Total knee joint CT image segmentation method, device, equipment and medium

Abstract

The invention relates to the technical field of medical images, and discloses a total knee joint CT image segmentation method and device, equipment and a medium, and the method comprises the steps: obtaining total knee joint CT image data, and carrying out the overall segmentation through a rapid segmentation model, and obtaining a femoral image and a tibia image preliminary result; extracting the maximum connected domain of the two through connected domain analysis to obtain a preliminary result of the stable continuous bone structure; positioning a complex key area, and cutting an original image to generate local data; carrying out fine segmentation through a fine segmentation model to obtain a key region fine result; and fusing the two types of results to generate a final segmentation result. A layering strategy of coarse segmentation and fine segmentation is adopted, the processing efficiency and precision are both considered, and the real-time analysis requirement of the total knee joint bone structure is met.

Description

Technical Field

[0001] This application relates to the field of medical imaging technology, and more specifically, to a method, apparatus, device, and medium for segmenting CT images of the entire knee joint. Background Technology

[0002] In the three-dimensional structural analysis of the entire knee joint, the regional segmentation of femoral and tibial images from knee CT images (such as DICOM format) is a crucial step, directly affecting the accuracy of bone structure model construction and related structural analysis. Existing techniques are mostly based on the nnU-Net framework for region segmentation, but they have significant shortcomings: these models employ a unified global segmentation strategy, which limits their attention to regions with complex and detailed anatomical structures, such as the mid-knee joint. When morphological abnormalities exist in the bone tissue, problems such as incomplete boundaries and discontinuous contours easily occur, resulting in poor region segmentation accuracy. Furthermore, processing full-length knee joint three-dimensional CT data requires calculating a large voxel range, leading to a large network size and computational load, resulting in long processing times and making it difficult to meet the needs of real-time analysis. Summary of the Invention

[0003] The main objective of this application is to provide a method, device, equipment, and medium for segmenting CT images of the entire knee joint, aiming to solve the technical problems of existing segmentation methods based on the nnU-Net framework, which have poor segmentation accuracy for complex areas of the knee joint and take a long time to process three-dimensional CT data, making it difficult to meet surgical requirements.

[0004] The first aspect of this application proposes a method for segmenting CT images of the entire knee joint, including: Acquire CT images of the entire knee joint, and use a fast segmentation model for overall segmentation to obtain preliminary segmentation results for the femoral and tibial images; Connectivity analysis is performed on the preliminary segmentation results to extract the largest connected components corresponding to the femoral and tibial images, thereby obtaining a stable and continuous preliminary segmentation result of the bone structure. Based on the preliminary segmentation results of the bone structure, key areas with complex anatomical structures are located, and the original CT image data is locally cropped to generate local image data. For the local image data, a fine segmentation model is used for segmentation processing to obtain fine segmentation results for each key local region; The preliminary segmentation results of the bone structure are fused with the fine segmentation results to generate the final segmentation result of the whole knee joint CT image.

[0005] Furthermore, the step of acquiring full knee CT image data and performing overall segmentation processing using a fast segmentation model to obtain preliminary segmentation results for the femoral and tibial images includes: Full knee joint CT image data were acquired, and a two-dimensional U-Net residual network model was used as the fast segmentation model. The image data of each slice is obtained by splitting the CT image along the Z-axis direction; The two-dimensional U-Net residual network model is used to perform segmentation operations on each slice of image data; The segmentation results of each slice are summarized to obtain preliminary segmentation results for the femur and tibia images.

[0006] Further, the step of performing connected component analysis on the preliminary segmentation results to extract the largest connected components corresponding to the femoral and tibial images, and obtaining stable and continuous preliminary segmentation results of bone structures, includes: The preliminary segmentation results are obtained, and connected component analysis is performed on them to identify all connected components and their corresponding categories. From the identified connected components, the sets of connected components belonging to the femoral and tibial images are selected respectively; For the connected component sets of femoral and tibial images, extract the corresponding maximum connected component for each. By integrating the largest connected regions of the femoral and tibial images, a stable and continuous preliminary segmentation result of the bone structure is obtained.

[0007] Furthermore, the step of locating key areas with complex anatomical structures based on the preliminary segmentation results of the bone structure, and locally cropping the original CT image data to generate local image data includes: The preliminary segmentation results of the bone structure and the original CT image data are obtained, and the key areas with complex anatomical structures in the total knee joint are determined based on the preliminary segmentation results of the bone structure. Identify the spatial location information of each key region in the original CT image data; The image range corresponding to each key area is delineated based on the spatial location information; The original CT image data is locally cropped according to the defined image range to generate local image data corresponding to each key region.

[0008] Furthermore, the step of acquiring the preliminary segmentation results of the bone structure and the original CT image data, and determining the key areas with complex anatomical structures in the total knee joint based on the preliminary segmentation results of the bone structure, includes: The preliminary segmentation results of the bone structure and the original CT image data are obtained, and the two are correlated and matched. Analyze the preliminary segmentation results of the correlated bone structure to identify the distribution information of various anatomical parts in the entire knee joint; Based on the distribution information of the anatomical sites, areas with complex morphological features and high segmentation difficulty are selected; The selected areas were defined as key regions with complex anatomy in the entire knee joint, and their positional relationship with the original CT image data was clarified.

[0009] Furthermore, the step of performing segmentation processing on the local image data using a fine segmentation model to obtain fine segmentation results for each key local region includes: Local image data corresponding to each key region are obtained, and the data are divided according to the key local regions. The fine segmentation model is determined to be a segmentation model based on nnU-Net. Match the corresponding nnU-Net-based fine segmentation model to each key local region; The local image data corresponding to each key local area are input into the matching fine segmentation model to perform targeted segmentation processing of bone tissue details; By summarizing the segmentation outputs of each fine segmentation model, the fine segmentation results of bone tissue corresponding to each key local region are obtained.

[0010] Furthermore, the step of fusing the preliminary segmentation result of the bone structure with the fine segmentation result to generate the final segmentation result of the whole knee joint CT image includes: The preliminary segmentation results of the bone structure and the fine segmentation results of each key local region are obtained, and the two are spatially aligned to maintain consistent spatial positions. Based on the aligned spatial relationships, the fine segmentation results of each key local region replace the content of the corresponding region in the preliminary segmentation results of the bone structure. Perform consistency verification on the replaced overall segmentation data, and integrate the preliminary segmentation content of the bone structure of non-critical areas with the fine segmentation content of critical areas; The segmented data is aggregated and integrated to generate the final segmentation result of bone tissue from the full knee CT image.

[0011] A second aspect of this application also proposes a whole-knee CT image segmentation device, comprising: The coarse segmentation module is used to acquire CT image data of the entire knee joint. A fast segmentation model is used for overall segmentation processing to obtain preliminary segmentation results of femoral and tibial images. The connected component analysis module is used to perform connected component analysis on the preliminary segmentation results, extract the largest connected component corresponding to the femur image and tibia image, and obtain a stable and continuous preliminary segmentation result of bone structure. The key region cropping module is used to locate key regions with complex anatomical structures based on the preliminary segmentation results of the bone structure, and to perform local cropping on the original CT image data to generate local image data. The fine segmentation module is used to perform segmentation processing on the local image data using a fine segmentation model to obtain fine segmentation results for each key local region; The result fusion module is used to fuse the preliminary segmentation results of the bone structure with the fine segmentation results to generate the final segmentation result of the whole knee joint CT image.

[0012] The first aspect of this plan brings the following benefits: This application employs a layered strategy of "coarse segmentation + fine segmentation" to rapidly segment the model and achieve efficient overall processing, significantly reducing the processing time of 3D CT data. At the same time, it improves the local segmentation accuracy by targeting and finely segmenting complex key areas. Furthermore, the fusion of these two approaches balances efficiency and accuracy, overcoming the problem of poor segmentation of complex areas in traditional methods and solving the pain point of excessive processing time. This approach can meet the dual requirements of segmentation speed and accuracy in the analysis of bone structure related to the entire knee joint. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating a method for segmenting CT images of the entire knee joint according to an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a total knee CT image segmentation device according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a computer device according to an embodiment of this application; The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0015] Those skilled in the art will understand that, unless explicitly stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in the specification of this application means the presence of features, integers, steps, operations, elements, modules, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, modules, components, and / or groups thereof. It should be understood that when an element is “connected” or “coupled” to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein may include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any modules and all combinations of one or more associated listed items.

[0016] Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0017] Reference Figure 1 This application provides a method for segmenting CT images of the entire knee joint, including: S1: Acquire full knee CT image data, use a fast segmentation model for overall segmentation processing, and obtain preliminary segmentation results for femoral and tibial images; S2: Perform connected component analysis on the preliminary segmentation results, extract the largest connected component corresponding to the femur image and tibia image, and obtain a stable and continuous preliminary segmentation result of the bone structure; S3: Based on the preliminary segmentation results of the bone structure, locate key areas with complex anatomical structures, perform local cropping on the original CT image data, and generate local image data; S4: For the local image data, a fine segmentation model is used for segmentation processing to obtain fine segmentation results for each key local region; S5: The preliminary segmentation results of the bone structure are fused with the fine segmentation results to generate the final segmentation result of the whole knee joint CT image.

[0018] In step S1, the entire knee CT image data is segmented using a two-dimensional U-Net residual network model, combined with the processing logic of CT (Computed Tomography) image Z-axis slice segmentation. This yields preliminary segmentation results for the femur and tibia images. Taking the clinically common DICOM format of the entire knee CT image as an example, which contains complete three-dimensional data of the femur, tibia, and surrounding soft tissues, when a patient with mild osteophyte CT images is input, the system first determines the fast segmentation model as a two-dimensional U-Net residual network (segmentation speed is improved by more than 30% compared to a three-dimensional model). Then, the three-dimensional data is split into 120 slice image data along the Z-axis of the CT image. Based on the model's ability to identify grayscale differences between bone and soft tissues, segmentation calculations are performed on each slice, focusing on distinguishing the femur, tibia, and surrounding muscles, synovial fluid, and other non-target tissues. For example, in slices containing the middle of the knee joint in layers 30-50, the model reduces the gradient vanishing problem through residual block structure, quickly locks the general outline of bone tissue, and finally summarizes the segmentation results of all slices to form preliminary segmentation results of the femur and tibia covering the entire length of the knee joint. This yields preliminary results of the image segmentation region corresponding to the target bone tissue (femur and tibia). Although this result can fully present the overall morphology of the bone structure, there is slight boundary blurring in the osteophyte region at the posterior end of the femur. This step uses a two-dimensional U-Net residual network as a fast segmentation model. The Z-axis slice segmentation strategy significantly reduces the complexity of three-dimensional data processing and significantly shortens the overall segmentation time, adapting to the real-time requirements of surgery. The residual block structure enhances the model's feature extraction capability, ensuring the overall accuracy of bone tissue localization and providing a reliable foundation for subsequent connected component analysis and key region localization. At the same time, it avoids interference from non-target tissues, balancing segmentation efficiency and preliminary localization accuracy, and solving the pain point of long processing time for large-scale data by traditional nnU-Net models.

[0019] In step S2, the preliminary segmentation results obtained in step S1 are denoised and purified using a connected component analysis algorithm, combined with the category distinction rules for femoral and tibial images, ultimately yielding a stable and continuous preliminary segmentation result of the bone structure. Taking the preliminary segmentation result of a patient's CT image with mild osteophyte formation as an example, in addition to the main outlines of the femoral and tibial images, the result also contains three small connected component noises caused by soft tissue grayscale interference (belonging to the "non-target tissue" category). The system first performs connected component analysis on the preliminary segmentation result, identifying all eight connected components through pixel connectivity determination, and labeling the category of each connected component (five belonging to the femoral, two to the tibia, and one to the non-target tissue). Subsequently, the largest connected component (accounting for 92% of the total segmented area of ​​the femoral image) is selected from the connected component set of the femoral image, and the corresponding largest connected component (accounting for 94% of the total segmented area of ​​the tibia image) is extracted from the connected component set of the tibia image, removing the three small non-target connected components and scattered pixel noise. Finally, by integrating the two types of maximum connected components, the preliminary segmentation results of the generated bone structure show that although the contour of the osteophyte region at the posterior end of the femur is still slightly rough, the overall structure is free of breaks and redundant noise, fully meeting the spatial constraints required for subsequent key region localization. This step, through connected component analysis, accurately removes noise and non-target regions, effectively solving the discontinuity problem in the preliminary segmentation results, ensuring the integrity and stability of the bone structure, providing reliable data support for subsequent key region localization and local trimming, and improving the coherence and accuracy of the overall segmentation process.

[0020] In step S3, through anatomical distribution analysis and spatial location positioning technology, combined with key area screening rules, and based on the preliminary segmentation results of stable bone structures obtained in step S2, complex key areas are located and cropped to generate local image data. Taking the preliminary segmentation results of bone structures in patients with mild osteophytes mentioned above as an example, the results have clearly defined the overall spatial distribution of the femur and tibia. The system first associates and matches the preliminary segmentation results with the original DICOM format CT image data, and identifies the spatial distribution information of various anatomical parts of the entire knee joint through coordinate mapping, highlighting four high-difficulty segmentation areas: the anterior end of the femur, the posterior end of the femur (including the osteophyte area), the anterior end of the tibia, and the posterior end of the tibia. According to the screening criteria of "complex morphology and high segmentation difficulty" defined in the technical disclosure, these four areas are confirmed as key areas, and their corresponding Z-axis slice range in the original CT images is layers 25-68. Subsequently, the image boundaries of each key region were delineated based on this spatial location information. The original CT data was then locally cropped, removing non-critical region data from layers 1-24 and 69-120, generating four sets of local image data corresponding to the four key regions. Each set of data only contains the bone tissue and necessary surrounding environmental information of the corresponding region, significantly reducing the scope of subsequent fine segmentation calculations. This step accurately locates and locally crops complex key regions, focusing on key segmentation points and significantly reducing the computational load of fine segmentation. At the same time, it provides accurate data for targeted fine segmentation, ensuring the efficiency and relevance of subsequent processing.

[0021] In step S4, a dedicated nnU-Net-based fine segmentation model is matched to each key region. Combined with bone tissue detail feature extraction logic, the local image data generated in step S3 is segmented in a targeted manner, ultimately obtaining the fine segmentation results for each key region. Taking the four sets of key region local image data mentioned above as examples, the local image of the posterior femur (including the osteophyte area) is irregular in shape due to the lesion, which is a segmentation challenge. The system first matches pre-trained dedicated nnU-Net fine segmentation models (each model optimizes parameters for the corresponding region's anatomical features) to the anterior femur, posterior femur, anterior tibia, and posterior tibia according to the key region category. After inputting the local image data of the posterior femur into its dedicated model, the model focuses on the gray-level gradient changes of the osteophyte area and captures subtle boundaries through multi-scale feature fusion, improving the attention to local details by 40% compared to the global model; the other three sets of local data are also input into their respective models for targeted segmentation processing. In the final, aggregated fine segmentation results, the boundaries of the osteophyte formation at the posterior end of the femur are complete and continuous, the contour of the articular surface at the anterior end of the tibia is accurate, and there are no structural defects in any key areas, completely compensating for the accuracy deficiencies of the coarse segmentation. This step uses a dedicated fine segmentation model to specifically process key areas, fully explore local detailed features, and significantly improve the segmentation accuracy of complex areas, laying a high-quality foundation for the final result fusion.

[0022] In step S5, the preliminary bone structure segmentation results from step S2 are fused with the fine segmentation results of key regions from step S4, using a spatial alignment algorithm, region replacement logic, and consistency verification mechanism, combined with the requirements for the integrity of the entire knee joint anatomical structure, to generate the final segmentation result. Taking the segmentation data of a patient with mild osteophyte formation as an example, the system first performs spatial alignment on the two types of results, calibrating the position based on the Z-axis coordinate system of CT images to ensure that the fine segmentation results of key regions such as the posterior femur are completely matched with the spatial coordinates of the preliminary segmentation results. Subsequently, according to the aligned positional relationship, the fine segmentation results of four key regions, including the posterior femur (including the osteophyte region) and the anterior tibia, replace the content of the corresponding regions in the preliminary segmentation results, while non-critical regions (such as the mid-shaft femur and mid-shaft tibia) retain the efficient results of the original preliminary segmentation. Next, a consistency check was performed on the replaced overall data to check for contour breaks and boundary misalignments at the junctions between critical and non-critical areas. Two minor junction deviations were corrected through grayscale smoothing. The final integrated whole-knee CT image segmentation result maintained the high efficiency of non-critical area segmentation while achieving precise detail representation of critical areas, fully meeting the bone model registration data requirements for surgical navigation. This step combines coarse segmentation efficiency with fine segmentation accuracy, and ensures the integrity of the results through consistency checks. It resolves the efficiency-accuracy contradiction of traditional methods while meeting the core needs of the surgical scenario.

[0023] In one embodiment, the step of acquiring full knee CT image data and performing overall segmentation processing using a fast segmentation model to obtain preliminary segmentation results for the femoral and tibial images includes: S10: Acquire full knee joint CT image data and use a two-dimensional U-Net residual network model as the fast segmentation model; S11: Split along the Z-axis of the CT image to obtain the image data of each slice; S12: The two-dimensional U-Net residual network model is used to perform segmentation operations on each slice of image data; S13: Summarize the segmentation results of each slice to obtain the preliminary segmentation results of the femur and tibia images.

[0024] In this embodiment, the core process of coarse segmentation for whole knee CT images is first defined. Taking a patient with mild osteophyte formation as an example, the whole knee CT images are in DICOM format, containing 120 Z-axis slices, covering complete three-dimensional information of the femur, tibia, and surrounding soft tissues. After acquiring the patient's whole knee CT image data in the first step, a two-dimensional U-Net residual network model is chosen as the fast segmentation model. This model alleviates the gradient vanishing problem of deep networks through residual block structures. Compared with traditional three-dimensional segmentation models, it improves segmentation speed by more than 30% while maintaining basic positioning accuracy, meeting the high-efficiency requirements of the coarse segmentation stage. Then, slices are split along the Z-axis of the CT image, decomposing the complete three-dimensional CT data into 120 independent two-dimensional slice image data, each slice being 0.6mm thick. Slices 25-68 correspond to the key area in the middle of the knee joint, including the osteophyte formation at the rear of the femur. The split data does not require complex three-dimensional feature mapping, significantly reducing the amount of data processed per step. Next, a two-dimensional U-Net residual network model was used to segment the image data of each slice. The model extracts the gray-level difference features between bone and soft tissue in each slice through the encoder, and restores the bone tissue contour in the decoding stage. For slices containing osteophytes in layers 30-50, the model focuses on identifying the gray-level gradient features between the osteophyte and normal bone tissue to quickly lock the approximate contour. Finally, the segmentation results of each slice were summarized, and the segmentation masks of 120 slices were recombined in Z-axis coordinate order to form a three-dimensional segmentation result covering the entire knee joint. The overall contours of the femoral and tibial images are complete. Although there is slight boundary blurring in the osteophyte areas, the bone structure in non-critical areas is accurately located. The preliminary segmentation results of the femoral and tibial images were successfully obtained, providing basic data for subsequent connected component analysis. This embodiment, through the two-dimensional U-Net residual network and Z-axis slice splitting strategy, significantly improves the segmentation efficiency while ensuring the overall accurate localization of bone tissue, avoiding the high computational cost of the three-dimensional model, and laying an efficient and reliable foundation for subsequent processes.

[0025] In one embodiment, the step of performing connected component analysis on the preliminary segmentation results to extract the largest connected components corresponding to the femoral and tibial images, and obtaining stable and continuous preliminary segmentation results of bone structures, includes: S20: Obtain the preliminary segmentation results, perform connected component analysis on them, and identify all connected components and their corresponding categories; S21: From the identified connected components, filter out the sets of connected components belonging to the femoral and tibial images respectively; S22: Extract the maximum connected component for each of the connected component sets of the femoral and tibial images; S23: Integrate the maximum connected regions of the femoral and tibial images to obtain a stable and continuous preliminary segmentation result of the bone structure.

[0026] In this embodiment, the core objective of connected component analysis is to refine the preliminary segmentation results obtained earlier. The preliminary segmentation results of femoral and tibial images from patients with mild osteophyte formation are used as the processing object. These results have been segmented and reconstructed using a 2D U-Net residual network with 120 Z-axis slices. While they can present the overall contour of the bone structure, they contain redundant information due to soft tissue grayscale interference and slice edge noise. After obtaining this preliminary segmentation result in the first step, the connected component analysis algorithm is started. Based on the 8-neighbor pixel connectivity rule, the entire 3D segmentation mask is traversed to identify all connected components and their corresponding categories. Through a preset grayscale threshold and bone tissue category determination criteria, a total of 11 connected components are identified, of which 6 belong to the femur (including 1 connected component associated with the osteophyte region), 3 belong to the tibia, and 2 belong to non-target tissues (muscle, residual joint fluid noise). The second step involves filtering the 11 identified connected components into femoral and tibial connected component sets based on their category: the femoral set contains 6 connected components with volumes of 2800, 320, 180, 90, 60, and 40 pixels; the tibial set contains 3 connected components with volumes of 2500, 210, and 70 pixels. The two connected components that are not part of the target tissue are excluded due to category mismatch. The third step extracts the largest connected component for each set: in the femoral set, the 2800-pixel connected component accounts for 93% of the total femoral segmentation volume, completely covering the femoral body and the associated bone hyperplasia area, representing the core of the true bone structure; in the tibial set, the 2500-pixel connected component accounts for 94.5% of the total tibial segmentation volume, corresponding to the main tibial structure, while the remaining smaller connected components are noise interference. Finally, by integrating the maximum connected components of the femoral and tibial images, a stable and continuous preliminary segmentation result of the bone structure is generated. In this result, although the contour of the osteophyte region at the posterior end of the femur still has slight roughness, it is free of breaks and redundant noise. The tibial structure is complete and smooth, and the spatial relationship between the two types of bone tissue is consistent with the anatomical structure, fully meeting the spatial constraints for subsequent key region localization. This forms a closed-loop data system that connects with the preceding process and is compatible with subsequent steps. This embodiment, through step-by-step connected component analysis and maximum connected component extraction, accurately eliminates non-target noise, solves the discontinuity problem of the preliminary segmentation result, ensures the integrity of the bone structure, and provides highly reliable data support for subsequent key region localization.

[0027] In one embodiment, the step of locating key areas with complex anatomical structures based on the preliminary segmentation results of the bone structure, and locally cropping the original CT image data to generate local image data includes: S30: Obtain the preliminary segmentation results of the bone structure and the original CT image data, and determine the key areas with complex anatomical structures in the total knee joint based on the preliminary segmentation results of the bone structure. S31: Identify the spatial location information of each key region in the original CT image data; S32: Determine the image range corresponding to each key area based on the spatial location information; S33: Perform local cropping of the original CT image data according to the defined image range to generate local image data corresponding to each key region.

[0028] In this embodiment, the core objective is first to accurately locate complex and critical regions and trim local data based on the preliminary bone structure segmentation results purified previously, laying the foundation for focused processing in the fine segmentation stage. Continuing from the diagnostic and treatment data of a patient with mild osteophyte formation mentioned earlier—the preliminary bone structure segmentation results for this patient have been noise-removed through connected component analysis; the femur (including the osteophyte region) and the main structure of the tibia are stable and continuous; the original CT images are DICOM format data of 120 Z-axis slices with a slice thickness of 0.6 mm. After obtaining the preliminary segmentation results of the bone structure and the original CT image data in the first step, the two were correlated and matched through coordinate mapping. Based on the anatomical standards of the total knee joint, the segmentation results after correlation were analyzed: four major anatomical parts were identified: the anterior femur, the posterior femur (including the osteophyte area), the anterior tibia, and the posterior tibia. Among them, the posterior femur has an irregular shape due to osteophyte formation, and the articular surface texture of the anterior tibia is complex. The segmentation difficulty of these two areas is significantly higher than that of smooth areas such as the mid-shaft femur and mid-shaft tibia. Combined with the screening rules of "complex shape, rich details, and high segmentation difficulty" defined in the technical disclosure, these four parts were identified as key areas with complex anatomical structures in the total knee joint. The second step is to identify the spatial location information of each key region in the original CT image data: By traversing the three-dimensional coordinates of the preliminary bone structure segmentation results, the anterior end of the femur corresponds to layers 15-28 on the Z-axis of the original CT image, the posterior end of the femur (including the osteophyte area) corresponds to layers 30-55, the anterior end of the tibia corresponds to layers 40-62, and the posterior end of the tibia corresponds to layers 65-80. The X and Y axis coordinate ranges of each region are bounded by the edge of the bone structure to ensure complete coverage of the target anatomical site. The third step is to delineate the image range corresponding to each key region based on the above spatial location information: Using the Z-axis slice range as the core, combined with the X and Y axis boundary coordinates, an independent rectangular image range is delineated for each key region. For example, the image range of the posterior end of the femur is layers 30-55 on the Z-axis, 120-380 pixels on the X-axis, and 150-420 pixels on the Y-axis, which includes the complete osteophyte area while avoiding excessive redundant soft tissue data. Finally, the original CT image data was locally cropped according to the defined image range: for the four key regions, corresponding slices and pixel ranges were extracted from the original 120-slice CT data, and non-key region data such as layers 1-14, 29, 56-59, 63-64, and 81-120 were removed, generating four sets of independent local image data. Each set of data only contains bone tissue and necessary surrounding environment information of the corresponding key region, forming a closed-loop data that connects with the previous segmentation results and is suitable for subsequent fine segmentation processing. This embodiment, by accurately locating key regions and performing local cropping, focuses on segmentation difficulties, significantly reduces the computational workload of subsequent fine segmentation, and provides accurate data for targeted fine processing, ensuring the efficiency and relevance of the segmentation process.

[0029] In one embodiment, the step of acquiring the preliminary segmentation results of the bone structure and the original CT image data, and determining the key regions with complex anatomical structures in the total knee joint based on the preliminary segmentation results of the bone structure, includes: S301: Obtain the preliminary segmentation results of the bone structure and the original CT image data, and perform correlation matching between the two; S302: Analyze the preliminary segmentation results of the bone structure after correlation to identify the distribution information of various anatomical parts in the entire knee joint; S303: Based on the distribution information of the anatomical parts, select areas with complex morphological features and high segmentation difficulty; S304: Define the selected areas as key regions with complex anatomy in the entire knee joint and clarify their positional relationship with the original CT image data.

[0030] In this embodiment, the core objective is first to accurately screen complex and critical regions from the associated bone structure data, providing a basis for subsequent local trimming. This continues the previous scenario involving the diagnostic data of a patient with mild osteophyte formation—the patient's preliminary bone structure segmentation results have been purified through connected component analysis; the femoral and tibial images show stable and continuous main structures; the original CT images are DICOM format data with 120 Z-axis slices and a slice thickness of 0.6 mm, and both contain complete anatomical information of the knee joint. After obtaining the preliminary bone structure segmentation results and the original CT image data in the first step, a coordinate association matching process is initiated: using the three-dimensional coordinate system of the original CT image (X-axis horizontal, Y-axis vertical, Z-axis layer-wise) as a reference, a pixel coordinate mapping algorithm is used to precisely bind each pixel in the preliminary bone structure segmentation results to the corresponding spatial coordinates of the original CT image, ensuring complete alignment of the segmentation results with the original image in terms of positional information, and avoiding errors in locating critical regions due to coordinate deviations. The second step is to analyze the preliminary segmentation results of the associated bone structure and identify the distribution information of various anatomical parts in the total knee joint: Based on the part identification rules established by the anatomical atlas of the total knee joint, the three-dimensional segmentation data is traversed, and eight major anatomical parts are successfully identified, including the femoral image (including the head, middle and posterior osteophyte areas), the tibia image (including the anterior articular surface, middle and posterior), and the patellar attachment area. The distribution range of each part on the Z-axis slice (e.g., the posterior femoral part is located in layers 30-55), the spatial span of the XY axis, and the adjacent relationship with surrounding parts are clarified. The third step involves filtering out areas with complex morphological features and high segmentation difficulty based on the distribution information of each anatomical location: the contour complexity and segmentation difficulty of each location are evaluated using morphological analysis algorithms. Among them, the posterior femur has an uneven contour and disordered gray-scale gradient due to bone hyperplasia, with a segmentation difficulty value of 0.87 (out of 1.0); the anterior tibia has fine articular surface texture and blurred edges, with a difficulty value of 0.82; the anterior femoral epiphyseal region has an irregular shape, with a difficulty value of 0.79; the posterior tibia has a deep glenoid fossa structure, with a difficulty value of 0.76. In contrast, the mid-shaft femur and mid-shaft tibia have smooth contours and difficulty values ​​below 0.6, and are therefore excluded from the screening range. Finally, the selected anterior and posterior femur (including areas of bone hyperplasia), anterior and posterior tibia were defined as key areas with complex anatomical structures. Through inverse coordinate mapping, the positional relationship between each key area and the original CT image data was clarified. For example, the key area of ​​the posterior femur corresponds to layers 30-55 on the Z-axis, 120-380 pixels on the X-axis, and 150-420 pixels on the Y-axis of the original CT image. This provides precise coordinates for subsequent image delineation, forming a closed-loop logic that connects with the purification results described above and supports the subsequent cropping process. This embodiment, through association matching and hierarchical screening, accurately identifies anatomically complex areas, clarifies their positional relationship with the original images, provides precise basis for local cropping, improves the targeting and efficiency of subsequent fine segmentation, and ensures the continuity of the segmentation process.

[0031] In one embodiment, the step of performing segmentation processing on the local image data using a fine segmentation model to obtain fine segmentation results for each key local region includes: S40: Obtain the local image data corresponding to each key region, divide them according to the key local regions, and determine that the fine segmentation model is a segmentation model based on nnU-Net; S41: Match the corresponding nnU-Net-based fine segmentation model to each key local region; S42: Input the local image data corresponding to each key local area into the matching fine segmentation model to perform targeted segmentation processing of bone tissue details; S43: Summarize the segmentation output results of each fine segmentation model to obtain the fine segmentation results of bone tissue corresponding to each key local region.

[0032] In this embodiment, the core objective is to perform targeted fine segmentation of the key regional local image data obtained from the previous cropping, compensating for the accuracy deficiencies in the coarse segmentation stage. This continues the scenario of a patient with mild osteophyte formation – this patient has already had four sets of key regional local image data generated through local cropping, corresponding to the anterior femur, posterior femur (including the osteophyte area), anterior tibia, and posterior tibia. Each set of data contains bone tissue and necessary surrounding environmental information for the corresponding region, with no redundant or non-critical data. After obtaining the local image data corresponding to these four key regions in the first step, they are clearly divided according to the anatomical category of the key local regions. The fine segmentation model is determined to be a segmentation model based on nnU-Net – this model has strong feature extraction capabilities in the segmentation of local details in medical images, especially suitable for boundary recognition of complex bone structures. Compared with traditional global models, it significantly improves the segmentation accuracy of fine structures, perfectly matching the fine processing requirements of this stage. The second step is to match a corresponding nnU-Net-based fine segmentation model for each key local region: based on the differences in anatomical features of each region, the anterior femur is matched with an nnU-Net model optimized for "irregular epiphyseal structure" (training data includes 1000+ anterior femoral images), the posterior femur (including osteophyte areas) is matched with a dedicated model optimized for "distinguishing the boundary between osteophyte tissue and normal bone tissue" (training focuses on incorporating osteophyte case data), the anterior tibia is matched with a model adapted to "fine texture of articular surfaces", and the posterior tibia is matched with a model trained for "deep structure of glenoid fossa", ensuring that each model is highly adapted to the segmentation requirements of the corresponding region. The third step involves inputting the local image data of each key local area into the matching fine segmentation model for targeted segmentation of bone tissue details: taking the local image data of the posterior femur as an example, after inputting it into the dedicated model, the model uses a multi-scale feature fusion algorithm to accurately capture the gray-scale gradient difference between the bone hyperplasia area and normal bone tissue, clearly defining the boundary contour of the hyperplasia tissue; after inputting the local data of the anterior tibia into the corresponding model, the model focuses on the subtle texture features of the articular surface, restores the smooth curve of the articular surface, and avoids blurred boundaries. Finally, the segmentation outputs of each fine segmentation model were summarized, resulting in four sets of fine segmentation results for key local regions: the boundary of the osteophyte formation at the posterior end of the femur was complete and continuous, with no omissions or oversegmentation; the contour of the articular surface at the anterior end of the tibia was precise and smooth, and the details of bone tissue in each key region were clearly discernible, with structural integrity and anatomical realism fully meeting the standards, forming a closed-loop data that connects with the previously described local trimming data and supports the fusion of subsequent results. This embodiment, through dedicated model matching and targeted segmentation processing, fully explores the detailed features of bone tissue in key regions, significantly improves the segmentation accuracy of complex regions, provides high-quality data for the final result fusion, and ensures the accuracy required for surgical navigation.

[0033] In one embodiment, the step of fusing the preliminary segmentation result of the bone structure with the fine segmentation result to generate the final segmentation result of the whole knee joint CT image includes: S50: Obtain the preliminary segmentation result of the bone structure and the fine segmentation result of each key local region, and perform spatial alignment processing on the two to maintain consistent spatial position; S51: Based on the aligned spatial relationship, replace the content of the corresponding area in the preliminary segmentation result of the bone structure with the fine segmentation results of each key local area; S52: Perform consistency verification on the replaced overall segmentation data, and integrate the preliminary segmentation content of the bone structure of non-critical areas with the fine segmentation content of critical areas; S53: Summarize and integrate the segmented data to generate the final segmentation result of bone tissue in the full knee CT image.

[0034] In this embodiment, the core objective is to integrate the efficiency advantages of coarse segmentation with the precision advantages of fine segmentation to generate complete and reliable CT image segmentation results for the entire knee joint. This follows the previous scenario involving clinical data from a patient with mild osteophyte formation—the initial bone structure segmentation results for this patient have been purified through connected component analysis, and the fine segmentation results for four key regions (anterior femur, posterior femur containing osteophyte formation, anterior tibia, and posterior tibia) have been processed using a dedicated nnU-Net model. These two types of results respectively cover the overall structure of the entire knee joint and local details. After obtaining the initial bone structure segmentation results and the four sets of fine segmentation results, spatial alignment processing is initiated: using the three-dimensional coordinate system of the original CT image (X-axis horizontal, Y-axis vertical, Z-axis layer-wise) as a unified benchmark, a coordinate calibration algorithm corrects the spatial deviation between the two types of results. For example, the Z-axis coordinate of the fine segmentation result for the posterior femur is precisely aligned with layers 30-55 of the initial segmentation result, with the XY-axis pixel coordinate error controlled within 1 pixel. This ensures that the fine results for the key regions are completely consistent with the overall initial results in spatial position, avoiding boundary misalignment during fusion. The second step, based on the aligned spatial relationships, involves region replacement: the refined segmentation results of the anterior femur replace the corresponding regions of layers 15-28 on the Z-axis in the preliminary segmentation results; the refined segmentation results of the posterior femur (including areas of bone hyperplasia) replace the corresponding content of layers 30-55; the refined segmentation results of the anterior tibia replace the regions of layers 40-62; and the refined segmentation results of the posterior tibia replace the regions of layers 65-80. Non-critical regions (such as the midshaft of the femur and midshaft of the tibia) retain the efficient results of the preliminary segmentation, ensuring accuracy without redundant calculations. The third step performs consistency verification on the overall segmentation data after replacement: using an anatomical structure continuity algorithm, it checks for issues such as contour breaks and abrupt grayscale changes at the junctions between critical and non-critical regions (such as the anterior and midshaft of the femur, and the posterior and midshaft of the tibia). Two minor connection deviations were found (inconsistent grayscale at the boundary between layers 30 and 29 of the femur, and slight misalignment of the contours between layers 65 and 64 of the tibia). After correction using grayscale smoothing and contour fitting algorithms, the overall segmentation data is integrated to obtain logically coherent and structurally complete data. Finally, the integrated segmentation data is summarized to generate the final segmentation result of the bone tissue from the total knee CT image. In this result, non-critical areas maintain a smooth overall outline, while critical areas exhibit precise details—the boundaries of osteophyte formation at the posterior femur are clear and continuous, the articular surfaces at the anterior tibia are smooth and regular, and the anatomical structure of the total knee joint bone tissue highly matches the patient's actual bone structure. This fully meets the accuracy requirements for bone model registration and surgical navigation in total knee replacement surgery, forming a closed-loop logic throughout the entire segmentation process. This embodiment efficiently integrates the two types of segmentation results through spatial alignment, region replacement, and consistency verification, preserving both the efficiency of coarse segmentation and highlighting the precision of fine segmentation. This resolves the efficiency-precision contradiction of traditional methods and meets the core needs of surgical scenarios.

[0035] refer to Figure 2 A total knee CT image segmentation device, comprising: The coarse segmentation module 100 is used to acquire CT image data of the entire knee joint. A fast segmentation model is used for overall segmentation processing to obtain preliminary segmentation results of the femoral and tibial images. The connected component analysis module 200 is used to perform connected component analysis on the preliminary segmentation results, extract the largest connected component corresponding to the femur image and the tibia image, and obtain a stable and continuous preliminary segmentation result of the bone structure. The key region cropping module 300 is used to locate key regions with complex anatomical structures based on the preliminary segmentation results of the bone structure, and to perform local cropping on the original CT image data to generate local image data. The fine segmentation module 400 is used to perform segmentation processing on the local image data using a fine segmentation model to obtain fine segmentation results for each key local region; The result fusion module 500 is used to fuse the preliminary segmentation result of the bone structure with the fine segmentation result to generate the final segmentation result of the whole knee joint CT image.

[0036] Furthermore, the aforementioned coarse segmentation module 100 includes: The CT image data acquisition unit is used to acquire CT image data of the entire knee joint, and a two-dimensional U-Net residual network model is used as the fast segmentation model. Z-axis slice splitting unit is used to split the CT image along the Z-axis direction to obtain slice image data; The slice segmentation operation unit is used to perform segmentation operations on each slice image data using the two-dimensional U-Net residual network model; The preliminary results summary unit is used to summarize the segmentation operation results of each slice to obtain the preliminary segmentation results of the femoral and tibial images.

[0037] Furthermore, the aforementioned connected component analysis module 200 includes: The connected component identification unit is used to obtain the preliminary segmentation result, perform connected component analysis on it, and identify all connected components and their corresponding categories. The target connected component filtering unit is used to filter out the sets of connected components belonging to the femoral and tibia images from the identified connected components, respectively. The maximum connected component extraction unit is used to extract the corresponding maximum connected component from the connected component sets of the femoral and tibial images, respectively. The bone structure result integration unit is used to integrate the maximum connected regions of the femoral and tibial images to obtain stable and continuous preliminary bone structure segmentation results.

[0038] Furthermore, the aforementioned key area clipping module 300 includes: The key region determination unit is used to acquire the preliminary segmentation results of the bone structure and the original CT image data, and to determine the key regions with complex anatomical structures in the total knee joint based on the preliminary segmentation results of the bone structure. The spatial location recognition unit is used to identify the spatial location information of each key area in the original CT image data; The image range delineation unit is used to delineate the image range corresponding to each key area based on the spatial location information. The local image generation unit is used to locally crop the original CT image data according to the defined image range, and generate local image data corresponding to each key area.

[0039] Furthermore, the aforementioned key area determination units include: The data association and matching subunit is used to obtain the preliminary segmentation results of the bone structure and the original CT image data, and to perform association and matching between the two. The anatomical location distribution identification subunit is used to analyze the preliminary segmentation results of the associated bone structure and identify the distribution information of each anatomical location in the entire knee joint. The complex region filtering subunit is used to filter out regions with complex morphological features and high segmentation difficulty based on the distribution information of the anatomical parts. The key region definition subunit is used to define the selected regions as key regions with complex anatomy in the entire knee joint and clarify their positional relationship with the original CT image data.

[0040] Furthermore, the aforementioned fine segmentation module 400 includes: The local image segmentation and model determination unit is used to acquire local image data corresponding to each key region, segment them according to the key local regions, and determine the fine segmentation model as a segmentation model based on nnU-Net. A dedicated model matching unit is used to match the corresponding nnU-Net-based fine segmentation model for each key local region; The targeted segmentation processing unit is used to input the local image data corresponding to each key local area into the matching fine segmentation model to perform targeted segmentation processing of bone tissue details. The fine results summary unit is used to summarize the segmentation output results of each fine segmentation model to obtain the fine segmentation results of bone tissue corresponding to each key local region.

[0041] Furthermore, the above-mentioned result fusion module 500 includes: The spatial alignment processing unit is used to obtain the preliminary segmentation result of the bone structure and the fine segmentation result of each key local region, and to perform spatial alignment processing on the two to maintain spatial consistency. The region content replacement unit is used to replace the content of the corresponding region in the preliminary segmentation result of the bone structure with the fine segmentation result of each key local region based on the aligned spatial position relationship. The overall data integration unit is used to perform consistency verification on the replaced overall segmentation data and integrate the preliminary segmentation content of the bone structure of non-critical areas with the fine segmentation content of critical areas. The final result generation unit is used to summarize and integrate the segmented data to generate the final segmentation result of bone tissue in the full knee CT image.

[0042] Reference Figure 3 This application also provides a computer device, which may be a server, and its internal structure may be as follows: Figure 3As shown, this computer device includes a processor, memory, network interface, and database connected via a bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores operations, computer programs, and the database. The internal memory provides an environment for the operation of the operations and computer programs stored in the non-volatile storage media. The database stores data such as whole-knee CT image segmentation methods. The network interface is used for communication with external terminals via a network connection. When executed by a processor, this computer program implements a method for segmenting CT images of the entire knee joint, comprising: acquiring CT image data of the entire knee joint; performing overall segmentation processing using a fast segmentation model to obtain preliminary segmentation results for femoral and tibial images; performing connected component analysis on the preliminary segmentation results to extract the largest connected components corresponding to the femoral and tibial images, thereby obtaining stable and continuous preliminary segmentation results of bone structures; based on the preliminary segmentation results of bone structures, locating key regions with complex anatomical structures, and performing local cropping on the original CT image data to generate local image data; performing segmentation processing on the local image data using a fine segmentation model to obtain fine segmentation results for each key local region; and fusing the preliminary segmentation results of bone structures with the fine segmentation results to generate the final segmentation result of the CT image of the entire knee joint.

[0043] One embodiment of this application also provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a method for segmenting CT images of the entire knee joint, comprising: acquiring CT image data of the entire knee joint; performing overall segmentation processing using a fast segmentation model to obtain preliminary segmentation results for femoral and tibial images; performing connected component analysis on the preliminary segmentation results to extract the largest connected components corresponding to the femoral and tibial images, thereby obtaining stable and continuous preliminary segmentation results for bone structures; based on the preliminary segmentation results for bone structures, locating key regions with complex anatomical structures, and performing local cropping on the original CT image data to generate local image data; performing segmentation processing on the local image data using a fine segmentation model to obtain fine segmentation results for each key local region; and fusing the preliminary segmentation results for bone structures with the fine segmentation results to generate the final segmentation result of the CT image of the entire knee joint.

[0044] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media provided in this application and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0045] The above description is only a preferred embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural changes made based on the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for segmenting CT images of the entire knee joint, characterized in that, include: Acquire CT images of the entire knee joint, and use a fast segmentation model for overall segmentation to obtain preliminary segmentation results for the femoral and tibial images; Connectivity analysis is performed on the preliminary segmentation results to extract the largest connected components corresponding to the femoral and tibial images, thereby obtaining a stable and continuous preliminary segmentation result of the bone structure. Based on the preliminary segmentation results of the bone structure, key areas with complex anatomical structures are located, and the original CT image data is locally cropped to generate local image data. For the local image data, a fine segmentation model is used for segmentation processing to obtain fine segmentation results for each key local region; The preliminary segmentation results of the bone structure are fused with the fine segmentation results to generate the final segmentation result of the whole knee joint CT image.

2. The method for segmenting total knee joint CT images according to claim 1, characterized in that, The steps of acquiring full knee CT image data, performing overall segmentation using a fast segmentation model, and obtaining preliminary segmentation results for the femoral and tibial images include: Full knee joint CT image data were acquired, and a two-dimensional U-Net residual network model was used as the fast segmentation model. The image data of each slice is obtained by splitting the CT image along the Z-axis direction; The two-dimensional U-Net residual network model is used to perform segmentation operations on each slice of image data; The segmentation results of each slice are summarized to obtain preliminary segmentation results for the femur and tibia images.

3. The method for segmenting total knee joint CT images according to claim 1, characterized in that, The step of performing connected component analysis on the preliminary segmentation results to extract the largest connected components corresponding to the femoral and tibial images, and obtaining stable and continuous preliminary segmentation results of bone structures, includes: The preliminary segmentation results are obtained, and connected component analysis is performed on them to identify all connected components and their corresponding categories. From the identified connected components, the sets of connected components belonging to the femoral and tibial images are selected respectively; For the connected component sets of femoral and tibial images, extract the corresponding maximum connected component for each. By integrating the largest connected regions of the femoral and tibial images, a stable and continuous preliminary segmentation result of the bone structure is obtained.

4. The method for segmenting total knee CT images according to claim 1, characterized in that, The steps of locating key areas with complex anatomical structures based on the preliminary segmentation results of the bone structure, and locally cropping the original CT image data to generate local image data include: The preliminary segmentation results of the bone structure and the original CT image data are obtained, and the key areas with complex anatomical structures in the total knee joint are determined based on the preliminary segmentation results of the bone structure. Identify the spatial location information of each key region in the original CT image data; The image range corresponding to each key area is delineated based on the spatial location information; The original CT image data is locally cropped according to the defined image range to generate local image data corresponding to each key region.

5. The method for segmenting total knee CT images according to claim 4, characterized in that, The steps of obtaining the preliminary segmentation results of the bone structure and the original CT image data, and determining the key areas with complex anatomical structures in the total knee joint based on the preliminary segmentation results of the bone structure, include: The preliminary segmentation results of the bone structure and the original CT image data are obtained, and the two are correlated and matched. Analyze the preliminary segmentation results of the correlated bone structure to identify the distribution information of various anatomical parts in the entire knee joint; Based on the distribution information of the anatomical sites, areas with complex morphological features and high segmentation difficulty are selected; The selected areas were defined as key regions with complex anatomy in the entire knee joint, and their positional relationship with the original CT image data was clarified.

6. The method for segmenting total knee CT images according to claim 1, characterized in that, The step of performing segmentation processing on the local image data using a fine segmentation model to obtain fine segmentation results for each key local region includes: Local image data corresponding to each key region are obtained, and the data are divided according to the key local regions. The fine segmentation model is determined to be a segmentation model based on nnU-Net. Match the corresponding nnU-Net-based fine segmentation model to each key local region; The local image data corresponding to each key local area are input into the matching fine segmentation model to perform targeted segmentation processing of bone tissue details; By summarizing the segmentation outputs of each fine segmentation model, the fine segmentation results of bone tissue corresponding to each key local region are obtained.

7. The method for segmenting total knee joint CT images according to claim 1, characterized in that, The step of fusing the preliminary segmentation result of the bone structure with the fine segmentation result to generate the final segmentation result of the whole knee joint CT image includes: The preliminary segmentation results of the bone structure and the fine segmentation results of each key local region are obtained, and the two are spatially aligned to maintain consistent spatial positions. Based on the aligned spatial relationships, the fine segmentation results of each key local region replace the content of the corresponding region in the preliminary segmentation results of the bone structure. Perform consistency verification on the replaced overall segmentation data, and integrate the preliminary segmentation content of the bone structure of non-critical areas with the fine segmentation content of critical areas; The segmented data is aggregated and integrated to generate the final segmentation result of bone tissue from the full knee CT image.

8. A total knee joint CT image segmentation device, characterized in that, include: The coarse segmentation module is used to acquire CT image data of the entire knee joint. A fast segmentation model is used for overall segmentation processing to obtain preliminary segmentation results of femoral and tibial images. The connected component analysis module is used to perform connected component analysis on the preliminary segmentation results, extract the largest connected component corresponding to the femur image and tibia image, and obtain a stable and continuous preliminary segmentation result of bone structure. The key region cropping module is used to locate key regions with complex anatomical structures based on the preliminary segmentation results of the bone structure, and to perform local cropping on the original CT image data to generate local image data. The fine segmentation module is used to perform segmentation processing on the local image data using a fine segmentation model to obtain fine segmentation results for each key local region; The result fusion module is used to fuse the preliminary segmentation results of the bone structure with the fine segmentation results to generate the final segmentation result of the whole knee joint CT image.

9. A method and device for segmenting CT images of the entire knee joint, characterized in that, include: processor; Memory, used to store computer programs; The processor executes the computer program to implement the whole knee CT image segmentation method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the total knee CT image segmentation method according to any one of claims 1 to 7.

Images