation threshold to obtain the bone mask ultra-high resolution CT image.

[0029] After obtaining the reconstructed ultra-high resolution CT image through super-resolution reconstruction, the global grayscale distribution characteristics of the reconstructed ultra-high resolution CT image can be calculated to determine the global segmentation threshold.

[0030] In embodiments of this application, the reconstructed ultra-high resolution CT image can be a three-dimensional grayscale voxel image, where bone tissue is typically represented as a set of high-grayscale voxels, while the background (including soft tissue, bone marrow, or air, etc.) is typically represented as a set of low-grayscale voxels. By calculating the global grayscale distribution features, statistical analysis can be performed on the voxel grayscale values ​​in the reconstructed ultra-high resolution CT image to determine the grayscale values ​​that can optimally divide the grayscale distribution of the entire image into two categories: foreground (bone tissue) and background (non-bone tissue). This grayscale value can then be determined as the global segmentation threshold. This global segmentation threshold allows for efficient preliminary differentiation between the main bone structure and the background.

[0031] After obtaining the global segmentation threshold, the reconstructed ultra-high resolution CT image can be converted into a preliminary bone tissue mask, i.e., the first bone mask, based on the global segmentation threshold.

[0032] In one possible implementation, the ultra-high resolution CT image can be reconstructed voxel by voxel, comparing its grayscale value with a global threshold: when the grayscale value of a voxel is higher than the threshold, it is marked as a candidate bone voxel; otherwise, it is marked as a background voxel. Through this voxel-by-voxel binarization operation, a complete three-dimensional binary mask can be generated. In this mask, the set of voxels marked as "bone" constitutes the initial bone region, and the voxel region marked as "background" constitutes the non-bone tissue or background part.

[0033] Because bone tissue typically exhibits spatial inhomogeneity in images—for example, frequent grayscale changes in internal cancellous bone regions and significant grayscale differences in edge regions—while some thin-layer bone tissues or low-contrast areas may have similar grayscale values ​​to the surrounding background, relying solely on a uniform global threshold often fails to accurately identify these complex regions. Therefore, further, the segmentation threshold can be dynamically adjusted based on the grayscale characteristics of each voxel's local region, allowing the segmentation strategy to adapt to image conditions at different locations.

[0034] 104: Calculate local gray-scale statistical features of voxels in ultra-high resolution CT images of bone masks, and construct local adaptive thresholds based on local gray-scale statistical features.

[0035] 105: Reclassify the edge regions or low-contrast regions in the first bone mask using local adaptive thresholding to obtain the second bone mask.

[0036] After obtaining the ultra-high resolution CT image of the bone mask corresponding to the first bone mask, the gray-level changes of each voxel in the image within its local area can be further analyzed to obtain local gray-level statistical information that reflects the characteristics of the voxel's region.

[0037] In one embodiment, for any voxel, a local reference region in space can be determined based on the voxel. By performing statistical processing on the grayscale data within the local reference region, local grayscale statistical features describing the brightness level, grayscale change trend, or local texture features around the voxel can be obtained.

[0038] After obtaining the local gray-level statistical features, a locally adaptive threshold matching the local gray-level characteristics of each voxel can be determined based on these features. This locally adaptive threshold is no longer a globally uniform value, but rather adapts to the local environment of the voxel. This allows regions with significant differences in brightness and darkness in the image to use more suitable segmentation conditions, while regions with similar gray levels, low contrast, or thin structures can be more accurately identified based on their local gray-level characteristics.

[0039] By using local adaptive thresholds to distinguish between bone and non-bone tissues, the misclassification phenomenon that is prone to occur when only a global threshold is used can be effectively improved, making the bone tissue boundary more closely match the real structure, while enhancing the recognition ability of thin beams, small-scale structures and low-contrast regions.

[0040] In this context, the boundary region typically refers to the voxel band adjacent to the mask and non-mask, i.e., the boundary between bone and non-bone areas after initial binarization; low-contrast regions refer to areas with small gray-level differences and insignificant local statistics in the original reconstructed image. These regions often have high uncertainty and are prone to misclassification. After identifying these regions, a local adaptive threshold can be applied to them, while voxels with high confidence within the mask are left unlabeled to avoid unnecessary perturbation to stable regions.

[0041] 106: Based on the preset volume constraints, isolated noise connected domains are removed from the second bone mask, and the fractures or pseudo-connections of the trabecular network are repaired based on the morphological characteristics of the bone tissue to obtain the bone tissue structure segmentation results.

[0042] Within the defined second bone mask, there may be isolated noise regions unrelated to actual bone tissue. These regions are typically caused by imaging noise, local segmentation errors, or extremely small non-bone high-grayscale points, and their volume is usually much smaller than that of the actual bone trabecular structure. To ensure the structural consistency and reliability of the segmentation results, the size of each independent region in the mask can be determined based on preset volume constraints, and isolated voxel sets that clearly do not conform to the size characteristics of actual bone tissue can be excluded from the mask. This effectively removes scattered noise, preventing it from interfering with subsequent morphological analysis or extraction of bone microstructure parameters.

[0043] Furthermore, due to resolution limitations or grayscale fluctuations during imaging, some trabeculae may exhibit minor breaks during initial segmentation, causing what was originally a continuous structure to be divided into multiple discontinuous segments. Simultaneously, some regions may form pseudo-connections that do not conform to the true physiological structure due to grayscale blurring or local misclassification. To ensure that the final bone tissue segmentation results objectively reflect the actual morphology of the trabecular network, these broken or pseudo-connected areas can be repaired by incorporating the inherent morphological characteristics of bone tissue.

[0044] In one implementation of this application, based on the continuity, orientation consistency and spatial extension trend of bone tissue, the connection and completion of positions that clearly belong to the same trabeculae but are slightly interrupted can be performed, while unreasonable pseudo-connections can be corrected to return to a form that is more in line with the real structure.

[0045] 107: Remove cortical bone from the bone tissue segmentation results to obtain the cancellous bone segmentation results.

[0046] Because cortical bone and cancellous bone differ significantly in anatomical structure, geometry, and grayscale characteristics—cortical bone typically appears as a dense, high-grayscale, thick-walled structure located on the outer layer of bone tissue, while cancellous bone is located internally and consists of a fine trabecular network—mixing the two types of bone tissue not only affects the accuracy of subsequent bone microstructural parameters but also interferes with the quantitative description of the trabecular network. Therefore, after bone tissue segmentation, cortical bone and cancellous bone can be distinguished, and the cortical bone can be removed.

[0047] In one implementation of this application, cortical bone can be removed based on the differences in spatial location and structural morphology between cortical bone and cancellous bone. First, since cortical bone generally continuously surrounds the bone surface, portions near the outer boundary that are distributed in a ring or shell shape can be identified as cortical bone regions in the overall bone tissue segmentation results. Second, the thickness of cortical bone is usually significantly greater than the size of a single trabecular bone, presenting as a wide and dense voxel layer in the segmentation results; therefore, its overall outline can be identified and removed based on this morphological characteristic. The removal process involves locating the outer voxels of the segmentation results, analyzing the structural thickness and continuity layer by layer inward, thereby separating the outer continuous region whose thickness conforms to the anatomical characteristics of cortical bone, and removing it as part of the cortical bone.

[0048] 108: Bone microstructure parameters were extracted from the cancellous bone segmentation results.

[0049] After obtaining the trabecular network structure containing only cancellous bone, various bone microstructure parameters reflecting its microstructural characteristics can be calculated based on the cancellous bone segmentation results.

[0050] The segmentation result of cancellous bone can be a three-dimensional binary image composed of voxels. By performing geometric measurements, topological analysis, and statistical calculations on this three-dimensional structure, various microstructural indices can be obtained. For example, based on the proportion of cancellous bone in the overall volume, the volume fraction reflecting bone mass can be obtained; based on the spatial distribution and geometric morphology of the trabeculae within the cancellous bone structure, thickness parameters reflecting structural thickness can be extracted; the porosity of the cancellous bone spatial structure can be calculated based on the spacing distribution between trabeculae; the completeness of its structural network can be reflected based on the overall number density or arrangement of bone trabeculae; and parameters measuring network connectivity can also be extracted based on the spatial connections between trabeculae.

[0051] Different types of bone microstructure parameters provide quantitative support for different aspects of bone quality, and all of them can be calculated based on the geometric structure and voxel statistics of segmented images.

[0052] These bone microstructure parameters can quantify the internal structure of cancellous bone, which is originally difficult to describe accurately from a visual perspective, into a set of indicators that are easier to compare, evaluate, and model, thereby providing a reliable data foundation for bone quality assessment, disease diagnosis, efficacy tracking, or related research.

[0053] This application employs a super-resolution reconstruction step to enhance the original ultra-high resolution CT images with high fidelity, effectively suppressing noise, blurring, and artifacts introduced by low-dose scanning. This significantly improves the sharpness and detail of trabecular and cortical bone boundaries. Secondly, a gray-scale-spatial consistency fusion strategy is used in the segmentation stage. Efficient initial separation is achieved through a global threshold, followed by accurate reclassification of edges and low-contrast regions using local gray-scale statistical features. This ensures accurate capture of subtle trabecular bone structures. Furthermore, isolated noise is eliminated through three-dimensional connected component analysis, and topological repair is performed based on the morphological characteristics of bone tissue. This effectively corrects trabecular network breaks or pseudo-connections caused by imaging and segmentation errors, significantly improving the accuracy and reliability of topological parameters such as connectivity density. Finally, by removing cortical bone and selecting cancellous bone from standardized regions for analysis, the extracted bone microstructure parameters are ensured to be based on clinically meaningful and pure microstructures, exhibiting good repeatability and comparability. This achieves accurate extraction of bone microstructure parameters under clinically acceptable conditions, strongly supporting the early diagnosis and risk prediction of diseases such as osteoporosis.

[0054] In one possible implementation, super-resolution reconstruction of ultra-high-resolution CT images can be achieved based on sparse representation and iterative optimization. Specifically, this can be implemented as follows: First, the ultra-high-resolution CT image is divided into blocks to obtain multiple overlapping image patches. Then, a Discrete Wavelet Transform (DWT) or similar sparsification transformation algorithm is applied to each image patch to transform it from the pixel domain to the transform domain, obtaining the sparse coefficients of the image patch, which represent the sparse representation of the image patch on the transform basis. Next, sparse coding can be performed using an overcomplete dictionary pre-trained on a large database of high-resolution ultra-high-resolution CT images using dictionary learning methods such as K-SVD, aiming to find an encoding that can represent the sparse coefficients with the fewest atomic linear combinations. Based on the sparse coding, an optimization objective function with total variation regularization can be constructed, formulating the super-resolution reconstruction problem as a constrained minimization problem. The core of this optimization function lies in balancing the data fidelity of the reconstruction with the quality constraints of the image. The objective function can include a fidelity term to ensure the reconstruction result closely approximates the original low-resolution observation, and a regularization term using total variational regularization (TVRegularization) to suppress noise and maintain the smoothness of trabecular bone edges. This objective function is solved using iterative optimization algorithms such as the Alternating Directional Multiplier Method (ADMM). In each iteration, high-resolution image patches and their sparse codes are alternately updated, gradually minimizing reconstruction errors and improving details. Finally, after the iteration converges, the optimized high-resolution image patches are seamlessly stitched together and subjected to post-processing steps such as anisotropic diffusion filtering to further smooth flat areas while sharpening edges, ultimately resulting in a reconstructed CT image with clear edges and significantly reduced artifacts. In another possible implementation, super-resolution reconstruction of ultra-high-resolution CT images can be achieved based on a Dense Convolutional Network (DCN). This DCN architecture maximizes the information flow within the network. Its core feature is that each layer receives feature maps from all preceding layers as input and passes its own output feature map as input to all subsequent layers. This feature reuse and cascading mechanism greatly enhances feature propagation efficiency and the network's ability to extract detailed features, enabling the network to efficiently capture and aggregate multi-scale degradation information from the original ultra-high-resolution CT image. During super-resolution processing, DCN can be combined with sub-pixel convolutional layers or efficient upsampling modules to map low-resolution feature maps to a high-resolution output space without introducing additional computational complexity, thereby generating high-precision reconstructed ultra-high-resolution CT images. The loss function of this method can also be pixel-level. A combination of loss and multi-scale structural similarity (MS-SSIM) loss is used to ensure that the topological integrity and structural consistency of bone microstructures are maintained while restoring high resolution. In this way, effective noise reduction and detail restoration are achieved on low-dose ultra-high resolution CT images, resulting in accurate input images for subsequent segmentation.

[0055] In some embodiments, calculating the global grayscale distribution features of the reconstructed ultra-high resolution CT image to determine the global segmentation threshold can be specifically implemented as follows: Statistical analysis of the gray-level histogram distribution of reconstructed ultra-high resolution CT images; From the gray-level histogram distribution, the gray-level value that maximizes the inter-class variance and / or minimizes the intra-class variance between the bone tissue pixel set and the background pixel set is determined as the global segmentation threshold.

[0056] In the embodiments of this application, the grayscale values ​​of all voxels in the reconstructed ultra-high resolution CT image can first be read, and a grayscale histogram can be constructed based on these grayscale values. The grayscale histogram can divide the grayscale range of the entire image into several continuous grayscale intervals and count the number of corresponding voxels in each interval, thereby forming a frequency distribution map that can reflect the grayscale distribution characteristics of the entire image. Since the density differences of different tissues in the reconstructed ultra-high resolution CT image are significant, bone tissue generally corresponds to higher grayscale values, while soft tissue or air background corresponds to lower grayscale values. Therefore, there will be two or more relatively independent grayscale distribution peaks in the histogram.

[0057] After obtaining the grayscale histogram, to determine a global segmentation threshold that can effectively distinguish between bone and non-bone tissues, threshold optimization can be performed based on the criteria of maximizing inter-class variance or minimizing intra-class variance. Specifically, when any possible grayscale value on the histogram is considered as a candidate threshold, the histogram can be divided into two sets by the threshold, representing the voxels considered as bone and the voxels considered as background, respectively. For each candidate threshold, the mean, variance, and corresponding weights of these two sets can be calculated, and then the inter-class variance between sets and the intra-class variance within sets can be obtained. When a certain grayscale value maximizes the inter-class variance, it indicates that the overall grayscale difference between bone tissue and background tissue is most significant at that threshold; or, when a certain grayscale value minimizes the intra-class variance, it indicates that the grayscale distribution within each set is most concentrated and the internal consistency is highest at that threshold, thus achieving good differentiation. Therefore, the grayscale value that satisfies the above conditions can be determined as the global segmentation threshold.

[0058] In some embodiments, generating a first bone mask on the reconstructed ultra-high resolution CT image based on a global segmentation threshold, and obtaining an ultra-high resolution CT image of the bone mask, can be specifically implemented as follows: The gray value of each voxel in the reconstructed ultra-high resolution CT image is compared with the global segmentation threshold. Voxels with gray values ​​greater than the global segmentation threshold are marked as candidate bone voxels and assigned the first identifier value; voxels with gray values ​​less than or equal to the global segmentation threshold are marked as background voxels and assigned the second identifier value, thus forming a binarized first bone mask.

[0059] In embodiments of this application, voxels in the reconstructed ultra-high resolution CT image can be read one by one and their corresponding gray values ​​obtained, which are then compared with a predetermined global segmentation threshold. When the gray value of a voxel is higher than the threshold, it is considered that the voxel is more likely to belong to the bone tissue region, and therefore it is marked as a candidate bone voxel and assigned a first identifier value to represent the bone tissue category; when the gray value of a voxel is less than or equal to the threshold, it is determined as a background voxel and assigned a second identifier value to represent the background category.

[0060] Through the aforementioned voxel-by-voxel comparison and labeling operations, all voxels in the reconstructed ultra-high resolution CT image can be classified into two categories: "bone" or "background." After all voxel-level classifications are completed, a binary image containing clear bone / non-bone labels is obtained, which constitutes the first bone mask. This mask maintains a one-to-one spatial correspondence with the original grayscale image, and therefore can be further combined with the reconstructed ultra-high resolution CT image to form a bone mask ultra-high resolution CT image. This provides a clear structural framework and preliminary spatial constraints for subsequent local feature analysis, boundary refinement, and bone tissue structure repair.

[0061] In some embodiments, calculating the local gray-level statistical features of voxels in ultra-high resolution CT images of bone masks can be specifically implemented as follows: For any bone cell, a three-dimensional neighborhood window of a preset size is created with the bone cell as the center. The gray values ​​of all voxels within the three-dimensional neighborhood window are statistically analyzed, and the local gray mean and local gray standard deviation are calculated. The mean and standard deviation of local gray levels are used as statistical features of local gray levels.

[0062] After determining the first bone mask, for any voxel labeled as a bone voxel, a three-dimensional neighborhood window of a preset size can be constructed centered on the voxel's position in three-dimensional space. This three-dimensional neighborhood window can be a cube or a regular region approximating a cube, and its size can be set according to the voxel spacing of the reconstructed ultra-high resolution CT image and the local structural scale of the bone tissue, so that the window can cover a sufficient number of neighboring voxels, thereby ensuring the stability and representativeness of the statistics. After the window is established, the gray values ​​of all voxels within it are traversed and statistically analyzed to obtain the gray value distribution of all voxels within the window.

[0063] During the statistical process, the sum of gray levels of a specified voxel in the local space can be obtained through cumulative calculation, and the local gray level mean can be further calculated. This mean is used to characterize the trend of several voxels around the voxel in the overall brightness level. At the same time, the deviation of the voxel gray level values ​​in this region from the mean can also be calculated, i.e., the local gray level standard deviation, thus reflecting the gray level fluctuation and local contrast level in the neighborhood. Since the gray level mean and gray level standard deviation describe the local gray level characteristics around the voxel from the two dimensions of brightness and fluctuation, respectively, the local gray level mean and local gray level standard deviation can be used as the local gray level statistical features of the voxel, and the local adaptive threshold can be determined based on the local gray level statistical features.

[0064] In some embodiments, constructing a local adaptive threshold based on local gray-level statistical features can be specifically implemented as follows: The grayscale fluctuation correction coefficient is calculated based on the ratio of the local grayscale standard deviation to the preset grayscale dynamic range normalization parameter. By using a grayscale fluctuation correction coefficient to weight and adjust the local grayscale mean, a local adaptive threshold is obtained.

[0065] After obtaining the local grayscale mean and local grayscale standard deviation, a correction coefficient characterizing the degree of local grayscale fluctuation can be obtained by first calculating the ratio of the local standard deviation to a preset grayscale dynamic range normalization parameter. This grayscale fluctuation correction coefficient numerically reflects the contrast characteristics of the local structure: when the grayscale fluctuation within the neighborhood is large, the correction coefficient increases accordingly; when the grayscale change in the neighborhood is small and the contrast is low, the correction coefficient decreases accordingly. Furthermore, the local grayscale mean can be weighted and adjusted based on the grayscale fluctuation correction coefficient. By applying an offset or scaling related to grayscale fluctuation to the local grayscale mean, the final determined local adaptive threshold can reflect both the overall brightness level of the area and adaptively increase or decrease according to changes in local contrast, thereby generating a local adaptive threshold suitable for that specific location.

[0066] By determining a local adaptive threshold, bone tissue segmentation can be flexibly performed in different grayscale environments. This avoids missing low-contrast trabeculae due to an excessively high fixed threshold, and also prevents misclassification of background noise as bone tissue due to an excessively low threshold. This provides a more targeted threshold basis for subsequent local reclassification, thereby effectively improving the accuracy of bone tissue segmentation in edge, thin layer, and weak contrast regions.

[0067] In some embodiments, based on preset volume constraints, removing isolated noise connected components from the second bone mask can be specifically implemented as follows: Perform a three-dimensional connected component analysis on the second bone mask to generate multiple bone voxel sets; Calculate the geometric volume of multiple bone matrix sets; Remove bone matrix sets with geometric volumes smaller than a preset volume threshold from the second bone mask.

[0068] After obtaining the second bone mask through local adaptive threshold reclassification, although the mask has accurately captured the fine edges of the trabeculae, it may inevitably contain some isolated voxel clusters caused by imaging noise or minor artifacts. These isolated voxel clusters, though extremely small, can contaminate subsequent bone microstructure parameters, especially interfering with the accuracy of indicators such as connectivity density and bone volume fraction, if not removed. Therefore, embodiments of this application can also remove isolated noisy connected domains from the second bone mask based on preset volume constraints.

[0069] In the embodiments of this application, a three-dimensional connected component analysis can first be performed on the second bone mask. Three-dimensional connected component analysis is a topology-based image processing technique designed to identify and aggregate all interconnected voxels (i.e., voxels with the identifier "bone") in a binary mask. This analysis can classify all spatially directly or indirectly adjacent bone voxels into an independent set by examining the connectivity between each bone voxel and its neighboring voxels in three-dimensional space, thereby generating multiple sets of bone voxels. Each set of bone voxels represents an independent connected structural unit in the bone mask.

[0070] Subsequently, the geometric volume of multiple bone voxel sets can be calculated. This geometric volume can be calculated, for example, by counting the number of voxels contained within each bone voxel set and multiplying this number by the actual physical volume of a single voxel (determined by CT scan parameters). This volume value represents the actual size of each connected structural unit.

[0071] Finally, voxel clusters can be screened using volume constraints. Screening voxel clusters can be implemented, for example, by removing bone voxel sets with geometric volumes smaller than a preset volume threshold from the second bone mask. The preset volume threshold can be the minimum effective structural volume determined based on clinical experience or experimental analysis, for example, set to a volume sufficient to accommodate a certain number of voxels, ensuring that any set smaller than this threshold is considered an isolated, clinically meaningless noisy connected region. By labeling voxels in these excessively small connected regions as background voxels (i.e., assigning them a second identifier value), denoising and refining of the segmentation results can be effectively achieved, ensuring that subsequent analysis is based only on structurally significant main bone tissue.

[0072] After obtaining the second bone mask with volume-constrained denoising, although most of the noise has been removed, minor structural breaks or unnecessary pseudo-connections may still exist in the trabecular bone network due to imaging blur, insufficient local contrast, and the accumulation of small errors in the segmentation threshold. These defects can severely affect key parameters (such as connectivity density) calculated based on topology. Therefore, embodiments of this application may also include a step of repairing the breaks or pseudo-connections in the trabecular bone network based on the morphological characteristics of bone tissue, aiming to restore the true connectivity of the bone network. In some embodiments, repairing the breaks or pseudo-connections in the trabecular bone network based on the morphological characteristics of bone tissue can be specifically implemented as follows: Extract the three-dimensional skeleton or central path of the second bone mask to determine the path continuity of the trabecular network structure; Identify endpoint pairs with a preset distance gap based on the connection status of the 3D skeleton or central path; The endpoint pairs are identified as breakpoints; Perform morphological closure transformations or distance-based fill operations on the breakpoints to establish connections between them.

[0073] In one implementation of this application, the three-dimensional skeleton or center path of the second bone mask can first be extracted to determine the path continuity of the trabecular bone network structure. This process can be implemented using distance transformation and skeletonization algorithms to refine the three-dimensional bone tissue structure into a centerline network of monomer widths. The skeleton accurately represents the geometric core path of the trabecular bone, facilitating subsequent localization of structural defects.

[0074] Subsequently, defect analysis can be performed based on the connection status of the three-dimensional skeleton or central path. By analyzing the endpoints, branch points, and gaps on the skeleton, endpoint pairs with a preset distance gap can be identified. An endpoint pair is the two ends of a trabecular rod. If the spatial distance between them is short (e.g., a gap less than or equal to 1 to 3 voxels), the gap can be identified as a potential split fracture, and the endpoint pair can be identified as the fracture point.

[0075] Finally, to repair the fractures, morphological closure transformations or distance-based filling operations can be performed on the fracture points to establish connections between them. Morphological closure transformations are morphological operations that involve dilation followed by erosion, with the core objective of "stitching" together tiny gaps without significantly altering the overall thickness of the trabecular bone. Distance-based filling operations utilize skeletal information to guide the filling area, ensuring the connection path is optimal. Through this repair process, the connectivity of the trabecular bone network structure is optimized, thus avoiding topological distortion caused by segmentation errors. This results in high-fidelity bone tissue segmentation, providing a foundation for accurate calculation of parameters such as connectivity density.

[0076] In the embodiments of this application, after obtaining the bone tissue structure segmentation result after topological repair, the result includes the outer cortical bone and the inner cancellous bone. In order to analyze the trabecular microstructure, which is most sensitive to osteoporosis, the embodiments of this application further include the step of removing cortical bone from the bone tissue segmentation result to obtain the cancellous bone segmentation result.

[0077] In one possible implementation, the removal operation can be achieved through a distance transformation distribution: first, a three-dimensional geometric distance transformation is performed on the bone tissue segmentation result, and the minimum distance from each voxel to its nearest external boundary is calculated. In this embodiment, the cortical portion is automatically removed based on a preset thickness (e.g., a cortical bone retention distance threshold set to 0.5 mm to 1 mm). Specifically, all voxels whose distance from the outer boundary of the bone tissue is less than or equal to this preset thickness can be identified as cortical bone voxels and removed from the segmentation result, ultimately retaining only the internal cancellous bone volume, resulting in a pure cancellous bone segmentation result.

[0078] Subsequently, to ensure the comparability of analysis results between different patients and different scans, this embodiment of the application also selects a region of interest (ROI) for the cancellous bone segmentation results. The ROI can be selected based on a standardized template of the anatomical location. For example, if the scan is of the proximal tibia or distal radius, the articular surface is used as a stable anatomical reference point. The ROI can be determined based on spatial coordinate constraints. For example, in the direction perpendicular to the principal axis, a standard cubic region located within 5 mm to 10 mm from the articular surface is selected, and its volume is typically fixed at 5. 5 This standardized selection process ensures consistency across different samples and eliminates parameter fluctuations caused by subjectivity in ROI selection. Finally, to guarantee the geometric integrity and repeatability of the input region, morphological filtering, such as smoothing, is applied to the ROI boundaries to avoid boundary artifacts caused by rigid trabecular bone truncation, resulting in the region ultimately used for parameter extraction having optimal geometric quality.

[0079] In the embodiments of this application, bone microstructure parameters may include any one or more of the following: bone volume fraction (BV / TV), trabecular thickness (Tb.Th), number of trabecular bones (Tb.N), trabecular spacing (Tb.Sp), and connectivity density (Conn.D), etc. These parameters together reflect the density, arrangement, connectivity, and structural integrity of the trabecular network.

[0080] The extraction process of the aforementioned bone microstructure parameters can be implemented as follows: Bone volume fraction (BV / TV): The proportion of voxel volume occupied by bone tissue in the total volume of the segmented cancellous bone. It can be calculated using the formula: ; in It can represent the volume of bone tissue, and this parameter can be calculated by multiplying the number of bone voxels in the cancellous bone segmentation result by the physical volume of a single voxel. It can represent the total volume of the segmented cancellous bone.

[0081] Trabecular thickness (Tb.Th): This index is used to measure the average thickness of trabecular plates or rods. It is usually obtained through methods such as three-dimensional distance transformation combined with midline skeleton analysis to reflect the density of bone.

[0082] Trabecular bone number (Tb.N) and trabecular bone spacing (Tb.Sp): Trabecular bone number (Tb.N) is defined as the average distribution density of trabecular bone per unit length, which is indirectly obtained through the calculated BV / TV and Tb.Th. The calculation formula can be: ; The trabecular spacing (Tb.Sp) is approximated using the following relationship: This is obtained to reflect the degree of sparseness between bone trabeculae.

[0083] Connectivity Density (Conn.D): This metric reflects the spatial connectivity and topological complexity of trabecular networks. It is calculated based on topological principles using the following formula: ; in Indicates the number of connected components. Indicates the number of holes. This represents the number of completely enclosed volumes. Because the structural repair in this embodiment was performed before segmentation, it ensures... The accuracy of topological parameters, therefore this The values ​​can reliably assess the integrity of bone structure and its resistance to mechanical damage. Finally, all extracted quantitative parameters of bone microstructure, including bone volume fraction (BV / TV), trabecular thickness (Tb.Th), trabecular number (Tb.N), trabecular spacing (Tb.Sp), and connectivity density (Conn.D), along with comparisons with standardized databases (such as risk level assessment) and annotations of local abnormal areas, can be integrated into a complete automated detection report, realizing a fully automated workflow from CT image acquisition to clinical parameter output.

[0084] Figure 2 A block diagram of a bone microstructure parameter extraction device provided in an embodiment of this application is shown below. Figure 2 As shown, the device may include: The first acquisition module 201 is used to acquire ultra-high resolution CT images of the bone tissue to be detected. Reconstruction module 202 is used to perform super-resolution reconstruction on ultra-high resolution CT images to obtain reconstructed ultra-high resolution CT images; The first mask extraction module 203 is used to calculate the global grayscale distribution features of the reconstructed ultra-high resolution CT image to determine the global segmentation threshold, and generate the first bone mask on the reconstructed ultra-high resolution CT image based on the global segmentation threshold to obtain the bone mask ultra-high resolution CT image. The second mask extraction module 204 is used to calculate local gray-level statistical features of voxels in the bone mask ultra-high resolution CT image and construct a local adaptive threshold based on the local gray-level statistical features. The third mask extraction module 205 is used to reclassify the edge region or low contrast region in the first bone mask using a local adaptive threshold to obtain the second bone mask. The first result generation module 206 is used to remove isolated noise connected domains from the second bone mask based on preset volume constraints, and repair the fractures or pseudo-connections of the bone trabecular network based on the morphological characteristics of bone tissue, so as to obtain the bone tissue structure segmentation result. The removal module 207 is used to remove cortical bone from the bone tissue segmentation results to obtain cancellous bone segmentation results; The parameter extraction module 208 is used to extract bone microstructure parameters from the cancellous bone segmentation results.

[0085] Figure 2 The bone microstructure parameter extraction device can perform... Figure 1 The implementation principle and technical effects of the bone microstructure parameter extraction method described in the illustrated embodiment will not be repeated here. The specific operation methods of each module and unit in the bone microstructure parameter extraction device in the above embodiments have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0086] It should be noted that some processes described in the above embodiments and accompanying drawings include multiple operations appearing in a specific order. However, it should be clearly understood that these operations may not be executed in the order they appear in this document, or they may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should also be noted that the descriptions such as "first" and "second" in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0087] Figure 3 This is a schematic diagram of the structure of one embodiment of a computing device provided in this application. Figure 3 As shown, in practice, the computing device may include a storage component 301 and a processing component 302.

[0088] Storage component 301 is used to store computer programs and can be configured to store various other data to support operation on a computing device. Examples of this data include instructions for any application or method used to operate on the computing device, data structures, contact data, phone book data, messages, pictures, videos, etc.

[0089] Processing component 302, coupled to storage component 301, is used to execute computer programs in storage component 301 for implementing, etc. Figure 1 The method for extracting bone microstructure parameters is shown.

[0090] Furthermore, such as Figure 3As shown, the computing device may also include other components such as a communication component 303, a display component 304, a power supply component 305, and an audio component 306. Figure 3 The diagram only shows some components and does not mean that the device includes only these components. Figure 3 The components shown. Additionally... Figure 3 The components within the dashed box are optional, not mandatory, and their specific requirements depend on the product form of the computing device. The computing device in this embodiment can be a terminal device such as a desktop computer, laptop computer, smartphone, or IoT (Internet of Things) device, or a server-side device such as a conventional server, cloud server, or server array. If the computing device in this embodiment is implemented as a terminal device such as a desktop computer, laptop computer, or smartphone, it may include... Figure 3 The components within the dashed box; if the computing device in this embodiment is implemented as a conventional server, cloud server, or server array, etc., it may be omitted. Figure 3 The component within the dashed box.

[0091] The processing component described above includes one or more processors to execute computer instructions to complete all or part of the steps in the method described above. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the method described above.

[0092] The aforementioned storage components can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0093] The aforementioned communication component is configured to facilitate wired or wireless communication between the device housing the communication component and other devices. The device housing the communication component can access wireless networks based on communication standards, such as mobile communication networks, or combinations thereof. In one exemplary embodiment, the communication component receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel.

[0094] The aforementioned display components may include a screen, which may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors can sense not only the boundaries of touch or swipe actions but also the duration and pressure associated with the touch or swipe operation.

[0095] The aforementioned power supply components provide power to various components within the device in which they reside. These power supply components may include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power to the device in which they reside.

[0096] The aforementioned audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC) configured to receive external audio signals when the device containing the audio component is in an operating mode, such as call mode, recording mode, or voice recognition mode. The received audio signals can be further stored in memory or transmitted via a communication component. In some embodiments, the audio component also includes a speaker for outputting audio signals.

[0097] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to implement the steps in the above-described method embodiments. The computer-readable storage medium includes volatile or non-volatile components, or a combination thereof, and can be removable or non-removable. Examples of computer-readable storage media include, but are not limited to, phase-change random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random-access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), flash memory or other memory technologies, CD-ROM, Digital Video Disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium. Accordingly, this application also provides a computer program product, which includes a computer program or instructions that, when executed by a processor, cause the processor to implement the steps in the above method embodiments. It should be understood that each step or combination of steps in the above method flow can be implemented by the computer program or instructions. Furthermore, these computer programs or instructions can be applied to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device, enabling the processor of the general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to function as an apparatus for implementing the corresponding functions in the above method embodiments.

[0098] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0099] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0100] Finally, it should be noted that the above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method of extracting bone microstructure parameters, characterized in that, The method comprises: obtaining an ultra-high resolution CT image of a bone tissue to be detected; performing super-resolution reconstruction on the ultra-high resolution CT image to obtain a reconstructed ultra-high resolution CT image; calculating a global gray distribution feature of the reconstructed ultra-high resolution CT image to determine a global segmentation threshold, and generating a first bone mask on the reconstructed ultra-high resolution CT image based on the global segmentation threshold to obtain a bone mask ultra-high resolution CT image; calculating a local gray statistical feature of a voxel in the bone mask ultra-high resolution CT image, and constructing a local adaptive threshold based on the local gray statistical feature; reclassifying an edge region or a low-contrast region in the first bone mask by using the local adaptive threshold to obtain a second bone mask; based on a preset volume constraint condition, removing isolated noise connected domains from the second bone mask, and repairing a fracture or a false connection of a bone trabecular network based on a morphological feature of the bone tissue to obtain a bone tissue structure segmentation result; removing cortical bone from the bone tissue segmentation result to obtain a cancellous bone segmentation result; extracting a bone microstructure parameter from the cancellous bone segmentation result.

2. The method of claim 1, wherein, The method further comprises: statistically analyzing a gray histogram distribution of the reconstructed ultra-high resolution CT image; determining a gray value that maximizes an inter-class variance between a bone tissue pixel set and a background pixel set and / or minimizes an intra-class variance from the gray histogram distribution as the global segmentation threshold.

3. The method of claim 2, wherein, The method further comprises: comparing a gray value of each voxel in the reconstructed ultra-high resolution CT image with the global segmentation threshold; marking a voxel with a gray value greater than the global segmentation threshold as a candidate bone voxel and assigning a first identification value to the voxel, and marking a voxel with a gray value less than or equal to the global segmentation threshold as a background voxel and assigning a second identification value to the voxel, thereby forming a binary first bone mask.

4. The method of claim 3, wherein, The method further comprises: for any bone voxel, establishing a three-dimensional neighborhood window of a preset size centered on the bone voxel; statistically analyzing gray values of all voxels in the three-dimensional neighborhood window to calculate a local gray mean value and a local gray standard deviation; taking the local gray mean value and the local gray standard deviation as the local gray statistical feature.

5. The method of claim 4, wherein, The method further comprises: calculating a gray fluctuation correction coefficient according to a ratio of the local gray standard deviation to a preset gray dynamic range normalization parameter; weighting and adjusting the local gray mean value by using the gray fluctuation correction coefficient to obtain the local adaptive threshold.

6. The method of claim 5, wherein, The method further comprises: performing three-dimensional connected domain analysis on the second bone mask to generate a plurality of bone voxel sets; calculating geometric volumes of the plurality of bone voxel sets; removing bone voxel sets with geometric volumes less than a preset volume threshold from the second bone mask.

7. The method of claim 6, wherein, the repairing of the broken or pseudo-connected trabecular network based on the morphological features of the bone tissue comprises: extracting a three-dimensional skeleton or center path of the second bone mask to determine path continuity of the trabecular network structure; identifying end point pairs with preset distance gaps according to a connection state of the three-dimensional skeleton or center path; determining the end point pairs as broken points; performing a morphological closing transformation or a distance transformation based filling operation on the broken points to establish connections between the broken points.

8. A computing device, comprising: comprising a processing component and a storage component; the storage component stores a computer program; the computer program is used to be called and executed by the processing component to implement the bone microstructure parameter extraction method according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, a computer program stored thereon, which, when executed by a processing component, implements the bone microstructure parameter extraction method according to any one of claims 1 to 7.

10. A computer program product, characterised in that, comprising a computer program or instructions, which, when executed by a processing component, implements the bone microstructure parameter extraction method according to any one of claims 1 to 7.

Images