Lumbar intervertebral disc herniation identification method and system based on CT image

Abstract

The invention discloses a lumbar disc herniation identification method and system based on a CT image, and the method comprises the steps: carrying out the multi-dimensional projection slice processing of input CT image data, and obtaining a multi-dimensional feature projection slice image data set; according to the multi-dimensional feature projection slice image data set, constructing an adaptive expansion bounding box model, and automatically generating a segmentation boundary of the lumbar intervertebral disc by using a nonlinear constraint iteration function to obtain a lumbar intervertebral disc three-dimensional voxel model of a target segmentation area; for the lumbar intervertebral disc three-dimensional voxel model, outputting a high-dimensional lumbar vertebra texture feature vector by using a fiber enhanced tensor network; and according to the high-dimensional lumbar vertebra texture feature vector, combining anatomical appearance features of the lumbar intervertebral disc at different positions in the multi-dimensional projection slice to construct a multi-dimensional fusion label classification model so as to output a normal or protruding identification result of the lumbar intervertebral disc. According to the embodiment of the invention, modern image processing and deep learning technologies can be combined, and the recognition rate and accuracy of lumbar disc herniation can be effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of image recognition, and in particular to a method and system for recognizing lumbar disc herniation based on CT images. Background Art

[0002] Lumbar disc herniation is a common spinal disease, usually manifested by symptoms such as low back pain and radiating pain in the lower limbs. In severe cases, it can lead to motor dysfunction, which has a significant impact on the patient's life. Therefore, early and accurate identification and diagnosis of lumbar disc herniation is crucial to improving the treatment effect and prognosis of patients.

[0003] Traditional diagnostic methods for lumbar disc herniation mainly rely on the doctor's experience and judgment, including analysis of clinical symptoms and imaging examinations such as X-rays, CT, and MRI. However, CT images have become a commonly used imaging examination method due to their fast imaging speed, high resolution, and clearer display of bone tissue. However, traditional CT image analysis has certain limitations, resulting in low recognition accuracy of lumbar disc herniation. Existing technologies often face problems such as noise interference, artifact influence, and insufficient understanding of the complexity of anatomical structures when processing CT images, making it difficult to fully extract the subtle features of the lumbar disc area. Summary of the invention

[0004] The purpose of the present invention is to provide a lumbar disc herniation identification method and system based on CT images to address the deficiencies in the prior art, and to combine modern image processing and deep learning technology to effectively improve the recognition rate and accuracy of lumbar disc herniation.

[0005] An embodiment of the present application provides a method for identifying lumbar disc herniation based on CT images, the method comprising: According to the anatomical characteristics of the lumbar intervertebral disc, the input CT image data is processed by multi-dimensional projection slicing. The two-dimensional cross-sectional images of multiple key projection angles are determined by projection. The high-contrast features of the lumbar intervertebral disc area are extracted from different key projection angles. The nonlinear filtering algorithm is combined to suppress noise and artifact interference while retaining edge details to obtain an optimized multi-dimensional feature projection slice image data set. According to the multi-dimensional feature projection slice image data set, an adaptive extended bounding box model is constructed. The area where the lumbar intervertebral disc is located is taken as the target, and the size and shape of the segmentation area are dynamically adjusted. The segmentation boundary of the lumbar intervertebral disc is automatically generated by using the nonlinear constraint iteration function in set theory, so as to obtain the target segmentation area based on the segmentation boundary segmentation. In the segmentation process, the intervertebral disc morphology continuity constraint calculated between adjacent layers is combined to further exclude the protruding non-target area, and obtain the lumbar intervertebral disc three-dimensional voxel model of the target segmentation area. For the 3D voxel model of lumbar intervertebral disc, the fiber-enhanced tensor network is used to extract the texture features of the internal microstructure of the lumbar intervertebral disc and output a high-dimensional lumbar texture feature vector. According to the high-dimensional lumbar texture feature vector and combined with the anatomical appearance features of the lumbar intervertebral disc in different positions and multi-dimensional projection slices, a multi-dimensional fusion label classification model is constructed to output the recognition results of whether the lumbar intervertebral disc is normal or protruding.

[0006] Optionally, the input CT image data is subjected to multi-dimensional projection slicing processing according to the anatomical structure characteristics of the lumbar intervertebral disc, two-dimensional cross-sectional images of multiple key projection angles are determined by projection, high contrast features of the lumbar intervertebral disc area are extracted from different key projection angles, and noise and artifact interference are suppressed by combining a nonlinear filtering algorithm while retaining edge details to obtain an optimized multi-dimensional feature projection slice image data set, including: Receiving CT image data to be processed and ensuring that the quality of the image data meets certain standards, wherein the CT image is three-dimensional volume data and is composed of multiple two-dimensional slices; According to the anatomical characteristics of the lumbar intervertebral disc, the lumbar intervertebral disc area is preliminarily marked using a recognition algorithm to ensure that the area of ​​interest can be located in the three-dimensional volume data; The marked lumbar intervertebral disc region is subjected to multi-dimensional projection slicing processing, by selecting a plurality of key projection angles in the three-dimensional CT image data for projection, and each projection angle corresponds to generating a set of two-dimensional projection slice images as two-dimensional cross-sectional images, wherein the two-dimensional cross-sectional images can provide cross-sectional views of the lumbar intervertebral disc observed from different directions, and help identify anatomical structural features with important diagnostic value, and the two-dimensional cross-sectional images at least include sagittal plane, coronal plane, and transverse plane; In each set of two-dimensional projection slice images generated, adaptive contrast enhancement technology is used to dynamically enhance the contrast of the image based on local histogram equalization or contrast-limited adaptive histogram equalization to more clearly identify the edges and fine structures of the lumbar intervertebral disc and improve the prominence of specific anatomical structures; The contrast-enhanced two-dimensional projection slice image is filtered by a nonlinear filtering algorithm to remove noise and artifact interference in the image while retaining edge details in the image, wherein the nonlinear filtering algorithm is bilateral filtering or non-local mean filtering. The bilateral filtering can retain edge features during the denoising process, and the non-local mean filtering can retain details by considering similar pixel blocks; The filtered two-dimensional projection slice images are combined into a multi-dimensional feature projection slice image dataset, in which each slice provides high-contrast features extracted from different projection angles, and all slices together form a complete, multi-angle observation of lumbar intervertebral disc image feature set.

[0007] Optionally, the method constructs an adaptive extended bounding box model based on a multi-dimensional feature projection slice image data set, takes the area where the lumbar intervertebral disc is located as a target, dynamically adjusts the size and shape of the segmented area, and uses a nonlinear constraint iterative function in set theory to automatically generate a segmentation boundary of the lumbar intervertebral disc, so as to obtain a target segmentation area based on the segmentation boundary segmentation, wherein, in the segmentation process, the intervertebral disc morphology continuity constraint calculated between adjacent layers is combined to further exclude the protruding non-target area, and obtain a lumbar intervertebral disc three-dimensional voxel model of the target segmentation area, including: Preliminarily extracting the position of the lumbar intervertebral disc from the multi-dimensional feature projection slice image data set, and enclosing the lumbar intervertebral disc with an initial bounding box, wherein the initial bounding box is determined based on prior information or obtained through morphological operations, wherein the prior information includes an average lumbar intervertebral disc size, and the morphological operations include edge detection of the image; Adopting an adaptive adjustment strategy based on image gradient or edge information, the bounding box is dynamically adjusted according to the specific shape and size of the lumbar intervertebral disc, and each face and corner of the bounding box is fine-tuned in real time to make the bounding box close to the edge of the target area; A nonlinear constrained iterative function in set theory is applied to optimize the boundary of the bounding box. The nonlinear constrained iterative function is based on multiple iterations. In each iteration, the bounding box is fine-tuned by calculating the similarity between the current bounding box and the target area pixels, so as to gradually approach the segmentation boundary, so as to obtain the target segmentation area based on the segmentation boundary segmentation. In each iteration, an energy minimization method is used to ensure that the bounding box will not be excessively disturbed by noise points during the convergence process. The energy minimization is achieved by a graph cut method or a level set method. In the process of bounding box adjustment, the morphological continuity constraint between adjacent layers is further combined. The morphological continuity constraint includes analyzing the morphological and positional changes of the lumbar intervertebral disc in adjacent slices, and predicting the possible position and shape in the current slice through historical information. The morphological continuity constraint is used to smooth the segmentation result so that the segmented lumbar intervertebral disc morphology is continuous in three-dimensional space without unreasonable mutations. During the segmentation process, multiple discrimination criteria are applied to filter out non-target areas, including boundary clarity, tissue density and expected anatomical position, and each segmented two-dimensional projection slice is integrated into a three-dimensional voxel model of the lumbar intervertebral disc of a complete target segmentation area to reflect the true three-dimensional morphology of the lumbar intervertebral disc.

[0008] Optionally, the three-dimensional voxel model of the lumbar intervertebral disc is used to extract the texture features of the internal microstructure of the lumbar intervertebral disc using a fiber-enhanced tensor network, and output a high-dimensional lumbar texture feature vector, including: The multi-layer convolution structure and tensor module of the fiber-reinforced tensor network FatNet are used to perform multi-scale convolution operations on the 3D voxel model data of the lumbar intervertebral disc. During the multi-scale convolution operation, the fiber-reinforced tensor network can extract texture features from the macroscopic to the microscopic level. The extracted texture features are subjected to high-dimensional modeling to transform the texture features into high-dimensional lumbar texture feature vectors with biological significance to reflect the internal structure and texture of the lumbar intervertebral disc.

[0009] Optionally, the method of constructing a multidimensional fusion label classification model based on the high-dimensional lumbar texture feature vector and the anatomical appearance features of the lumbar intervertebral disc in different positions and multidimensional projection slices to output the recognition result of whether the lumbar intervertebral disc is normal or protruding includes: Combined with the anatomical features of the lumbar intervertebral disc in different positions and multi-dimensional projection slices, a deep learning model was selected for training to obtain a multi-dimensional fusion label classification model; According to the high-dimensional lumbar texture feature vector, the trained multi-dimensional fusion label classification model is used to output the lumbar intervertebral disc recognition results corresponding to the CT image data, wherein the recognition results include: normal and protruding.

[0010] Another embodiment of the present application provides a lumbar disc herniation identification system based on CT images, the system comprising: The projection module is used to perform multi-dimensional projection slicing processing on the input CT image data according to the anatomical structure characteristics of the lumbar intervertebral disc, determine the two-dimensional cross-sectional images of multiple key projection angles by projection, extract the high-contrast features of the lumbar intervertebral disc area from different key projection angles, and combine the nonlinear filtering algorithm to suppress noise and artifact interference while retaining edge details to obtain an optimized multi-dimensional feature projection slice image data set; The segmentation module is used to construct an adaptive extended bounding box model based on a multi-dimensional feature projection slice image data set, take the area where the lumbar intervertebral disc is located as the target, dynamically adjust the size and shape of the segmentation area, and use the nonlinear constraint iteration function in set theory to automatically generate the segmentation boundary of the lumbar intervertebral disc, so as to obtain the target segmentation area based on the segmentation boundary segmentation. In the segmentation process, the intervertebral disc morphology continuity constraint calculated between adjacent layers is combined to further exclude the protruding non-target area, so as to obtain a 3D voxel model of the lumbar intervertebral disc in the target segmentation area; The extraction module is used to extract the texture features of the internal microstructure of the lumbar intervertebral disc using a fiber-enhanced tensor network for the three-dimensional voxel model of the lumbar intervertebral disc, and output a high-dimensional lumbar texture feature vector; The output module is used to construct a multi-dimensional fusion label classification model based on the high-dimensional lumbar texture feature vector and the anatomical appearance features of the lumbar intervertebral disc in different positions and multi-dimensional projection slices to output the recognition results of whether the lumbar intervertebral disc is normal or protruding.

[0011] Yet another embodiment of the present application provides a storage medium, wherein the storage medium stores a computer program, wherein the computer program is configured to execute any of the above methods when running.

[0012] Yet another embodiment of the present application provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute any of the methods described above.

[0013] Compared with the prior art, the present invention provides a lumbar disc herniation recognition method based on CT images, which performs multi-dimensional projection slicing processing on the input CT image data to obtain a multi-dimensional feature projection slicing image data set; based on the multi-dimensional feature projection slicing image data set, an adaptive extended bounding box model is constructed, and a nonlinear constrained iterative function is used to automatically generate a segmentation boundary of the lumbar disc to obtain a three-dimensional voxel model of the lumbar disc in the target segmentation area; for the three-dimensional voxel model of the lumbar disc, a fiber-reinforced tensor network is used to output a high-dimensional lumbar texture feature vector; based on the high-dimensional lumbar texture feature vector, combined with the anatomical appearance features of the lumbar disc in different positions and multi-dimensional projection slices, a multi-dimensional fusion label classification model is constructed to output the recognition result of normal or herniated lumbar disc, so as to combine modern image processing and deep learning technology to effectively improve the recognition rate and accuracy of lumbar disc herniation. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A hardware structure block diagram of a computer terminal for a method for identifying lumbar disc herniation based on CT images provided in an embodiment of the present invention; Figure 2 A flow chart of a method for identifying lumbar disc herniation based on CT images provided by an embodiment of the present invention; Figure 3 A schematic structural diagram of a lumbar disc herniation identification system based on CT images provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, but should not be construed as limiting the present invention.

[0016] The embodiment of the present invention firstly provides a method for identifying lumbar disc herniation based on CT images. The method can be applied to electronic devices such as computer terminals, specifically ordinary computers, etc.

[0017] The following describes it in detail by taking running on a computer terminal as an example. Figure 1The hardware structure block diagram of a computer terminal for a method for identifying lumbar disc herniation based on CT images provided by an embodiment of the present invention. Figure 1 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0018] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any method for identifying lumbar disc herniation based on CT images.

[0019] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0020] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any lumbar disc herniation identification method based on CT images.

[0021] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0022] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0023] See also Figure 2 The embodiment of the present invention provides a method for identifying lumbar disc herniation based on CT images, which may include the following steps: S201, according to the anatomical structure characteristics of the lumbar intervertebral disc, the input CT image data is subjected to multi-dimensional projection slicing processing, two-dimensional cross-sectional images of multiple key projection angles are determined by projection, high-contrast features of the lumbar intervertebral disc area are extracted from different key projection angles, and noise and artifact interference are suppressed by combining a nonlinear filtering algorithm while retaining edge details, thereby obtaining an optimized multi-dimensional feature projection slicing image data set; The key to this step is to make full use of the anatomical features of the lumbar intervertebral disc and obtain its cross-sectional images at different viewing angles through multi-dimensional projection technology, laying the foundation for subsequent feature extraction and recognition. By extracting high-contrast features from multiple key projection angles, the morphological changes and pathological characteristics of the lumbar intervertebral disc can be better displayed. At the same time, the nonlinear filtering algorithm is combined to effectively suppress noise and artifacts in the image and retain details in the image, with the aim of ensuring the data quality and recognizability for subsequent analysis.

[0024] The significance of this step is to provide a high-quality multi-dimensional feature data set based on CT images for the image analysis of lumbar intervertebral discs. The optimized projection images not only improve the visibility and information content of the images, but also provide reliable data support for subsequent lumbar intervertebral disc segmentation and morphological analysis. This method helps doctors quickly and accurately identify abnormal lesions during the diagnosis process, and improves the scientificity and accuracy of clinical decision-making.

[0025] Specifically, CT image data to be processed may be received, and quality of the image data may be ensured to meet certain standards, wherein the CT image is three-dimensional volume data and is composed of a plurality of two-dimensional slices; The main task of this step is to conduct a preliminary quality check and confirmation of the CT images to be processed. Only CT images that meet certain standards can be included in the subsequent processing flow. Confirming that the CT image is three-dimensional volume data is the key to this step, because only three-dimensional data contains enough information for subsequent multi-dimensional processing. Ensuring image quality can ensure the accuracy and reliability of subsequent analysis. If the image quality does not meet the standards, it may lead to deviations in data processing, thereby affecting the final diagnosis results.

[0026] In this step, the CT image data to be processed must first be retrieved from the image storage system. Usually, this data is stored in DICOM format and contains three-dimensional volume data composed of multiple two-dimensional slices. During the retrieval process, it should be ensured that the image data has clear identification in terms of time, patient, and examination type to facilitate subsequent processing and analysis. At the same time, basic format verification of the image data is required to ensure its integrity and consistency to avoid analysis errors caused by data corruption or format incompatibility.

[0027] After verifying the integrity of the image data, the next step is to conduct a preliminary assessment of the image quality. This step can be achieved with the help of computer image processing tools, such as using custom algorithms to analyze the contrast, clarity, and noise level of the image. By calculating indicators such as the signal-to-noise ratio (SNR) and the standard deviation of contrast, it can be determined whether the image meets the standards for subsequent analysis. If the quality of the image does not meet the requirements, it may be necessary to improve its quality through image enhancement techniques, such as applying image filtering algorithms to remove unnecessary noise.

[0028] Finally, the quality-assessed imaging data will be classified and stored in a structured database to ensure standardized data management and easy retrieval. The database should contain basic information about the imaging, such as shooting date, equipment model, patient ID, etc., to improve data traceability and availability. Successful implementation at this stage will lay a solid foundation for subsequent analysis and ensure that all subsequent steps are supported by high-quality data.

[0029] According to the anatomical characteristics of the lumbar intervertebral disc, the lumbar intervertebral disc area is preliminarily marked using a recognition algorithm to ensure that the area of ​​interest can be located in the three-dimensional volume data; This step aims to mark the region of interest using the anatomical features of the lumbar disc, laying the foundation for subsequent projection slice processing. In this process, computer vision or morphological methods are used to perform adaptive region marking based on anatomical knowledge. Preliminary marking helps improve the efficiency and accuracy of subsequent processing, ensures that subsequent analysis focuses on relevant structures, reduces unnecessary computational burden, and avoids interference with non-target areas.

[0030] In this step, you first need to build a model that contains the anatomical features of the lumbar disc and prepare the training data set required for labeling. Use the labeled images as the training set and select a suitable machine learning algorithm, such as a convolutional neural network (CNN), so that the model can learn the anatomical features of the lumbar disc. Through a programmatic method, a large amount of sample data is input for training, so that the model can automatically identify the area of ​​the lumbar disc when receiving new CT images.

[0031] The recognition algorithm usually needs to be trained and validated multiple times to improve its accuracy. After each training, an independent validation set is needed to evaluate the performance of the model, and it is iteratively optimized by adjusting hyperparameters, increasing the number of training times, etc. When the model performs well on the validation set, it can be applied to actual CT image data. At this point, the model will automatically identify the area of ​​the lumbar disc and generate a preliminary labeled image indicating the location of the lumbar disc.

[0032] Finally, the resulting labeled image is post-processed to optimize its accuracy. This step can include applying morphological operations such as dilation and erosion to the identified areas to remove small noise and smooth the boundaries. Through these operations, possible erroneous labels can be further filtered out. At the same time, some analysis rules can be set to ensure the rationality of the labeled area, such as judging the size and shape of the lumbar intervertebral disc and its position relative to the surrounding structures based on conventional anatomical knowledge. After processing, the final labeling results will be stored in the database for subsequent multi-dimensional projection slice processing.

[0033] The marked lumbar intervertebral disc region is subjected to multi-dimensional projection slicing processing, by selecting a plurality of key projection angles in the three-dimensional CT image data for projection, and each projection angle corresponds to generating a set of two-dimensional projection slice images as two-dimensional cross-sectional images, wherein the two-dimensional cross-sectional images can provide cross-sectional views of the lumbar intervertebral disc observed from different directions, and help identify anatomical structural features with important diagnostic value, and the two-dimensional cross-sectional images at least include sagittal plane, coronal plane, and transverse plane; This step generates two-dimensional cross-sectional images in different directions by performing multi-dimensional projection slicing on the marked lumbar disc area. This process makes it possible to observe the anatomical structure of the lumbar disc from multiple perspectives for comprehensive diagnostic analysis. This slicing process can provide multi-angle cross-sectional views to help doctors better understand the biomechanical properties and pathological changes of the lumbar disc, making the location of the lesion area and the extraction of its characteristics more accurate.

[0034] In this step, it is first necessary to select multiple key projection angles from the three-dimensional CT image data according to the marked lumbar disc area. The projection angles usually selected include sagittal plane, coronal plane and cross-section, etc. These angles can cover the anatomical characteristics of the lumbar disc to the greatest extent. Using three-dimensional reconstruction technology, the algorithm is applied to convert the image data of the selected projection angle into the corresponding two-dimensional slice image, and each slice clearly shows the lumbar disc and its surrounding anatomical structure.

[0035] In the process of generating two-dimensional cross-sectional images, interpolation algorithms, such as linear interpolation or cubic interpolation, can be used to improve the smoothness and detail clarity of slice images. By performing appropriate color mapping and enhancement processing on images in different directions, it is ensured that the images not only provide the necessary information of the anatomical structure, but also enhance the visual expression, making it easier for clinicians to identify and diagnose. In addition, by applying a variety of image processing techniques, it can be ensured that each projection slice retains edge details while minimizing information loss.

[0036] Finally, all the generated two-dimensional projection slice images are organized into a structured data set for subsequent analysis and processing. For each slice, the corresponding projection angle information should be recorded for classification and retrieval in subsequent steps. This multidimensional data set will serve as an important basis for subsequent image analysis and provide important support and reference for feature extraction and pathological research.

[0037] In each set of two-dimensional projection slice images generated, adaptive contrast enhancement technology is used to dynamically enhance the contrast of the image based on local histogram equalization or contrast-limited adaptive histogram equalization to more clearly identify the edges and fine structures of the lumbar intervertebral disc and improve the prominence of specific anatomical structures; Adaptive contrast enhancement technology is used to improve the visual effect of each set of two-dimensional projection slice images, so as to more clearly present the edge and subtle structural features of the lumbar intervertebral disc. This processing is extremely important for the identification and analysis of pathological structures. Improving image contrast can significantly enhance the visibility of edge features, making it easier for diagnosticians to identify abnormal changes and key features of anatomical structures, thereby enhancing the accuracy of medical decisions.

[0038] The core of this step is to apply adaptive contrast enhancement technology to improve the visibility of the generated 2D projection slice images. First, each 2D slice image is divided into multiple small blocks (for example, 8x8 or 16x16 pixel grids), and then the local histogram is calculated for each small block. By equalizing the local histogram, the contrast of the image within the small block can be effectively improved, making the details in the image more prominent. This process emphasizes local characteristics and can more effectively display the anatomical structure of interest.

[0039] Next, the contrast-limited adaptive histogram equalization (CLAHE) algorithm is used to avoid the loss of image details that may be caused by global equalization. The CLAHE algorithm dynamically adjusts the enhancement process according to the contrast and brightness of the local area, so that the contrast between different areas is widened and it is not easy to produce visual artifacts or over-saturation. The processed image not only shows clearer edge features, but also more accurately presents the differences in the tissue structure of the lumbar disc and its adjacent areas.

[0040] Finally, the enhanced image will be saved and added to the multidimensional feature projection slice image dataset to provide visual support for subsequent analysis. The enhanced image provides clinicians with clear anatomical information, which helps improve the accuracy and efficiency of diagnosis. The successful implementation of this step not only improves the quality of the image, but also lays the foundation for subsequent feature extraction and pathological analysis.

[0041] The contrast-enhanced two-dimensional projection slice image is filtered by a nonlinear filtering algorithm to remove noise and artifact interference in the image while retaining edge details in the image, wherein the nonlinear filtering algorithm is bilateral filtering or non-local mean filtering. The bilateral filtering can retain edge features during the denoising process, and the non-local mean filtering can retain details by considering similar pixel blocks; On the contrast-enhanced image, a nonlinear filtering algorithm is used to further clarify it, with the focus on removing noise and artifacts while retaining the edge details in the image to the maximum extent. This step ensures that the final output image can truly reflect the structure of the lumbar intervertebral disc. Effective noise suppression can significantly improve the usability of the image, reduce the risk of misidentification during image analysis, provide a purer image foundation for subsequent segmentation and analysis steps, and ensure the accuracy of the diagnostic results.

[0042] In this processing step, choosing a suitable nonlinear filtering algorithm is crucial for image clarity and detail preservation. First, a bilateral filtering algorithm is applied. The core of this algorithm is to make full use of spatial distance and color difference to perform weighted processing on pixels. By setting appropriate spatial and color domain standards, similar pixels are aggregated together to achieve effective denoising while retaining important edge details in the image. This method is particularly suitable for complex structures in medical images and can minimize the impact of noise without losing key anatomical information.

[0043] Secondly, non-local mean filtering can be considered. This algorithm compares similar blocks in the image and uses the information of the entire image to average a single pixel, thereby effectively suppressing noise and maintaining details. This processing method is suitable for images with complex backgrounds. It can accurately retain subtle structural features while removing noise, ensuring that the edges and tiny features of the lumbar intervertebral disc can be presented more clearly.

[0044] The filtered images will be saved in the database and used for subsequent feature extraction and analysis. This step ensures the clarity and usability of the images, provides clean data support for subsequent diagnosis, and significantly reduces the risk of misinterpretation. Images processed by nonlinear filtering not only enable doctors to perform pathological analysis more accurately, but also provide a solid foundation for subsequent model training and clinical decision-making.

[0045] The filtered two-dimensional projection slice images are combined into a multi-dimensional feature projection slice image dataset, in which each slice provides high-contrast features extracted from different projection angles, and all slices together form a complete, multi-angle observation of lumbar intervertebral disc image feature set.

[0046] This step combines each processed two-dimensional projection slice image to form a multidimensional feature projection slice image dataset, thereby gathering information from different angles and providing an integrated view of multidimensional observation. This collection provides a comprehensive data foundation for subsequent functional analysis and model training. The construction of a multidimensional feature projection slice image dataset ensures the comprehensive presentation of the structural characteristics of the lumbar intervertebral disc from different perspectives, providing irreplaceable information support for subsequent segmentation, feature extraction and recognition, making the entire analysis process more systematic and coherent.

[0047] In this step, all filtered 2D projection slice images need to be integrated first. In order to ensure the organization and efficiency of the data, a unified storage format needs to be designed so that images from different perspectives can be effectively classified and managed. An image database system can usually be used to ensure that the meta-information of each slice (such as projection angle, processing date, processing algorithm, etc.) can be clearly recorded to facilitate subsequent retrieval and analysis.

[0048] Next, each slice is labeled and indexed to facilitate subsequent data retrieval and call. Slices can be classified by projection angle, processing time, or patient information by setting a unified naming rule to ensure the ease of use of the data set. In addition, in order to improve the overall data access efficiency, consider compressing the slices for storage while ensuring that the image quality is not affected, which can provide a fast response time for subsequent analysis.

[0049] Finally, the integrated multi-dimensional feature projection slice image dataset will serve as the basis for subsequent analysis, providing comprehensive support for feature extraction and imaging research of the lumbar intervertebral disc. This process ensures the consistency and integrity of the structure of the lumbar intervertebral disc observed from different angles in the same dataset, allowing doctors to obtain comprehensive imaging information on a single platform when making diagnoses, thereby improving the scientificity and accuracy of clinical decision-making.

[0050] S202, constructing an adaptive extended bounding box model based on the multi-dimensional feature projection slice image data set, taking the area where the lumbar intervertebral disc is located as the target, dynamically adjusting the size and shape of the segmented area, and automatically generating the segmentation boundary of the lumbar intervertebral disc by using the nonlinear constraint iteration function in set theory, so as to obtain the target segmentation area based on the segmentation boundary segmentation, wherein, in the segmentation process, combining the intervertebral disc morphology continuity constraint calculated between adjacent layers, further excluding the protruding non-target area, and obtaining the lumbar intervertebral disc three-dimensional voxel model of the target segmentation area; The core of this step is to adapt to the changing shape and size of the lumbar intervertebral disc by constructing a flexible bounding box. The initial bounding box is obtained based on prior knowledge and image processing, and the box is made to fit the target area through an adaptive adjustment algorithm. Combined with the iterative function of nonlinear constraints, the boundary of the bounding box is continuously optimized to ensure that the anatomical structure of the lumbar intervertebral disc is accurately captured to form an accurate three-dimensional voxel model. The implementation of this step not only improves the segmentation accuracy of the lumbar intervertebral disc, but also ensures that the final generated three-dimensional voxel model has good continuity and real biological characteristics. By applying the intervertebral disc morphology continuity constraint between adjacent layers, the erroneous areas introduced by noise or other irrelevant factors during the segmentation process can be effectively avoided, thereby improving the accuracy and reliability of the model. In addition, the constructed three-dimensional voxel model will become an important basis for subsequent analysis and pathological research, and has important guiding value for the research and clinical diagnosis of medical imaging.

[0051] Specifically, the position of the lumbar intervertebral disc can be initially extracted from the multi-dimensional feature projection slice image data set, and the lumbar intervertebral disc is enclosed by an initial bounding box, wherein the initial bounding box is determined based on prior information or obtained through morphological operations, wherein the prior information includes an average lumbar intervertebral disc size, and the morphological operations include edge detection of the image; In this step, the preliminary position of the lumbar intervertebral disc is first extracted from the multidimensional feature projection slice image dataset, and an initial bounding box is created for it. The construction of this bounding box can be based on known prior information, such as the average size of the lumbar intervertebral disc, or through image processing techniques such as morphological operations and edge detection to detect significant structures in the image. Through these methods, it can be ensured that the preliminary bounding box can cover the lumbar intervertebral disc well and serve as the starting point for subsequent adjustments. The accuracy of the initial bounding box directly affects the subsequent segmentation results. If the bounding box can surround the target area as accurately as possible, it will be able to effectively reduce the complexity and amount of calculation of subsequent adjustments, and lay the foundation for efficient and accurate lumbar intervertebral disc segmentation. The successful implementation of this stage provides the necessary conditions for subsequent dynamic adjustment and refined segmentation.

[0052] First, the image processing algorithm is applied to extract the preliminary location of the lumbar intervertebral disc from the multidimensional feature projection slice image data set. The threshold-based segmentation method is used to identify the high-density areas in the CT images, that is, the possible lumbar intervertebral disc areas, by setting appropriate thresholds. For this purpose, automatic threshold algorithms such as Otsu's threshold method can be used to automatically determine the appropriate segmentation value. Then, the morphological operation method is used to enhance the recognition effect, and the target structure is refined using techniques such as expansion and erosion, so as to more accurately determine the location of the lumbar intervertebral disc.

[0053] After the preliminary position is determined, an initial bounding box is configured based on prior information. The prior information can be the average size of the lumbar disc in historical data, or a reasonable size range derived from anatomical knowledge. Through geometric operations, the length, width and height of the bounding box are set to values ​​that match the average lumbar disc size. At this point, the initial bounding box can serve as the basis for subsequent adjustments, covering the main area of ​​the lumbar disc and ensuring that no important structures are missed in subsequent processing.

[0054] Finally, the accuracy of the bounding box is further optimized using an edge detection algorithm. By applying the Canny edge detection algorithm, edge features in the image can be accurately identified. The identified edge information will be used to adjust the position and shape of the bounding box to better fit the actual contour of the lumbar disc. This integration process will provide a reasonable starting point for subsequent steps, ensuring that the foundation for accurate segmentation is laid from the beginning.

[0055] Adopting an adaptive adjustment strategy based on image gradient or edge information, the bounding box is dynamically adjusted according to the specific shape and size of the lumbar intervertebral disc, and each face and corner of the bounding box is fine-tuned in real time to make the bounding box close to the edge of the target area; This step mainly dynamically adjusts the initial bounding box by analyzing the image gradient or edge information. By calculating the gradient information of the lumbar disc area, the algorithm can evaluate which areas are edges. Based on this edge information, adaptive adjustments are implemented to make the boundaries of the bounding box fit the target area more closely. This adjustment strategy ensures that the bounding box not only covers the entire area of ​​the target, but also accurately follows the complex shape of the target. The process of dynamically adjusting the bounding box is the key to ensuring segmentation accuracy. Through real-time fine-tuning, errors caused by shape changes can be effectively eliminated, so that the bounding box tightly surrounds the target area. This adaptive method can flexibly respond to the morphological changes of the lumbar disc in different slices, ensure the accuracy of the subsequent segmentation boundaries, and provide a basis for forming a high-quality three-dimensional model.

[0056] Before starting to adaptively adjust the bounding box, we first need to calculate the gradient information of the lumbar disc area in the image. By using gradient calculation methods such as the Sobel operator or the Scharr operator, we can extract the edge part with the most dramatic changes in the image. After obtaining the gradient information, we can combine it with the edge detection results of the image to identify the actual boundary of the lumbar disc. With this information, we can set the dynamic update strategy of the bounding box.

[0057] Based on the edge information obtained, an adaptive adjustment strategy is implemented. First, for each face of the bounding box, the distance from the edge is determined, and the boundary of the bounding box is adjusted in real time based on this distance. For example, if the edge of one side is too far from the boundary of the bounding box, the edge can be pulled to the real boundary by translation or extension. At the same time, the weighted average method can also be used for smooth adjustment to reduce excessive oscillation or instability during the adjustment process, ensuring the smoothness and effectiveness of the adjustment process.

[0058] During the real-time update process, a feedback mechanism is used to continuously monitor the adaptability of the bounding box. After each adjustment, the similarity and distance between the bounding box and the actual boundary of the lumbar disc are calculated, and then the next adjustment strategy is determined based on the monitoring results. Through multiple iterative updates, the bounding box is gradually optimized to fit closely to the edge of the target area. This adaptive adjustment scheme will make the final segmentation result more accurate and provide a reliable foundation for the subsequent segmentation process.

[0059] A nonlinear constrained iterative function in set theory is applied to optimize the boundary of the bounding box. The nonlinear constrained iterative function is based on multiple iterations. In each iteration, the bounding box is fine-tuned by calculating the similarity between the current bounding box and the target area pixels, so as to gradually approach the segmentation boundary, so as to obtain the target segmentation area based on the segmentation boundary segmentation. In each iteration, an energy minimization method is used to ensure that the bounding box will not be excessively disturbed by noise points during the convergence process. The energy minimization is achieved by a graph cut method or a level set method. In this step, the boundary of the bounding box is optimized through the nonlinear constrained iterative function in set theory. The algorithm transforms the conventional image segmentation problem into an energy minimization problem, and calculates the similarity between the pixels of the current bounding box and the target area through multiple iterations, so as to accurately adjust the bounding box. In each iteration, the boundary of the bounding box is gradually optimized according to the similarity information of the pixels to achieve the ideal segmentation effect. The implementation of this process helps to further improve the accuracy of segmentation, ensure that the bounding box generated in the end can effectively fit the target area, and minimize the impact of noise on the segmentation results. Through this iterative optimization, the true morphology of the lumbar intervertebral disc can be captured more efficiently, providing high-quality data support for subsequent three-dimensional modeling.

[0060] First, an initial target energy function is set, which is used to measure the similarity between the bounding box and the target area of ​​the lumbar disc. It can be set as the sum of squares of the difference between the pixel intensity in the bounding box and the pixel intensity in the target area. In this way, the relationship between the bounding box and the target area can be quantified and provide a basis for subsequent iterations.

[0061] In each iteration, the effect of the current bounding box is evaluated by calculating the value of the energy function. Subsequently, the iterative optimization algorithm is applied to fine-tune the bounding box according to the principle of energy minimization. For example, the graph cut method can be used to construct a minimum cut graph to distinguish the pixels inside and outside the bounding box, so as to accurately capture the target boundary. After each iteration, the change in the energy function value is checked to determine the adjustment direction and magnitude to ensure that the bounding box gradually approaches the true segmentation boundary.

[0062] In order to avoid the interference of noise on the segmentation results, it is necessary to introduce smoothness constraints in the energy minimization process. The drastic changes of the bounding box boundary can be suppressed by setting the smoothness term to ensure that the bounding box will not be disturbed by noise points during the stabilization process. Finally, after multiple rounds of iterations, the boundary of the bounding box will converge to an accurate segmentation result, providing a high-quality model for subsequent steps.

[0063] In the process of bounding box adjustment, the morphological continuity constraint between adjacent layers is further combined. The morphological continuity constraint includes analyzing the morphological and positional changes of the lumbar intervertebral disc in adjacent slices, and predicting the possible position and shape in the current slice through historical information. The morphological continuity constraint is used to smooth the segmentation result so that the segmented lumbar intervertebral disc morphology is continuous in three-dimensional space without unreasonable mutations. In the process of adjusting the bounding box, it is necessary to combine the morphological continuity constraints of adjacent slices to optimize the segmentation results. By analyzing the morphological and positional changes of the lumbar intervertebral disc in adjacent slices, historical information can be used to predict the possible shape and position of the lumbar intervertebral disc in the current slice. This continuity constraint makes the segmentation results appear coherent in three-dimensional space and eliminates unreasonable breaks or mutations. By introducing morphological continuity constraints, the segmentation quality of the lumbar intervertebral disc can be effectively improved, ensuring that the generated three-dimensional model is biologically reasonable and continuous. This method not only improves the realism of the model, but also provides a more reliable data basis for subsequent research and analysis.

[0064] During the segmentation process, multiple discrimination criteria are applied to filter out non-target areas, including boundary clarity, tissue density and expected anatomical position, and each segmented two-dimensional projection slice is integrated into a three-dimensional voxel model of the lumbar intervertebral disc of a complete target segmentation area to reflect the true three-dimensional morphology of the lumbar intervertebral disc.

[0065] During the segmentation process, multiple discrimination criteria are applied to filter out non-target areas. These criteria include detection of boundary clarity, analysis of tissue density, and rationality of its expected anatomical position. Through layer-by-layer screening, it is ensured that the final segmented area only includes the real lumbar disc structure, while also effectively excluding possible mislabeled areas. Finally, each screened two-dimensional projection slice is integrated to form a complete three-dimensional voxel model of the lumbar disc. This discrimination method helps to improve the accuracy of segmentation, so that the final generated three-dimensional voxel model better reflects the real lumbar disc morphology. This will provide an important basis for subsequent medical image analysis, diagnosis, and treatment, and enhance the scientific nature and clinical value of image recognition.

[0066] When performing segmentation, the segmentation results are evaluated by setting the discrimination criteria. First, the boundary clarity can be judged by calculating the gradient strength of the pixels in the bounding box. The clearer the boundary, the more reliable the segmentation result. Therefore, a threshold can be set. Only when the boundary clarity exceeds the threshold can it be considered as a valid segmentation area.

[0067] Secondly, the determination of tissue density is achieved by analyzing the CT values ​​in the target area. Since normal lumbar intervertebral discs have a specific density range in CT images, the determination criteria for tissue density can be set based on this information. When the tissue density is higher or lower than the preset standard, it can be considered as a potential non-target area. In this way, the accuracy of segmentation can be further improved through density analysis, and the wrongly marked areas can be effectively excluded.

[0068] Finally, combined with the expected information of the anatomical position, each segmented area is evaluated to see if it conforms to the anatomical characteristics. If the position or shape of a segmented area deviates significantly from the normal anatomical structure, it can be marked as a non-target area for removal. After multiple judgments, the remaining valid two-dimensional slices will be integrated to form a three-dimensional voxel model of the lumbar intervertebral disc. This process ensures that the final generated model truly reflects the anatomical characteristics of the lumbar intervertebral disc and provides high-quality data support for subsequent analysis.

[0069] S203, for the lumbar intervertebral disc three-dimensional voxel model, using a fiber enhanced tensor network to extract the texture features of the internal microstructure of the lumbar intervertebral disc, and output a high-dimensional lumbar texture feature vector; The fiber-enhanced tensor network (FatNet) uses a multi-layer convolution structure to capture features of different scales in the image, especially using tensor modules to analyze the microstructure of the lumbar intervertebral disc. This process enables the network to extract complex texture features including fiber bundles and cell structures by performing multi-scale convolution operations on the three-dimensional voxel model. These features can effectively reflect the biological structure and pathological changes of the lumbar intervertebral disc.

[0070] The significance of this step is that by extracting high-dimensional lumbar texture feature vectors, more abundant and detailed information can be provided for subsequent classification models. Texture features are important indicators for characterizing details and structures in images, and can reveal the health status and potential lesions of the lumbar disc, such as herniation or degeneration. Therefore, accurately extracting and modeling the texture features of these fine structures will significantly improve the accuracy of lumbar disc herniation identification, provide an important basis for clinical diagnosis, and ultimately help formulate personalized treatment plans.

[0071] Specifically, the multi-layer convolution structure and tensor module of the fiber-reinforced tensor network FatNet can be used to perform multi-scale convolution operations on the 3D voxel model data of the lumbar intervertebral disc. During the multi-scale convolution operation, the fiber-reinforced tensor network can extract texture features from the macroscopic to the microscopic level. This step uses a fiber-enhanced tensor network with a multi-layer convolutional structure to analyze the 3D voxel model of the lumbar disc. The network is specially designed with tensor modules to extract detailed information from the model at different levels and scales. Through multi-scale convolution operations, the network can capture various microstructures inside the lumbar disc, including fiber arrangement, tissue density, etc., thereby forming a deep understanding of the image.

[0072] The significance of this step is that it allows the model to obtain richer feature expressions and enhances the perception of the microstructure of the lumbar disc. Through multi-scale feature extraction, clinically important anatomical features can be better captured, thereby improving the accuracy of subsequent classification and recognition. This process helps to achieve higher diagnostic efficiency in image analysis, ensuring that doctors can quickly identify lumbar disc lesions for timely intervention and treatment.

[0073] In this step, we first need to build and initialize the fiber-reinforced tensor network (FatNet). The network adopts a multi-layer convolutional structure, which usually includes multiple convolutional layers, activation layers, and normalization layers. The three-dimensional voxel model data of the lumbar intervertebral disc is input into the network, and features are extracted through continuous convolution operations. In each layer of convolution, the network filters the input data and identifies feature information of different scales and levels.

[0074] During this process, the network uses tensor modules to enhance its ability to capture the fine structure inside the lumbar disc. The tensor module can not only effectively process multidimensional data, but also maintain sensitivity to details while improving feature extraction efficiency. Through multiple convolution operations, the network will go deeper layer by layer and gradually obtain multiple features from macro to micro. This process not only improves the model's ability to fully mine image information, but also ensures the retention of important features.

[0075] Finally, the extracted features are integrated and mapped through the fully connected layer to form a set of high-dimensional feature vectors. These feature vectors will contain texture information from different scales, providing sufficient basic data for subsequent feature analysis and classification. Through in-depth analysis of three-dimensional voxel data, the FatNet network ensures a panoramic understanding of the internal structure of the lumbar disc, thus providing a scientific basis for diagnosis and treatment.

[0076] The extracted texture features are subjected to high-dimensional modeling to transform the texture features into high-dimensional lumbar texture feature vectors with biological significance to reflect the internal structure and texture of the lumbar intervertebral disc.

[0077] After extracting the texture features of the lumbar disc, this step will further perform high-dimensional modeling on these features in order to convert them into more biologically meaningful feature vectors. This modeling process usually uses dimensionality reduction techniques such as principal component analysis (PCA) and t-SNE to compress a large number of high-dimensional features into a relatively small feature vector while maintaining the integrity of the original information as much as possible. Through this process, a representative high-dimensional lumbar texture feature vector can be formed, which can be used in subsequent model learning.

[0078] The significance of this step is to improve the interpretability and efficient utilization of features. High-dimensional modeling not only reduces the computational complexity of subsequent processing, but also ensures that the model can focus on the features with the most biological explanatory power. This provides a solid foundation for the final classification decision, ensuring that the classification results can accurately reflect the actual condition of the lumbar disc, thereby improving the reliability of clinical diagnosis.

[0079] After the texture features are extracted, this step will perform high-dimensional modeling on these features. First, all feature information obtained through multi-scale convolution operations is collected. At this point, these features can be numerically integrated with the help of data processing tools to form a unified data set for subsequent processing.

[0080] Next, dimensionality reduction techniques such as principal component analysis (PCA) are used to process high-dimensional features, and the main components are determined by analyzing the correlation between features. When performing PCA, special attention should be paid to selecting an appropriate variance explanation ratio to ensure that the results after dimensionality reduction can accurately reflect the characteristics of the original data. In this process, the covariance matrix of the features is calculated and sorted according to the size of the eigenvalues ​​to extract the main components and generate a reduced eigenvector. This process not only helps reduce the dimensionality of the feature data, but also improves the training efficiency of subsequent models while retaining important information.

[0081] Finally, the high-dimensional lumbar texture feature vector will be incorporated into the subsequent recognition and classification process. Through collection and modeling, the feature vector can not only describe the subtle texture of the lumbar disc, but also provide support for understanding its biological significance. The realization of this process ensures that researchers and clinicians can make faster and more accurate judgments on the health status of the lumbar disc.

[0082] S204, based on the high-dimensional lumbar texture feature vector and in combination with the anatomical appearance features of the lumbar intervertebral disc in different positions and multi-dimensional projection slices, a multi-dimensional fusion label classification model is constructed to output the recognition result of whether the lumbar intervertebral disc is normal or protruding.

[0083] One of the key steps of this method is to construct a multidimensional fusion label classification model based on the high-dimensional lumbar texture feature vector and the anatomical appearance features of the lumbar intervertebral disc in different positions and multidimensional projection slices. In this process, it is first necessary to comprehensively analyze the lumbar texture features extracted by multi-scale convolution and the anatomical feature data at different projection angles. These features include not only the morphology of the lumbar intervertebral disc, but also its spatial distribution relationship in the voxel model. Then, a deep learning algorithm is used to train a classification model that integrates multiple features to identify the healthy state or protrusion of the lumbar intervertebral disc.

[0084] The significance of building a multi-dimensional fusion label classification model is to improve the accuracy and efficiency of the diagnosis of lumbar disc status. By integrating multiple feature data, this model can capture the complexity and diversity of lumbar discs, so that when processing new CT image data, it can effectively distinguish between normal and protruding lumbar discs. This not only provides a scientific basis for clinical diagnosis, but also provides data support for subsequent treatment decisions, which helps to improve the treatment effect and quality of life of patients.

[0085] Specifically, the anatomical appearance features of the lumbar intervertebral disc in different positions and multi-dimensional projection slices can be combined to select a deep learning model for training to obtain a multi-dimensional fusion label classification model; In this step, researchers first need to collect comprehensive data related to lumbar intervertebral discs, including their anatomical features at different locations and multi-dimensional projection slice images. These data will provide a basis for the training of deep learning models. Then, select a suitable deep learning framework, such as a convolutional neural network (CNN), and design a network architecture that can effectively integrate feature data from different sources. By iteratively training using labeled data sets, the model will gradually learn the mapping relationship between the morphological characteristics of various types of lumbar intervertebral discs and the corresponding pathological conditions.

[0086] The purpose of this training process is to obtain an efficient multi-dimensional fusion label classification model that can quickly and accurately identify the status of the lumbar disc when processing new data. Through the training of the deep learning model, it can have good generalization ability, which helps to improve the accuracy of clinical diagnosis, reduce the risk of misdiagnosis and missed diagnosis, and provide patients with better medical services.

[0087] The first step of this step is to collect a large amount of lumbar disc imaging data to ensure that these data cover the normal and protruding states of various lumbar discs. These data should include CT scan results at different angles and positions to ensure sample diversity and representativeness. Each piece of data needs to be accurately labeled, for example, to clearly indicate whether it is a normal or protruding category. In order to enhance the generalization ability of the model, the dataset should contain samples from different groups (such as age, gender, weight, etc.) to ensure that the model can adapt to various patient backgrounds in practical applications.

[0088] Next, the collected imaging data is preprocessed. This includes steps such as denoising, normalizing, and enhancing the images to improve data quality. For example, histogram equalization can be used to enhance the contrast of the image to make the structure of the lumbar disc more obvious. Data augmentation techniques can also be used to expand the training set through methods such as rotation, scaling, and local cropping to prevent model overfitting. At this stage, these processing steps ensure the consistency and reliability of the data, laying the foundation for subsequent deep learning model training.

[0089] Then, the deep learning model architecture is designed and selected for training. Taking the convolutional neural network (CNN) as an example, multiple layers of convolution and pooling layers are constructed to extract high-dimensional features from the image. During the training process, the labeled high-dimensional lumbar texture feature vector set is used for learning, and the classification accuracy is improved by minimizing the loss function and optimizing the network parameters. The cross-validation technique is used to evaluate the model performance and adjust the hyperparameters. Finally, through high-quality training, the model can effectively recognize high-dimensional features, thereby forming a multi-dimensional fusion label classification model that can accurately classify based on the experience of historical data.

[0090] According to the high-dimensional lumbar texture feature vector, the trained multi-dimensional fusion label classification model is used to output the lumbar intervertebral disc recognition results corresponding to the CT image data, wherein the recognition results include: normal and protruding.

[0091] In this step, the multi-dimensional fusion label classification model trained in the previous step is used to analyze the new CT image data. When the new image data is input into the model, the model will calculate the corresponding feature vector through forward propagation and classify the state of the lumbar intervertebral disc according to the existing learning results. The recognition result will indicate whether the input lumbar intervertebral disc is normal or protruding, thus providing data support for subsequent clinical judgment and treatment decisions.

[0092] The implementation of this step is directly related to the accuracy and timeliness of clinical decision-making. Through the automated recognition of the model, doctors can quickly obtain a health status assessment of the lumbar disc, and then adopt corresponding treatment plans based on the recognition results, reducing the patient's waiting time for medical treatment and improving medical efficiency, ultimately improving the overall quality of medical care.

[0093] In this step, the CT image data to be identified must first be preprocessed to ensure that it conforms to the format required by the training model. Preprocessing includes image standardization, denoising, and contrast adjustment to make the quality of the input data consistent with the training data, thereby improving the accuracy and robustness of model recognition. The processed CT image will be converted into a high-dimensional lumbar texture feature vector. This step usually includes using the feature extraction method determined during the training process, such as extracting relevant features through the CNN network trained above.

[0094] Next, the generated high-dimensional lumbar texture feature vector is input into the trained multi-dimensional fusion label classification model. The model will infer based on these features and output the recognition results. In this process, the model will convert the feature vector into a classification result through the forward propagation algorithm, which can reflect the health status of the lumbar intervertebral disc. The model's analysis of the input features will cover multiple layers of calculations, and efficient classification will be achieved by extracting features layer by layer, so as to determine whether the lumbar intervertebral disc is normal or protruding.

[0095] Finally, the output will include a label indicating whether the lumbar disc is normal or protruding. This result can also be accompanied by a confidence score to provide a reference for the reliability of the classification. The efficiency of the processing enables doctors to quickly obtain recognition results, thereby speeding up the formulation of clinical diagnosis and treatment plans, and ultimately improving patient health management. This automated recognition process not only improves diagnostic efficiency, but also reduces human errors, which is of great significance to improving patient treatment outcomes.

[0096] It can be seen that the input CT image data is processed by multi-dimensional projection slicing to obtain a multi-dimensional feature projection slicing image data set; based on the multi-dimensional feature projection slicing image data set, an adaptive extended bounding box model is constructed, and the segmentation boundary of the lumbar intervertebral disc is automatically generated by using a nonlinear constrained iterative function to obtain a three-dimensional voxel model of the lumbar intervertebral disc in the target segmentation area; for the three-dimensional voxel model of the lumbar intervertebral disc, a fiber-reinforced tensor network is used to output a high-dimensional lumbar texture feature vector; based on the high-dimensional lumbar texture feature vector, combined with the anatomical appearance features of the lumbar intervertebral disc in different positions and multi-dimensional projection slices, a multi-dimensional fusion label classification model is constructed to output the recognition result of normal or protruding lumbar intervertebral disc, so as to combine modern image processing and deep learning technology to effectively improve the recognition rate and accuracy of lumbar disc protrusion.

[0097] Another embodiment of the present invention provides a lumbar disc herniation recognition system based on CT images, see Figure 3 , the system may include: The projection module 301 is used to perform multi-dimensional projection slicing processing on the input CT image data according to the anatomical structure characteristics of the lumbar intervertebral disc, determine the two-dimensional cross-sectional images of multiple key projection angles by projection, extract the high contrast features of the lumbar intervertebral disc area from different key projection angles, and combine the nonlinear filtering algorithm to suppress noise and artifact interference while retaining edge details, so as to obtain an optimized multi-dimensional feature projection slice image data set; The segmentation module 302 is used to construct an adaptive extended bounding box model based on the multi-dimensional feature projection slice image data set, take the area where the lumbar intervertebral disc is located as the target, dynamically adjust the size and shape of the segmentation area, and use the nonlinear constraint iteration function in set theory to automatically generate the segmentation boundary of the lumbar intervertebral disc, so as to obtain the target segmentation area based on the segmentation boundary segmentation, wherein, in the segmentation process, the intervertebral disc morphology continuity constraint calculated between adjacent layers is combined to further exclude the protruding non-target area, so as to obtain the lumbar intervertebral disc three-dimensional voxel model of the target segmentation area; Extraction module 303, for extracting texture features of the internal microstructure of the lumbar intervertebral disc using a fiber-enhanced tensor network for the three-dimensional voxel model of the lumbar intervertebral disc, and outputting a high-dimensional lumbar texture feature vector; The output module 304 is used to construct a multi-dimensional fusion label classification model based on the high-dimensional lumbar texture feature vector and the anatomical appearance features of the lumbar intervertebral disc in different positions and multi-dimensional projection slices to output the recognition result of whether the lumbar intervertebral disc is normal or protruded.

[0098] It can be seen that the input CT image data is processed by multi-dimensional projection slicing to obtain a multi-dimensional feature projection slicing image data set; based on the multi-dimensional feature projection slicing image data set, an adaptive extended bounding box model is constructed, and the segmentation boundary of the lumbar intervertebral disc is automatically generated by using a nonlinear constrained iterative function to obtain a three-dimensional voxel model of the lumbar intervertebral disc in the target segmentation area; for the three-dimensional voxel model of the lumbar intervertebral disc, a fiber-reinforced tensor network is used to output a high-dimensional lumbar texture feature vector; based on the high-dimensional lumbar texture feature vector, combined with the anatomical appearance features of the lumbar intervertebral disc in different positions and multi-dimensional projection slices, a multi-dimensional fusion label classification model is constructed to output the recognition result of normal or protruding lumbar intervertebral disc, so as to combine modern image processing and deep learning technology to effectively improve the recognition rate and accuracy of lumbar disc protrusion.

[0099] An embodiment of the present invention further provides a storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when running.

[0100] Specifically, in this embodiment, the above storage medium may be configured to store a computer program for performing the following steps: S201, according to the anatomical structure characteristics of the lumbar intervertebral disc, the input CT image data is subjected to multi-dimensional projection slicing processing, two-dimensional cross-sectional images of multiple key projection angles are determined by projection, high-contrast features of the lumbar intervertebral disc area are extracted from different key projection angles, and noise and artifact interference are suppressed by combining a nonlinear filtering algorithm while retaining edge details, thereby obtaining an optimized multi-dimensional feature projection slicing image data set; S202, constructing an adaptive extended bounding box model based on the multi-dimensional feature projection slice image data set, taking the area where the lumbar intervertebral disc is located as the target, dynamically adjusting the size and shape of the segmented area, and automatically generating the segmentation boundary of the lumbar intervertebral disc by using the nonlinear constraint iteration function in set theory, so as to obtain the target segmentation area based on the segmentation boundary segmentation, wherein, in the segmentation process, combining the intervertebral disc morphology continuity constraint calculated between adjacent layers, further excluding the protruding non-target area, and obtaining the lumbar intervertebral disc three-dimensional voxel model of the target segmentation area; S203, for the lumbar intervertebral disc three-dimensional voxel model, using a fiber enhanced tensor network to extract the texture features of the internal microstructure of the lumbar intervertebral disc, and output a high-dimensional lumbar texture feature vector; S204, based on the high-dimensional lumbar texture feature vector and in combination with the anatomical appearance features of the lumbar intervertebral disc in different positions and multi-dimensional projection slices, a multi-dimensional fusion label classification model is constructed to output the recognition result of whether the lumbar intervertebral disc is normal or protruding.

[0101] It can be seen that the input CT image data is processed by multi-dimensional projection slicing to obtain a multi-dimensional feature projection slicing image data set; based on the multi-dimensional feature projection slicing image data set, an adaptive extended bounding box model is constructed, and the segmentation boundary of the lumbar intervertebral disc is automatically generated by using a nonlinear constrained iterative function to obtain a three-dimensional voxel model of the lumbar intervertebral disc in the target segmentation area; for the three-dimensional voxel model of the lumbar intervertebral disc, a fiber-reinforced tensor network is used to output a high-dimensional lumbar texture feature vector; based on the high-dimensional lumbar texture feature vector, combined with the anatomical appearance features of the lumbar intervertebral disc in different positions and multi-dimensional projection slices, a multi-dimensional fusion label classification model is constructed to output the recognition result of normal or protruding lumbar intervertebral disc, so as to combine modern image processing and deep learning technology to effectively improve the recognition rate and accuracy of lumbar disc protrusion.

[0102] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0103] Specifically, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0104] Specifically, in this embodiment, the processor may be configured to perform the following steps through a computer program: S201, according to the anatomical structure characteristics of the lumbar intervertebral disc, the input CT image data is subjected to multi-dimensional projection slicing processing, two-dimensional cross-sectional images of multiple key projection angles are determined by projection, high-contrast features of the lumbar intervertebral disc area are extracted from different key projection angles, and noise and artifact interference are suppressed by combining a nonlinear filtering algorithm while retaining edge details, thereby obtaining an optimized multi-dimensional feature projection slicing image data set; S202, constructing an adaptive extended bounding box model based on the multi-dimensional feature projection slice image data set, taking the area where the lumbar intervertebral disc is located as the target, dynamically adjusting the size and shape of the segmented area, and automatically generating the segmentation boundary of the lumbar intervertebral disc by using the nonlinear constraint iteration function in set theory, so as to obtain the target segmentation area based on the segmentation boundary segmentation, wherein, in the segmentation process, combining the intervertebral disc morphology continuity constraint calculated between adjacent layers, further excluding the protruding non-target area, and obtaining the lumbar intervertebral disc three-dimensional voxel model of the target segmentation area; S203, for the lumbar intervertebral disc three-dimensional voxel model, using a fiber enhanced tensor network to extract the texture features of the internal microstructure of the lumbar intervertebral disc, and output a high-dimensional lumbar texture feature vector; S204, based on the high-dimensional lumbar texture feature vector and in combination with the anatomical appearance features of the lumbar intervertebral disc in different positions and multi-dimensional projection slices, a multi-dimensional fusion label classification model is constructed to output the recognition result of whether the lumbar intervertebral disc is normal or protruding.

[0105] It can be seen that the input CT image data is processed by multi-dimensional projection slicing to obtain a multi-dimensional feature projection slicing image data set; based on the multi-dimensional feature projection slicing image data set, an adaptive extended bounding box model is constructed, and the segmentation boundary of the lumbar intervertebral disc is automatically generated by using a nonlinear constrained iterative function to obtain a three-dimensional voxel model of the lumbar intervertebral disc in the target segmentation area; for the three-dimensional voxel model of the lumbar intervertebral disc, a fiber-reinforced tensor network is used to output a high-dimensional lumbar texture feature vector; based on the high-dimensional lumbar texture feature vector, combined with the anatomical appearance features of the lumbar intervertebral disc in different positions and multi-dimensional projection slices, a multi-dimensional fusion label classification model is constructed to output the recognition result of normal or protruding lumbar intervertebral disc, so as to combine modern image processing and deep learning technology to effectively improve the recognition rate and accuracy of lumbar disc protrusion.

[0106] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings. The above is only a preferred embodiment of the present invention, but the present invention is not limited to the scope of implementation shown in the drawings. Any changes made according to the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which still do not exceed the spirit covered by the description and drawings, should be within the protection scope of the present invention.

Claims

1. A method for identifying lumbar disc herniation based on CT images, characterized in that: The method comprises: According to the anatomical characteristics of the lumbar intervertebral disc, the input CT image data is processed by multi-dimensional projection slicing. The two-dimensional cross-sectional images of multiple key projection angles are determined by projection. The high-contrast features of the lumbar intervertebral disc area are extracted from different key projection angles. The nonlinear filtering algorithm is combined to suppress noise and artifact interference while retaining edge details to obtain an optimized multi-dimensional feature projection slice image data set. According to the multi-dimensional feature projection slice image data set, an adaptive extended bounding box model is constructed. The area where the lumbar intervertebral disc is located is taken as the target, and the size and shape of the segmentation area are dynamically adjusted. The segmentation boundary of the lumbar intervertebral disc is automatically generated by using the nonlinear constraint iteration function in set theory, so as to obtain the target segmentation area based on the segmentation boundary segmentation. In the segmentation process, the intervertebral disc morphology continuity constraint calculated between adjacent layers is combined to further exclude the protruding non-target area, and obtain the lumbar intervertebral disc three-dimensional voxel model of the target segmentation area. For the 3D voxel model of lumbar intervertebral disc, the fiber-enhanced tensor network is used to extract the texture features of the internal microstructure of the lumbar intervertebral disc and output a high-dimensional lumbar texture feature vector. According to the high-dimensional lumbar texture feature vector and combined with the anatomical appearance features of the lumbar intervertebral disc in different positions and multi-dimensional projection slices, a multi-dimensional fusion label classification model is constructed to output the recognition results of whether the lumbar intervertebral disc is normal or protruding.

2. The method according to claim 1, characterized in that According to the anatomical structure characteristics of the lumbar intervertebral disc, the input CT image data is subjected to multi-dimensional projection slicing processing, and two-dimensional cross-sectional images of multiple key projection angles are determined by projection, and high-contrast features of the lumbar intervertebral disc area are extracted from different key projection angles, and noise and artifact interference are suppressed by combining a nonlinear filtering algorithm, while retaining edge details, to obtain an optimized multi-dimensional feature projection slice image data set, including: Receiving CT image data to be processed and ensuring that the quality of the image data meets certain standards, wherein the CT image is three-dimensional volume data and is composed of multiple two-dimensional slices; According to the anatomical characteristics of the lumbar intervertebral disc, the lumbar intervertebral disc area is preliminarily marked using a recognition algorithm to ensure that the area of ​​interest can be located in the three-dimensional volume data; The marked lumbar intervertebral disc region is subjected to multi-dimensional projection slicing processing, by selecting a plurality of key projection angles in the three-dimensional CT image data for projection, and each projection angle corresponds to generating a set of two-dimensional projection slice images as two-dimensional cross-sectional images, wherein the two-dimensional cross-sectional images can provide cross-sectional views of the lumbar intervertebral disc observed from different directions, and help identify anatomical structural features with important diagnostic value, and the two-dimensional cross-sectional images at least include sagittal plane, coronal plane, and transverse plane; In each set of two-dimensional projection slice images generated, adaptive contrast enhancement technology is used to dynamically enhance the contrast of the image based on local histogram equalization or contrast-limited adaptive histogram equalization to more clearly identify the edges and fine structures of the lumbar intervertebral disc and improve the prominence of specific anatomical structures; The contrast-enhanced two-dimensional projection slice image is filtered by a nonlinear filtering algorithm to remove noise and artifact interference in the image while retaining edge details in the image, wherein the nonlinear filtering algorithm is bilateral filtering or non-local mean filtering. The bilateral filtering can retain edge features during the denoising process, and the non-local mean filtering can retain details by considering similar pixel blocks; The filtered two-dimensional projection slice images are combined into a multi-dimensional feature projection slice image dataset, in which each slice provides high-contrast features extracted from different projection angles, and all slices together form a complete, multi-angle observation of lumbar intervertebral disc image feature set.

3. The method according to claim 2, characterized in that The method constructs an adaptive extended bounding box model based on a multi-dimensional feature projection slice image data set, takes the area where the lumbar intervertebral disc is located as a target, dynamically adjusts the size and shape of the segmented area, and automatically generates the segmentation boundary of the lumbar intervertebral disc by using a nonlinear constraint iteration function in set theory, so as to obtain a target segmentation area based on the segmentation boundary segmentation. In the segmentation process, the intervertebral disc morphology continuity constraint calculated between adjacent layers is combined to further exclude the protruding non-target area, so as to obtain a lumbar intervertebral disc three-dimensional voxel model of the target segmentation area, including: Preliminarily extracting the position of the lumbar intervertebral disc from the multi-dimensional feature projection slice image data set, and enclosing the lumbar intervertebral disc with an initial bounding box, wherein the initial bounding box is determined based on prior information or obtained through morphological operations, wherein the prior information includes an average lumbar intervertebral disc size, and the morphological operations include edge detection of the image; Adopting an adaptive adjustment strategy based on image gradient or edge information, the bounding box is dynamically adjusted according to the specific shape and size of the lumbar intervertebral disc, and each face and corner of the bounding box is fine-tuned in real time to make the bounding box close to the edge of the target area; A nonlinear constrained iterative function in set theory is applied to optimize the boundary of the bounding box. The nonlinear constrained iterative function is based on multiple iterations. In each iteration, the bounding box is fine-tuned by calculating the similarity between the current bounding box and the target area pixels, so as to gradually approach the segmentation boundary, so as to obtain the target segmentation area based on the segmentation boundary segmentation. In each iteration, an energy minimization method is used to ensure that the bounding box will not be excessively disturbed by noise points during the convergence process. The energy minimization is achieved by a graph cut method or a level set method. In the process of bounding box adjustment, the morphological continuity constraint between adjacent layers is further combined. The morphological continuity constraint includes analyzing the morphological and positional changes of the lumbar intervertebral disc in adjacent slices, and predicting the possible position and shape in the current slice through historical information. The morphological continuity constraint is used to smooth the segmentation result so that the segmented lumbar intervertebral disc morphology is continuous in three-dimensional space without unreasonable mutations. During the segmentation process, multiple discrimination criteria are applied to filter out non-target areas, including boundary clarity, tissue density and expected anatomical position, and each segmented two-dimensional projection slice is integrated into a three-dimensional voxel model of the lumbar intervertebral disc of a complete target segmentation area to reflect the true three-dimensional morphology of the lumbar intervertebral disc.

4. The method according to claim 3, characterized in that The method uses a fiber-enhanced tensor network to extract the texture features of the internal microstructure of the lumbar intervertebral disc for the three-dimensional voxel model of the lumbar intervertebral disc and output a high-dimensional lumbar texture feature vector, including: The multi-layer convolution structure and tensor module of the fiber-reinforced tensor network FatNet are used to perform multi-scale convolution operations on the 3D voxel model data of the lumbar intervertebral disc. During the multi-scale convolution operation, the fiber-reinforced tensor network can extract texture features from the macroscopic to the microscopic level. The extracted texture features are subjected to high-dimensional modeling to transform the texture features into high-dimensional lumbar texture feature vectors with biological significance to reflect the internal structure and texture of the lumbar intervertebral disc.

5. The method according to claim 4, characterized in that The multi-dimensional fusion label classification model is constructed based on the high-dimensional lumbar texture feature vector and the anatomical appearance features of the lumbar intervertebral disc in different positions and multi-dimensional projection slices to output the recognition results of whether the lumbar intervertebral disc is normal or protruding, including: Combined with the anatomical features of the lumbar intervertebral disc in different positions and multi-dimensional projection slices, a deep learning model was selected for training to obtain a multi-dimensional fusion label classification model; According to the high-dimensional lumbar texture feature vector, the trained multi-dimensional fusion label classification model is used to output the lumbar intervertebral disc recognition results corresponding to the CT image data, wherein the recognition results include: normal and protruding.

6. A lumbar disc herniation recognition system based on CT images, characterized in that: The system comprises: The projection module is used to perform multi-dimensional projection slicing processing on the input CT image data according to the anatomical structure characteristics of the lumbar intervertebral disc, determine the two-dimensional cross-sectional images of multiple key projection angles by projection, extract the high-contrast features of the lumbar intervertebral disc area from different key projection angles, and combine the nonlinear filtering algorithm to suppress noise and artifact interference while retaining edge details to obtain an optimized multi-dimensional feature projection slice image data set; The segmentation module is used to construct an adaptive extended bounding box model based on a multi-dimensional feature projection slice image data set, take the area where the lumbar intervertebral disc is located as the target, dynamically adjust the size and shape of the segmentation area, and use the nonlinear constraint iteration function in set theory to automatically generate the segmentation boundary of the lumbar intervertebral disc, so as to obtain the target segmentation area based on the segmentation boundary segmentation. In the segmentation process, the intervertebral disc morphology continuity constraint calculated between adjacent layers is combined to further exclude the protruding non-target area, so as to obtain a 3D voxel model of the lumbar intervertebral disc in the target segmentation area; The extraction module is used to extract the texture features of the internal microstructure of the lumbar intervertebral disc using a fiber-enhanced tensor network for the three-dimensional voxel model of the lumbar intervertebral disc, and output a high-dimensional lumbar texture feature vector; The output module is used to construct a multi-dimensional fusion label classification model based on the high-dimensional lumbar texture feature vector and the anatomical appearance features of the lumbar intervertebral disc in different positions and multi-dimensional projection slices to output the recognition results of whether the lumbar intervertebral disc is normal or protruding.

7. The system according to claim 6, characterized in that The projection module is specifically used for: Receiving CT image data to be processed and ensuring that the quality of the image data meets certain standards, wherein the CT image is three-dimensional volume data and is composed of multiple two-dimensional slices; According to the anatomical characteristics of the lumbar intervertebral disc, the lumbar intervertebral disc area is preliminarily marked using a recognition algorithm to ensure that the area of ​​interest can be located in the three-dimensional volume data; The marked lumbar intervertebral disc region is subjected to multi-dimensional projection slicing processing, by selecting a plurality of key projection angles in the three-dimensional CT image data for projection, and each projection angle corresponds to generating a set of two-dimensional projection slice images as two-dimensional cross-sectional images, wherein the two-dimensional cross-sectional images can provide cross-sectional views of the lumbar intervertebral disc observed from different directions, and help identify anatomical structural features with important diagnostic value, and the two-dimensional cross-sectional images at least include sagittal plane, coronal plane, and transverse plane; In each set of two-dimensional projection slice images generated, adaptive contrast enhancement technology is used to dynamically enhance the contrast of the image based on local histogram equalization or contrast-limited adaptive histogram equalization to more clearly identify the edges and fine structures of the lumbar intervertebral disc and improve the prominence of specific anatomical structures; The contrast-enhanced two-dimensional projection slice image is filtered by a nonlinear filtering algorithm to remove noise and artifact interference in the image while retaining edge details in the image, wherein the nonlinear filtering algorithm is bilateral filtering or non-local mean filtering. The bilateral filtering can retain edge features during the denoising process, and the non-local mean filtering can retain details by considering similar pixel blocks; The filtered two-dimensional projection slice images are combined into a multi-dimensional feature projection slice image dataset, in which each slice provides high-contrast features extracted from different projection angles, and all slices together form a complete, multi-angle observation of lumbar intervertebral disc image feature set.

8. The system according to claim 7, characterized in that The segmentation module is specifically used for: Preliminarily extracting the position of the lumbar intervertebral disc from the multi-dimensional feature projection slice image data set, and enclosing the lumbar intervertebral disc with an initial bounding box, wherein the initial bounding box is determined based on prior information or obtained through morphological operations, wherein the prior information includes an average lumbar intervertebral disc size, and the morphological operations include edge detection of the image; Adopting an adaptive adjustment strategy based on image gradient or edge information, the bounding box is dynamically adjusted according to the specific shape and size of the lumbar intervertebral disc, and each face and corner of the bounding box is fine-tuned in real time to make the bounding box close to the edge of the target area; A nonlinear constrained iterative function in set theory is applied to optimize the boundary of the bounding box. The nonlinear constrained iterative function is based on multiple iterations. In each iteration, the bounding box is fine-tuned by calculating the similarity between the current bounding box and the target area pixels, so as to gradually approach the segmentation boundary, so as to obtain the target segmentation area based on the segmentation boundary segmentation. In each iteration, an energy minimization method is used to ensure that the bounding box will not be excessively disturbed by noise points during the convergence process. The energy minimization is achieved by a graph cut method or a level set method. In the process of bounding box adjustment, the morphological continuity constraint between adjacent layers is further combined. The morphological continuity constraint includes analyzing the morphological and positional changes of the lumbar intervertebral disc in adjacent slices, and predicting the possible position and shape in the current slice through historical information. The morphological continuity constraint is used to smooth the segmentation result so that the segmented lumbar intervertebral disc morphology is continuous in three-dimensional space without unreasonable mutations. During the segmentation process, multiple discrimination criteria are applied to filter out non-target areas, including boundary clarity, tissue density and expected anatomical position, and each segmented two-dimensional projection slice is integrated into a three-dimensional voxel model of the lumbar intervertebral disc of a complete target segmentation area to reflect the true three-dimensional morphology of the lumbar intervertebral disc.

9. A storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to execute the method according to any one of claims 1 to 5 when executed.

10. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to perform the method according to any one of claims 1 to 5.

Images