A CT image segmentation method for spinal surgery

Abstract

The present invention discloses a CT image segmentation method for spinal surgery, which relates to the technical field of medical image processing, and comprises the following steps: obtaining a spinal CT image of a patient, and preprocessing the spinal CT image; using a vertebral space detection method to detect the vertebral space area of ​​the spine in the preprocessed CT image, and automatically segmenting each vertebra based on the vertebral space area information to obtain a preliminary segmentation result of the vertebra; extracting local texture features of intervertebral discs and nerve roots in the vertebral space area, and classifying them through a classifier to generate preliminary segmentation results of the intervertebral discs and nerve roots; rigidly and non-rigidly registering the preliminary segmentation results of the vertebrae, intervertebral discs and nerve roots with a standard three-dimensional model, and identifying abnormal areas in the segmentation; and accurately segmenting a variety of structures by extracting local texture features of intervertebral discs and nerve roots and combining them with a support vector machine classifier.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a CT image segmentation method for spinal surgery. Background Art

[0002] In recent years, spinal CT images have been increasingly used in spinal surgery, especially in preoperative planning and intraoperative navigation. CT image segmentation technology provides doctors with accurate anatomical information. However, traditional spinal CT image segmentation methods mostly rely on manual or semi-automatic segmentation techniques, which are cumbersome to operate and rely on expert experience for accuracy, making it difficult to meet the segmentation requirements of complex lesion areas. In addition, existing automatic segmentation methods usually only target a single anatomical structure, such as a vertebra or intervertebral disc, and it is difficult to simultaneously segment multiple important structures such as vertebral bodies, intervertebral discs, and nerve roots, and the processing of abnormal areas is insufficient.

[0003] The shortcomings of the existing technology mainly lie in the problem of balancing segmentation accuracy and efficiency. On the one hand, when dealing with complex spinal anatomical structures, traditional segmentation algorithms are easily affected by noise, grayscale unevenness and morphological changes, resulting in inaccurate segmentation results; on the other hand, existing automatic segmentation methods usually lack an effective correction mechanism for abnormal shape areas and are unable to handle abnormal shape areas caused by lesions or surgical requirements, resulting in insufficient reliability of segmentation results in clinical applications. The present invention overcomes these shortcomings and significantly improves segmentation accuracy through comprehensive image preprocessing, automatic segmentation and shape correction technology. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a CT image segmentation method for spinal surgery to solve the problem of automatic segmentation of vertebral bodies, intervertebral discs and nerve roots in spinal CT images and correct abnormal shape areas.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a CT image segmentation method for spinal surgery, which includes obtaining a spinal CT image of a patient and preprocessing the spinal CT image;

[0008] The intervertebral space detection method is used to detect the intervertebral space area of ​​the spine in the preprocessed CT image, and the upper and lower boundaries of the vertebra are automatically determined by analyzing the gray gradient changes and morphological characteristics between the vertebrae.

[0009] Each vertebra is automatically segmented based on the intervertebral space area information to obtain the preliminary segmentation result of the vertebra;

[0010] Extract local texture features of the intervertebral disc and nerve root in the intervertebral space area, and classify them through a classifier to generate preliminary segmentation results of the intervertebral disc and nerve root;

[0011] The preliminary segmentation results of vertebral bodies, intervertebral discs and nerve roots are rigidly and non-rigidly registered with the standard 3D model to identify abnormal areas in the segmentation;

[0012] The shape of the abnormal segmentation area is corrected based on the shape prior information of the standard 3D model to generate a corrected segmentation result.

[0013] As a preferred solution of the CT image segmentation method for spinal surgery described in the present invention, the preprocessing includes denoising, grayscale normalization, local contrast enhancement and resampling processing.

[0014] As a preferred solution of the CT image segmentation method for spinal surgery of the present invention, wherein: the denoising refers to removing image noise by using bilateral filtering method to obtain the gray value of the filtered image;

[0015] Grayscale normalization refers to obtaining the grayscale value in the denoised image and performing grayscale normalization on each pixel in the image;

[0016] The local contrast enhancement refers to dividing the normalized image into sub-blocks of equal size, performing local histogram equalization on each sub-block, and performing bilinear interpolation to reassemble the enhanced image into a complete image;

[0017] The resampling refers to determining the slice thickness of the combined complete image, performing trilinear interpolation on the image, and adjusting the uneven slice thickness to a uniform thickness.

[0018] As a preferred embodiment of the CT image segmentation method for spinal surgery described in the present invention, the vertebral space detection method is used to detect the vertebral space area of ​​the spine in the preprocessed CT image, and the upper and lower boundaries of the vertebra are automatically determined by analyzing the gray gradient changes and morphological characteristics between the vertebrae, including the following steps:

[0019] The grayscale gradient of each pixel of the preprocessed CT image is calculated to generate a gradient amplitude image;

[0020] In the gradient amplitude image, find the area with obvious gradient changes between the upper and lower boundaries of the spine and the intervertebral disc, and mark the area with the most significant gradient change as the location of the intervertebral space;

[0021] For the detected intervertebral gap area, small noise points are removed and the broken gap area is connected;

[0022] The distance between the upper and lower borders of each vertebra was recorded by calculating the width of the gap area after connection;

[0023] For each vertebral space area, the gray value distribution of its internal pixels is analyzed to generate a gray histogram to further determine the upper and lower boundaries of the vertebra.

[0024] As a preferred embodiment of the CT image segmentation method for spinal surgery of the present invention, each vertebra is automatically segmented based on the vertebral space area information to obtain the preliminary segmentation result of the vertebral body, including the following steps:

[0025] Between the upper and lower boundaries of each vertebra, M seed points are selected for region growing;

[0026] According to the characteristics of the spine CT image, the grayscale similarity threshold is set;

[0027] For each seed point, a growth region is initialized, and pixels adjacent to the seed point and with grayscale value differences less than the grayscale similarity threshold are included in the growth region;

[0028] Starting from the boundary point of the current growing region, the neighborhood check and region update are repeated until the grayscale difference exceeds the grayscale similarity threshold, then the region growth is stopped to form the final vertebral segmentation region.

[0029] As a preferred embodiment of the CT image segmentation method for spinal surgery described in the present invention, the local texture features of the intervertebral disc and the nerve root are extracted in the intervertebral space area, and classified by a classifier to generate preliminary segmentation results of the intervertebral disc and the nerve root, including the following steps:

[0030] Based on the vertebral segmentation area, the upper limit boundary of each vertebra is determined, and the middle area between each pair of adjacent vertebrae is used as the target area of ​​the intervertebral disc;

[0031] The target area of ​​the nerve root is estimated by analyzing the upper and lower borders of the vertebral body and the target area of ​​the intervertebral disc;

[0032] Extracting gray-level co-occurrence matrix-based features from each target region;

[0033] For each pixel in the target area, the texture features of each target area are obtained by calculating the local binary pattern histogram;

[0034] Calculate the gradient direction histogram for each target area to obtain the gradient direction information of the local area;

[0035] Combine all the extracted features in sequence to form the final high-dimensional feature vector;

[0036] Select support vector machine as the classifier and train the support vector machine using the CT image dataset with known annotations;

[0037] The extracted high-dimensional feature vector is input into the trained support vector machine. According to the output of the classifier, the areas belonging to the intervertebral disc and nerve root in the test image are marked to generate the segmentation results of the intervertebral disc and nerve root.

[0038] As a preferred embodiment of the CT image segmentation method for spinal surgery described in the present invention, the rigid registration and non-rigid registration of the preliminary segmentation results of the vertebral body, intervertebral disc and nerve root with the standard three-dimensional model include the following steps:

[0039] 3D reconstruction is performed using the 2D segmentation data of the segmented vertebral body, intervertebral disc and nerve root to generate a 3D voxel model;

[0040] Select a standard 3D model similar to the patient's age, gender and body shape from the medical database and set rigid transformation conditions;

[0041] Based on the rigid transformation condition, the generated 3D voxel model and the standard 3D model are aligned;

[0042] The segmentation result and the segmentation model are rigidly registered in space by iterating the closest point algorithm.

[0043] Establish control point grids on the segmentation result after rigid registration and the standard three-dimensional model respectively;

[0044] Using free-shape deformation models, local adjustments are made to standard 3D models through a grid of control points;

[0045] The shape similarity measure between the segmentation result and the standard 3D model is defined to minimize the shape difference between the segmentation result and the standard 3D model, and the displacement vector of the control point is optimized by the gradient descent method to complete the non-rigid registration.

[0046] As a preferred solution of the CT image segmentation method for spinal surgery described in the present invention, wherein: identifying the abnormal shape segmentation area based on the shape prior information of the standard three-dimensional model and correcting it, generating the corrected segmentation result includes the following steps:

[0047] Perform shape analysis on the vertebrae of the segmentation results, identify the abnormal vertebral shape areas, as well as the areas of the boundary models of the intervertebral disc and the nerve root, and mark all detected abnormal shape areas;

[0048] For the detected abnormal vertebral shape areas, three-dimensional shape operations are used to correct them;

[0049] For the areas with blurred intervertebral disc boundaries, a boundary smoothing algorithm was used for correction;

[0050] For the nerve root region, the three-dimensional template matching technology was used to adjust the position of the nerve root region to be consistent with the position in the standard spine model;

[0051] Based on the shape correction, a corrected segmentation result is generated.

[0052] In a second aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the CT image segmentation method for spinal surgery as described in the first aspect of the present invention is implemented.

[0053] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the CT image segmentation method for spinal surgery as described in the first aspect of the present invention is implemented.

[0054] The beneficial effects of the present invention are as follows: image quality is improved through denoising, grayscale normalization and local contrast enhancement processing, which lays a foundation for subsequent vertebral space detection and segmentation and avoids the influence of noise and grayscale unevenness on segmentation accuracy; in the automatic segmentation link, the vertebral space detection method based on grayscale gradient and morphological features is used to automatically determine the upper and lower boundaries of the vertebral body, quickly obtain preliminary segmentation results, and greatly improve the segmentation efficiency; by extracting the local texture features of the intervertebral disc and nerve root and combining with the support vector machine classifier, accurate segmentation of various structures can be achieved; in addition, rigid and non-rigid registration technology is used, combined with the shape prior information of the standard three-dimensional model, the detected shape abnormal area is corrected to ensure the accuracy and consistency of the final segmentation result. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0056] Figure 1 is a flow chart of the CT image segmentation method for spinal surgery in Example 1;

[0057] Figure 2 This is a schematic diagram of the preliminary segmentation of the vertebral body in Example 1. DETAILED DESCRIPTION

[0058] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0059] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0060] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0061] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a CT image segmentation method for spinal surgery, comprising the following steps:

[0062] S1. Obtain a spinal CT image of the patient and preprocess the spinal CT image.

[0063] S1.1. Preprocessing includes denoising, grayscale normalization, local contrast enhancement and resampling.

[0064] S1.1.1. Denoising refers to the use of bilateral filtering to remove image noise and obtain the grayscale value of the filtered image; grayscale normalization refers to obtaining the grayscale value in the denoised image and normalizing the grayscale of each pixel in the image; local contrast enhancement refers to dividing the normalized image into sub-blocks of equal size, performing local histogram equalization on each sub-block, and performing bilinear interpolation to recombine the enhanced image into a complete image; resampling refers to determining the slice thickness of the combined complete image, performing trilinear interpolation on the image, and adjusting the uneven slice thickness to a uniform thickness.

[0065] Specifically, the operation formula of bilateral filtering is:

[0066] ;

[0067] in, Indicates the pixel position of the filtered image The gray value at Indicates the pixel position currently being processed, represents the pixel position in the neighborhood, Represents the neighborhood pixel position Gray value, indicating the original image at the pixel position The gray value at represents the normalization factor, which is the sum of all weights in the filter window, represents the pixel neighborhood of the filter window, Represents the standard deviation in the spatial domain, controls the spatial range of the filter, and is usually set to 1% to 2% of the image size. It represents the standard deviation of the grayscale domain, controls the similarity of grayscale values, and is usually set to 5% to 10% of the dynamic range of the image grayscale value;

[0068] The formula for grayscale normalization is:

[0069] ;

[0070] in, is the normalized image at pixel position The gray value at is the minimum gray value in the image, is the maximum gray value in the image.

[0071] S2, using the intervertebral space detection method to detect the intervertebral space area of ​​the spine in the preprocessed CT image, and automatically determining the upper and lower boundaries of the vertebra by analyzing the gray gradient changes and morphological characteristics between the vertebrae, including the following steps:

[0072] The grayscale gradient of each pixel in the preprocessed CT image was calculated to generate a gradient amplitude image. In the gradient amplitude image, the area with obvious gradient changes between the upper and lower boundaries of the spine and the intervertebral disc was found, and the area with the most significant gradient change was marked as the location of the vertebral space. For the detected vertebral space area, small noise points were removed and the broken space area was connected. The distance between the upper and lower boundaries of each vertebra was recorded by calculating the width of the connected space area. For each vertebral space area, the grayscale value distribution of the internal pixels was analyzed to generate a grayscale histogram to further determine the upper and lower boundaries of the vertebra.

[0073] Specifically, the gray gradient calculation formula is:

[0074] ;

[0075] in, Indicates pixel position The gradient amplitude at Indicates the pixel position of the CT image The gray value at represents partial derivative.

[0076] The calculation formula for the width of the gap area after connection is:

[0077] ;

[0078] in, Represents the average width of the entire intervertebral space. The number of columns representing the gap area, Indicates The width of the gap between columns, The column index representing the gap area.

[0079] The formula for the grayscale histogram is:

[0080] ;

[0081] in, Represents grayscale value The histogram frequency of Indicates the total number of pixels, Indicates the use for judgment Is it equal to The indicator function, Represents grayscale value;

[0082] Furthermore, the indicator function is defined as follows:

[0083] ;

[0084] if , then it returns 1, indicating that the gray value of the pixel is , otherwise return 0.

[0085] S3, automatically segmenting each vertebra based on the vertebral space area information, and obtaining the preliminary segmentation result of the vertebra includes the following steps:

[0086] Between the upper and lower boundaries of each vertebra, M seed points are selected for regional growing. According to the characteristics of spinal CT images, the grayscale similarity threshold is set. For each seed point, a growing region is initialized, and pixels adjacent to the seed point and with grayscale value differences less than the grayscale similarity threshold are included in the growing region. Starting from the boundary point of the current growing region, the neighborhood check and region update are repeated until the grayscale difference exceeds the grayscale similarity threshold, then the regional growing is stopped to form the final vertebral segmentation region.

[0087] Specifically, the grayscale similarity threshold is set to 25. Because in CT images, the grayscale value of bones is usually higher, while the grayscale values ​​of intervertebral discs and soft tissues are lower, a grayscale similarity threshold of 25 is selected to effectively distinguish the bone area and intervertebral disc area of ​​the vertebra, and avoid mistakenly including intervertebral discs with large grayscale value differences into the vertebral segmentation area; the grayscale value inside the vertebral body is relatively uniform, and a grayscale difference of up to 25 units is allowed to cover different parts of the vertebral body, but it will not extend to the grayscale range that does not belong to the bone area. A grayscale similarity threshold of less than 25 will cause the range of regional growth to be too limited, and may not be able to fully cover the entire vertebral body, resulting in incomplete segmentation. A grayscale similarity threshold greater than 25 will cause the growth area to be too large, and may mistakenly include areas with lower grayscale values ​​(such as intervertebral discs or soft tissues), resulting in inaccurate segmentation.

[0088] It should be noted that automated vertebral segmentation reduces the workload and time of manual segmentation and improves segmentation efficiency. Traditional manual segmentation methods rely on the experience of experts, are time-consuming and highly subjective. The region growing algorithm can generate preliminary vertebral segmentation results in a short time through automated processing, dynamically adjust the segmentation region according to the local grayscale similarity, and adapt to the spinal anatomical structure of different patients, which significantly improves the segmentation efficiency and is particularly suitable for processing large-scale medical imaging data sets; by gradually growing regions from seed points, it can effectively control the expansion of the segmented region and avoid over-segmentation.

[0089] S4, extracting local texture features of the intervertebral disc and nerve root in the intervertebral space area, and classifying them through a classifier to generate preliminary segmentation results of the intervertebral disc and nerve root includes the following steps:

[0090] Based on the vertebral segmentation area, the upper and lower boundaries of each vertebra are determined, and the middle area between each pair of adjacent vertebrae is used as the target area of ​​the intervertebral disc; the target area of ​​the nerve root is estimated by analyzing the upper and lower boundaries of the vertebral body and the target area of ​​the intervertebral disc; the features based on the grayscale co-occurrence matrix are extracted from each target area; for each pixel in the target area, the texture features of each target area are obtained by calculating the local binary pattern histogram; for each target area, the gradient direction histogram is calculated to obtain the gradient direction information of the local area; all the extracted features are combined in sequence to form the final high-dimensional feature vector; the support vector machine is selected as the classifier, and the support vector machine is trained using a known annotated CT image data set; the extracted high-dimensional feature vector is input into the trained support vector machine, and according to the output of the classifier, the areas belonging to the intervertebral disc and nerve root in the test image are marked to generate the segmentation results of the intervertebral disc and nerve root.

[0091] Specifically, the gray-level co-occurrence matrix features include energy, contrast, and entropy; energy reflects the uniformity of the image, contrast reflects the sharpness of the image edge, and entropy is used to measure the randomness of the image.

[0092] The local binary pattern histogram is defined as:

[0093] ;

[0094] in, represents the local binary pattern value, represents the center pixel coordinates, represents the neighborhood pixel index, represents the total number of neighborhood pixels, Represents the neighborhood pixels The gray value of represents the gray value of the center pixel, Represents a sign function used to compare the grayscale values ​​of neighboring pixels and the center pixel gray value The size relationship between them.

[0095] Furthermore, the symbolic function is defined as follows:

[0096] ;

[0097] If the neighboring pixels The gray value of the pixel is greater than or equal to the gray value of the center pixel. ,but =1; if the neighboring pixel The gray value of the pixel is less than the gray value of the center pixel. ,but =0.

[0098] All features include gray-level co-occurrence matrix features, local binary pattern (LBP) histogram to obtain the texture features of each target area and gradient orientation histogram (HOG) features.

[0099] It should be noted that through the application of automated image processing and classifiers, the preliminary segmentation results of the intervertebral disc and nerve roots can be automatically generated, significantly reducing the time and workload of manual labeling; the texture features of the intervertebral disc and nerve root area can be extracted through the gray-level co-occurrence matrix (GLCM), which can improve the accuracy of segmentation; the local binary pattern (LBP) histogram can describe the local texture information of the image, further improving the segmentation effect of the intervertebral disc and nerve root; the gradient direction information of the local area is extracted through the gradient direction histogram (HOG), which helps to capture the edge and shape characteristics of the intervertebral disc and nerve root; combining different types of texture features (GLCM, LBP, HOG) into high-dimensional feature vectors can provide a richer feature description for the support vector machine, thereby improving the learning ability of the classifier and the accuracy of the segmentation results; as a classifier, the support vector machine (SVM) can effectively handle classification problems in high-dimensional feature space and show good performance in image segmentation tasks.

[0100] S5, rigidly registering and non-rigidly registering the preliminary segmentation results of the vertebral body, intervertebral disc and nerve root with the standard three-dimensional model includes the following steps:

[0101] 3D reconstruction is performed using the 2D segmentation data of the segmented vertebrae, intervertebral discs and nerve roots to generate a 3D voxel model; a standard 3D model similar to the patient's age, gender and body shape is selected from the medical database, and a rigid transformation condition is set. Based on the rigid transformation condition, the generated 3D voxel model and the standard 3D model are aligned; the segmentation result and the segmentation model are iterated by iterating the iterative nearest point algorithm to complete the rigid registration in space; control point grids are established on the segmentation result and the standard 3D model after rigid registration respectively; a free shape deformation model is used to locally adjust the standard 3D model through the control point grid; a shape similarity measure between the segmentation result and the standard 3D model is defined to minimize the shape difference between the segmentation result and the standard 3D model, and the displacement vector of the control point is optimized by the gradient descent method to complete the non-rigid registration.

[0102] It should be noted that the free shape deformation model makes local adjustments to the standard three-dimensional model through the control point grid, which can finely handle local deformation and adapt to complex anatomical structure differences; it defines the shape similarity measure between the segmentation result and the standard model, and by minimizing the shape difference, it can ensure that the shape matching between the segmentation model and the standard model is optimal; non-rigid registration enables the segmentation model to flexibly adapt to the shape of the standard model, improving the flexibility and accuracy of registration; through rigid and non-rigid registration, the generated three-dimensional model can be accurately aligned with the standard model, thereby providing a reliable reference for clinical diagnosis and surgical planning.

[0103] S6, based on the shape prior information of the standard three-dimensional model, identifying the abnormal shape segmentation area and correcting it, generating the corrected segmentation result includes the following steps:

[0104] The shape prior information of the standard three-dimensional spine model includes that the shape of the vertebra is usually a regular elliptical cylinder with smooth boundaries, regular boundaries of the intervertebral disc, moderate thickness, and clear boundaries with the upper and lower vertebrae, and the nerve root area is small and relatively fixed in position on both sides of the vertebrae; the vertebrae of the segmentation results are subjected to shape analysis to identify abnormal vertebral shape areas and areas with blurred boundaries between the intervertebral disc and the nerve root, and all detected abnormal shape areas are marked; for the detected abnormal vertebral shape areas, three-dimensional shape operations are used to correct them; for the areas with blurred boundaries of the intervertebral disc, boundary smoothing algorithms are used to correct them; for the nerve root area, three-dimensional template matching technology is used to make the position of the nerve root area consistent with that in the standard spine model; based on the shape correction, a corrected segmentation result is generated.

[0105] Specifically, vertebral shape correction:

[0106] For the detected abnormal vertebral shape area, three-dimensional shape operations (such as expansion, erosion, closing, etc.) are used to correct it. The specific operation steps are:

[0107] According to the shape prior of the standard vertebra, an ideal vertebra shape template is constructed.

[0108] The shape difference between the vertebral body and the template in the segmentation result is calculated, and the abnormal area is gradually approximated to the template shape using shape dilation or erosion operations.

[0109] Disc border correction:

[0110] For the areas with blurred or irregular boundaries of the intervertebral disc, a boundary smoothing algorithm is used for correction. The specific operations are as follows:

[0111] The gradient field of the disc boundary is calculated to identify areas with blurred boundaries.

[0112] Use curve fitting techniques (such as B-spline fitting) to smooth the boundaries and ensure that the boundaries between the intervertebral disc and the upper and lower vertebral bodies are clear.

[0113] Nerve root position correction:

[0114] For the nerve root area, use 3D template matching technology to ensure that the position of the nerve root area is consistent with that in the standard spine model. The specific operations are as follows:

[0115] Using the nerve root position in the standard model as a reference, the nerve root position in the segmentation result was adjusted to ensure that it was located in the correct anatomical position.

[0116] This embodiment also provides a computer device, which is suitable for the CT image segmentation method used for spinal surgery, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the CT image segmentation method for spinal surgery proposed in the above embodiment.

[0117] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0118] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the CT image segmentation method for spinal surgery proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0119] In summary, the present invention improves the image quality through denoising, grayscale normalization and local contrast enhancement processing, lays the foundation for subsequent vertebral space detection and segmentation, and avoids the influence of noise and grayscale unevenness on segmentation accuracy; in the automatic segmentation link, the vertebral space detection method based on grayscale gradient and morphological features is used to automatically determine the upper and lower boundaries of the vertebral body, quickly obtain preliminary segmentation results, and greatly improve the segmentation efficiency; by extracting the local texture features of the intervertebral disc and nerve root, and combining with the support vector machine classifier, it can achieve accurate segmentation of various structures; in addition, rigid and non-rigid registration technology is used, combined with the shape prior information of the standard three-dimensional model, the detected shape abnormal area is corrected to ensure the accuracy and consistency of the final segmentation result.

[0120] Example 2, referring to Table 1, is the second example of the present invention. To further verify the technical solution of the present invention, experimental simulation data of a CT image segmentation method for spinal surgery is provided.

[0121] The spinal CT image data of 10 patients were selected to compare the segmentation effects of the prior art and the solution of the present invention. The prior art adopts a spinal CT segmentation method based on traditional threshold segmentation and edge detection, which segments the vertebral body, intervertebral disc and nerve root area by setting a fixed grayscale threshold. However, this method is easily affected by noise and uneven grayscale when dealing with complex structures (such as vertebral lesions or fuzzy boundaries of nerve root areas), resulting in inaccurate segmentation.

[0122] The segmentation method adopted by the present invention first pre-processes the CT image, including denoising, grayscale normalization, local contrast enhancement and resampling. The specific steps are as follows: First, the CT image is denoised using the bilateral filtering method to remove noise while retaining edge information; then, the denoised image is grayscale normalized to standardize the grayscale value of each pixel, especially to correct the grayscale differences of CT images of different patients; then, the image is locally contrast enhanced using local histogram equalization and bilinear interpolation to improve the visibility of image details; finally, the image is resampled using the trilinear interpolation method to ensure the consistency of slice thickness, thereby providing consistent input data for subsequent segmentation.

[0123] In the segmentation stage, the present invention adopts the intervertebral space detection method, automatically determines the upper and lower boundaries of the vertebra by analyzing the gray gradient changes between the vertebrae, and automatically segments the vertebrae based on the intervertebral space area information. Through the region growing algorithm, seed points are selected for the upper and lower boundary areas of each vertebra, and the gray similarity threshold is set to gradually expand the segmentation area to form a preliminary vertebral segmentation result. For the intervertebral disc and nerve root areas, local texture features are extracted, and the support vector machine classifier is used for classification, and finally a preliminary segmentation result is generated. Finally, the segmentation result is rigidly and non-rigidly aligned with the standard three-dimensional model, and the abnormal shape area is corrected to generate a corrected segmentation result.

[0124] The details are shown in Table 1 below:

[0125] Table 1 Experimental data comparison table

[0126]

[0127] By comparing the experimental data, it can be clearly seen that the CT image segmentation method of the present invention is superior to the prior art in many aspects. First, in terms of segmentation accuracy, the vertebral segmentation accuracy of the prior art is generally around 78% to 80%, while the segmentation accuracy of the present invention is significantly improved to 94% to 96%. This is because the present invention adopts an automatic segmentation method based on grayscale gradient and regional growth, which can more accurately identify the upper and lower boundaries of the vertebral body and avoid the grayscale unevenness problem in the traditional threshold segmentation method.

[0128] In addition, in terms of the segmentation accuracy of the intervertebral disc and nerve root, the existing technology has poor segmentation effects due to the use of fixed thresholds and simple edge detection algorithms, which are between 70% and 74% and 65% and 68% respectively. However, the present invention can effectively distinguish the different tissue characteristics of the vertebral body, intervertebral disc and nerve root by extracting local texture features and combining the support vector machine classifier, significantly improving the segmentation accuracy to more than 90%, especially in the segmentation of the nerve root area, the accuracy is improved by more than 20%. This is of great significance for nerve protection in spinal surgery and reduces the risk of accidental nerve injury during surgery.

[0129] In terms of lesion recognition rate, the recognition rate of the present invention has also been significantly improved, reaching more than 85%, while the existing technology is only about 60%. This is because the present invention combines the shape prior information of the standard three-dimensional model and performs rigid and non-rigid registration, which can automatically identify and correct the shape abnormal area, greatly improving the recognition accuracy of the lesion area. The existing technology is prone to misjudgment of the lesion area due to the lack of an effective shape correction mechanism.

[0130] In terms of segmentation time, the present invention optimizes the input quality of the image through a series of preprocessing steps (such as denoising, grayscale normalization, etc.), reduces the computational burden in the segmentation process, and thus reduces the segmentation time by about 30% to 40% compared with the prior art. This is particularly important in actual clinical applications, as it can speed up preoperative planning and shorten surgical preparation time.

[0131] Finally, in terms of the error segmentation rate, the existing technology has a high error segmentation rate of about 11% to 13%, which is mainly due to the fact that the existing technology is easily affected by noise and uneven grayscale when processing complex spinal structures. However, the present invention uses multi-level segmentation and shape correction to reduce the error segmentation rate to only about 3%, greatly improving the reliability of the segmentation results.

[0132] In summary, compared with the existing technology, the present invention has significant advantages in segmentation accuracy, lesion recognition rate, segmentation time and mis-segmentation rate, especially in the segmentation and shape correction of complex spinal structures, showing unique practicality.

[0133] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A CT image segmentation method for spinal surgery, characterized in that: include, Obtaining the patient's spinal CT images and preprocessing the spinal CT images; The intervertebral space detection method is used to detect the intervertebral space area of ​​the spine in the preprocessed CT image, and the upper and lower boundaries of the vertebra are automatically determined by analyzing the gray gradient changes and morphological characteristics between the vertebrae. Each vertebra is automatically segmented based on the intervertebral space area information to obtain the preliminary segmentation result of the vertebra; Extract local texture features of the intervertebral disc and nerve root in the intervertebral space area, and classify them through a classifier to generate preliminary segmentation results of the intervertebral disc and nerve root; The preliminary segmentation results of vertebral bodies, intervertebral discs and nerve roots are rigidly and non-rigidly registered with the standard 3D model to identify abnormal areas in the segmentation; Based on the shape prior information of the standard 3D model, the shape of the abnormal segmentation area is corrected to generate a corrected segmentation result; Extracting local texture features of the intervertebral disc and nerve root in the intervertebral space area and classifying them through a classifier to generate preliminary segmentation results of the intervertebral disc and nerve root includes the following steps: Based on the vertebral segmentation area, the upper limit boundary of each vertebra is determined, and the middle area between each pair of adjacent vertebrae is used as the target area of ​​the intervertebral disc; The target area of ​​the nerve root is estimated by analyzing the upper and lower borders of the vertebral body and the target area of ​​the intervertebral disc; Extracting gray-level co-occurrence matrix-based features from each target region; For each pixel in the target area, the texture features of each target area are obtained by calculating the local binary pattern histogram; Calculate the gradient direction histogram for each target area to obtain the gradient direction information of the local area; Combine all the extracted features in sequence to form the final high-dimensional feature vector; Select support vector machine as the classifier and train the support vector machine using the CT image dataset with known annotations; The extracted high-dimensional feature vector is input into the trained support vector machine. According to the output of the classifier, the areas belonging to the intervertebral disc and nerve root in the test image are marked to generate the segmentation results of the intervertebral disc and nerve root.

2. The CT image segmentation method for spinal surgery according to claim 1, characterized in that: The preprocessing includes denoising, grayscale normalization, local contrast enhancement and resampling processing.

3. The CT image segmentation method for spinal surgery according to claim 2, characterized in that: Denoising refers to removing image noise using a bilateral filtering method to obtain a grayscale value of a filtered image; Grayscale normalization refers to obtaining the grayscale value in the denoised image and performing grayscale normalization on each pixel in the image; The local contrast enhancement refers to dividing the normalized image into sub-blocks of equal size, performing local histogram equalization on each sub-block, and performing bilinear interpolation to reassemble the enhanced image into a complete image; The resampling refers to determining the slice thickness of the combined complete image, performing trilinear interpolation on the image, and adjusting the uneven slice thickness to a uniform thickness.

4. The CT image segmentation method for spinal surgery according to claim 3, characterized in that: The vertebral gap detection method is used to detect the vertebral gap area of ​​the spine in the preprocessed CT image. By analyzing the gray gradient changes and morphological characteristics between vertebrae, the upper and lower boundaries of the vertebrae are automatically determined. The following steps are included: The grayscale gradient of each pixel of the preprocessed CT image is calculated to generate a gradient amplitude image; In the gradient amplitude image, find the area with obvious gradient changes between the upper and lower boundaries of the spine and the intervertebral disc, and mark the area with the most significant gradient change as the location of the intervertebral space; For the detected intervertebral gap area, small noise points are removed and the broken gap area is connected; The distance between the upper and lower borders of each vertebra was recorded by calculating the width of the gap area after connection; For each vertebral space area, the gray value distribution of its internal pixels is analyzed to generate a gray histogram to further determine the upper and lower boundaries of the vertebra.

5. The CT image segmentation method for spinal surgery according to claim 4, characterized in that: Automatically segment each vertebra based on the intervertebral space area information to obtain the preliminary segmentation result of the vertebra, including the following steps: Between the upper and lower boundaries of each vertebra, M seed points are selected for region growing; According to the characteristics of the spine CT image, the grayscale similarity threshold is set; For each seed point, a growth region is initialized, and pixels adjacent to the seed point and with grayscale value differences less than the grayscale similarity threshold are included in the growth region; Starting from the boundary point of the current growing region, the neighborhood check and region update are repeated until the grayscale difference exceeds the grayscale similarity threshold, then the region growth is stopped to form the final vertebral segmentation region.

6. The CT image segmentation method for spinal surgery according to claim 5, characterized in that: The rigid and non-rigid registration of the preliminary segmentation results of the vertebral body, intervertebral disc and nerve root with the standard 3D model includes the following steps: 3D reconstruction is performed using the 2D segmentation data of the segmented vertebral body, intervertebral disc and nerve root to generate a 3D voxel model; Select a standard 3D model similar to the patient's age, gender and body shape from the medical database and set rigid transformation conditions; Based on the rigid transformation condition, the generated 3D voxel model is aligned with the standard 3D model; The segmentation result and the segmentation model are rigidly registered in space by iterating the closest point algorithm. Establish control point grids on the segmentation result after rigid registration and the standard three-dimensional model respectively; Using free-shape deformation models, local adjustments are made to standard 3D models through a grid of control points; The shape similarity measure between the segmentation result and the standard 3D model is defined to minimize the shape difference between the segmentation result and the standard 3D model, and the displacement vector of the control point is optimized by the gradient descent method to complete the non-rigid registration.

7. The CT image segmentation method for spinal surgery according to claim 6, characterized in that: Based on the shape prior information of the standard 3D model, the abnormal shape segmentation area is identified and corrected, and the corrected segmentation result is generated, including the following steps: Perform shape analysis on the vertebrae of the segmentation results, identify the abnormal vertebral shape areas, as well as the areas of the boundary models of the intervertebral disc and the nerve root, and mark all detected abnormal shape areas; For the detected abnormal vertebral shape areas, three-dimensional shape operations are used to correct them; For the areas with blurred intervertebral disc boundaries, a boundary smoothing algorithm was used for correction; For the nerve root region, the three-dimensional template matching technology was used to adjust the position of the nerve root region to be consistent with the position in the standard spine model; Based on the shape correction, a corrected segmentation result is generated.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the CT image segmentation method for spinal surgery described in any one of claims 1 to 7 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the CT image segmentation method for spinal surgery described in any one of claims 1 to 7 are implemented.

Images