Brain tumor focus area segmentation method based on CT image analysis

Abstract

The invention relates to the technical field of image processing, in particular to a brain tumor focus area segmentation method based on CT image analysis. The method comprises the following steps: segmenting a meninx CT image to obtain an initial segmentation region, and preprocessing the initial segmentation region to obtain an enhanced image; obtaining a gray gradient direction vector of each pixel point in the enhanced image; skeleton lines in each foreground area and tangent lines of all end points of all the skeleton lines are obtained, and then suspected cut-off end points of all the end points are obtained; according to the gray gradient direction vector of each pixel point of the foreground area where each suspected truncation end point of one end point is located, screening the pixel points of the area where each suspected truncation end point is located to obtain disordered pixel points of the end point, and further obtaining each tumor part area of the end point; and taking the central point of each tumor partial region of each end point as a seed point of region growth, and respectively carrying out region growth in each tumor partial region to obtain a tumor region. According to the invention, the brain tumor region can be effectively segmented.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and specifically to a method for segmenting brain tumor lesion regions based on CT image analysis. Background Technology

[0002] Precise segmentation of brain tumors is one of the core challenges in the field of medical image analysis, as its accuracy directly affects the precision of clinical diagnosis, the reliability of surgical planning, and the effectiveness of treatment outcome evaluation. Currently, brain tumor segmentation techniques based on multi-slice spiral CT images mainly rely on traditional methods such as image grayscale features, texture analysis, and region growing. For tumor lesions located within the brain with significant grayscale contrast to surrounding normal brain tissue, these methods can achieve relatively accurate localization and segmentation by analyzing local grayscale differences. In particular, threshold segmentation and traditional region growing methods, when processing multi-slice CT images with homogeneous texture within the brain, can utilize the grayscale differences between lesions and normal tissue to achieve relatively accurate tumor localization and segmentation, and have achieved certain successes in clinical applications.

[0003] While existing technologies demonstrate good performance in segmenting intracranial brain tumors, traditional segmentation methods face significant challenges when tumors are located in the cerebral cortex (especially meningiomas). The surface of the cerebral cortex is not a smooth plane but rather covered with complex and tortuous sulci structures. These sulci have a natural, continuous orientation and connectivity in CT images. When using traditional segmentation methods in cortical regions, if the tumor and the sulci overlap, the natural presence of the sulci can cause unreasonable contraction or expansion of the segmentation boundary at the sulci, making it difficult to accurately determine the actual tumor boundary. Although existing preprocessing techniques can obtain the approximate area of ​​the tumor, they cannot achieve fine segmentation of the lesion boundary, resulting in a significant deviation between the segmentation result and the actual tumor area, seriously affecting subsequent clinical decisions. Summary of the Invention

[0004] To address the aforementioned technical problems, the present invention aims to provide a method for segmenting brain tumor lesion regions based on CT image analysis. The specific technical solution adopted is as follows:

[0005] One embodiment of the present invention provides a method for segmenting brain tumor lesion regions based on CT image analysis, the method comprising:

[0006] The meningeal CT image is segmented to obtain the initial segmentation region, and the initial segmentation region is mapped to a separate image for preprocessing to obtain the enhanced image; the gray-level gradient direction vector of each pixel is obtained based on the gray-level changes of each pixel in the enhanced image.

[0007] The brain sulcus region in the enhanced image is used as the foreground region, and other regions are used as the background region. The images are then binarized to obtain a binary image. The Zhang-Suen algorithm is used to process the foreground region in the binary image to obtain the skeleton lines in each foreground region.

[0008] Get the tangent line of a skeleton line by performing curve fitting based on the pixel point adjacent to an endpoint of the skeleton line and the coordinates of the endpoint; get the candidate endpoints of an endpoint based on the distance between an endpoint of a skeleton line and the endpoints of other skeleton lines; get the suspected truncated endpoints of an endpoint based on the angle between an endpoint and the tangent lines of the candidate endpoints of the endpoint.

[0009] The pixels in the foreground region of each suspected truncated endpoint are filtered to obtain the disordered pixels of the endpoint. Based on all the disordered pixels of an endpoint, the tumor regions of the endpoint are obtained. The center point of each tumor region of each endpoint is used as the seed point for region growth. Region growth is performed in each tumor region to obtain the tumor region.

[0010] Preferably, the gray-level gradient direction vector of each pixel is obtained based on the gray-level changes of each pixel in the enhanced image, including:

[0011] Obtain the gradient components of a pixel in the enhanced image in the x and y directions, and the gradient components in the x and y directions constitute the direction vector of the pixel.

[0012] Preferably, obtaining the tangent line at a skeleton line by performing curve fitting based on the pixel point adjacent to an endpoint of the skeleton line and the coordinates of that endpoint includes:

[0013] For an endpoint of a skeleton line, a preset number of pixels on the skeleton line are sequentially taken from the endpoint to the skeleton line where the endpoint is located, and recorded as the neighboring pixels of the endpoint; curve fitting is performed using the coordinates of the endpoint and each neighboring pixel of the endpoint to obtain a fitted curve, and the tangent of the fitted curve is obtained through the endpoint.

[0014] Preferably, obtaining candidate endpoints for an endpoint based on the distance between an endpoint on one skeleton line and endpoints on other skeleton lines includes:

[0015] Calculate the distance between an endpoint on a skeleton line and all endpoints on other skeleton lines, and select endpoints whose distance is less than a distance threshold as candidate endpoints for that endpoint.

[0016] Preferably, obtaining the suspected truncated endpoint of an endpoint based on the angle between an endpoint and the tangents of each candidate endpoint of that endpoint includes:

[0017] Obtain the angle between the tangent of an endpoint and the tangent of a candidate endpoint of that endpoint, and use it as the tangent angle; use an exponential function with the natural constant as the base to perform a negative correlation mapping on the ratio of the square of the tangent angle to the angle decay parameter to obtain the directional consistency index of a candidate endpoint of that endpoint; if the directional consistency index of a candidate endpoint of that endpoint is greater than the directional consistency threshold, then the candidate endpoint is a suspected truncated endpoint of that endpoint.

[0018] Preferably, the pixels in the region where each suspected truncated endpoint is located are filtered based on the grayscale gradient direction vector of each pixel in the foreground region where each suspected truncated endpoint is located to obtain the disordered pixels of that endpoint, including:

[0019] For an endpoint, each pixel in the foreground region where each suspected truncated endpoint is located is recorded as a pixel to be analyzed. A set of pixels corresponding to the pixel to be analyzed and all pixels in its eight-neighborhood are formed. The arithmetic mean of the gray-level gradient direction vectors of all pixels in the set of pixels corresponding to the pixel to be analyzed is calculated to obtain the average gray-level gradient direction vector. The average cosine of the angle between the gray-level gradient direction vector of each pixel in the set of pixels corresponding to the pixel to be analyzed and the average gray-level gradient direction vector is calculated and recorded as the average angle cosine value. The disorder level of the pixel to be analyzed is obtained by subtracting the average angle cosine value from a first preset value. If the disorder level of a pixel to be analyzed is greater than the disorder level threshold, then the pixel to be analyzed is a disordered pixel.

[0020] Preferably, obtaining each tumor region of an endpoint based on all disordered pixels at that endpoint includes:

[0021] Morphological processing is performed on all the disordered pixels of an endpoint to obtain different mask regions; if a mask region contains a suspected truncated endpoint of the endpoint, then the mask region is a tumor region.

[0022] The embodiments of the present invention have at least the following beneficial effects: This application segments a meningeal CT image to obtain an initial segmented region, and maps the initial segmented region onto a separate image for preprocessing to obtain an enhanced image; based on the grayscale changes of each pixel in the enhanced image, the grayscale gradient direction vector of each pixel is obtained to represent the gradient changes of the pixel; furthermore, the sulcus region in the enhanced image is used as the foreground region, and other regions are used as the background region for binarization to obtain a binary image; the Zhang-Suen algorithm is used to process the foreground region in the binary image to obtain the skeleton in each foreground region. Lines represent the trend of sulci on the surface of the cerebral cortex. Further, based on the essential difference between the natural continuity of the trend of sulci in intact cortical regions and the interruption of connectivity caused by tumor invasion, suspected truncation endpoints are obtained at each endpoint. Then, based on the suspected truncation endpoints, further analysis is performed. By detecting the areas in the trend of sulci that are truncated by the tumor, the tumor regions at each endpoint are obtained. Finally, the center point of each tumor region at each endpoint is used as the seed point for region growth. Region growth is performed in each tumor region to obtain the tumor region, thereby segmenting a more refined tumor lesion region and achieving precise segmentation of the meningioma boundary. Attached Figure Description

[0023] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating a method for segmenting brain tumor lesions based on CT image analysis, as provided in an embodiment of the present invention. Detailed Implementation

[0025] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a brain tumor lesion region segmentation method based on CT image analysis proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0027] The following description, in conjunction with the accompanying drawings, details a specific scheme for a brain tumor lesion region segmentation method based on CT image analysis provided by the present invention.

[0028] Example:

[0029] The main application scenario of this invention is: This application is mainly used for the segmentation of brain tumor lesion areas.

[0030] Please see Figure 1 The diagram illustrates a method flowchart for brain tumor lesion region segmentation based on CT image analysis, provided by an embodiment of the present invention. The method includes the following steps:

[0031] Step S1: Segment the meningeal CT image to obtain the initial segmentation region, and map the initial segmentation region onto a separate image for preprocessing to obtain the enhanced image; obtain the gray-level gradient direction vector of each pixel based on the gray-level changes of each pixel in the enhanced image.

[0032] First, the acquired coarsely segmented regions need to be optimized to provide high-quality, high-contrast regions of interest for subsequent fine segmentation algorithms based on sulcus trend analysis. By performing precise image alignment, structural enhancement, and noise suppression on the coarsely segmented regions, this step effectively highlights the detailed features of the sulcus structures within the cortical region, while eliminating artifacts and registration errors that may have been introduced in the earlier processing. This ensures that subsequent sulcus connectivity analysis is based on stable and clean local data. This process accurately defines the spatial extent for subsequent analysis and clearly presents the continuous directional features of the sulci in the enhanced local images.

[0033] First, a CT image of the meninges is obtained. Then, existing automated or semi-automated segmentation methods based on CT images (such as threshold-based segmentation or lightweight neural network segmentation) are used to coarsely segment the meninges CT image to obtain an initial segmentation region, which includes the tumor, normal meningeal tissue, and sulci.

[0034] Furthermore, the acquired initial segmented region needs to be mapped onto a separate image, which is designated as the image to be processed for subsequent analysis. Next, the image to be processed is preprocessed. Multislice spiral CT is used to rapidly capture brain CT images of each layer within the initial segmented region (the image to be processed). Utilizing the spatial correspondence between the cerebral cortex and its adjacent layers, the image to be processed is precisely aligned locally to ensure geometric consistency, thus obtaining the aligned image to be processed.

[0035] Then, within the aligned image to be processed, a filter specifically designed to enhance linear and sulcular structures (such as a vascular ridge enhancement filter based on the Hessian matrix) is applied to filter the aligned image to obtain the filtered image to enhance the continuous orientation of brain sulci while suppressing irrelevant tissue texture interference.

[0036] Finally, the filtered image needs to be denoised to further enhance its quality. Therefore, nonlocal mean denoising or anisotropic diffusion filtering is applied to the filtered image to enhance the direction of the sulci, resulting in the final enhanced image. This effectively smooths noise in uniform regions while preserving the integrity of the sulci edges to the maximum extent.

[0037] This yields an enhanced image containing the meningeal region, sulci, and tumor. It should be noted that the preprocessing performed on the initial segmented region mapped to an image is a well-known technique and will not be elaborated upon here.

[0038] Furthermore, even after preprocessing and optimization, while the brain sulcus structure is enhanced, its features, such as its direction, remain vague concepts implicit in the image grayscale and cannot be directly processed. Therefore, it is necessary to transform this anatomical feature of brain sulcus direction into a quantitative and structured mathematical model to establish a reliable computational benchmark for subsequent connectivity analysis. By calculating the direction and angle of image gradients, the local direction of the brain sulcus at each pixel can be precisely quantified, clarifying the "direction" as the direction and magnitude of a vector. This provides indispensable input data and judgment criteria for subsequent steps such as truncated region identification.

[0039] Therefore, the gray-level gradient direction vector of each pixel is obtained based on the gray-level changes of each pixel in the enhanced image.

[0040] Specifically, the gradient components of a pixel in the enhanced image in the x and y directions are obtained, and the gradient components in the x and y directions constitute the direction vector of the pixel.

[0041] The gradient components of one pixel in the x and y directions need to be obtained using the Sobel operator or a more precise central difference algorithm, and denoted as G respectively. x and G y Then the grayscale gradient direction vector of this pixel is (G x G y ).

[0042] The angle of the gray-level gradient direction vector is (the vector direction is perpendicular to the edge of the brain sulcus and points in the direction of gray-level decrease): The magnitude of the gray-level gradient direction vector is The direction of the gray-level gradient can characterize the significance of the trend. At each pixel, the direction of the gray-level gradient direction vector is perpendicular to the local brain sulcus edge, pointing in the direction of the fastest gray-level decrease. The larger the magnitude of the gray-level gradient direction vector, the clearer the brain sulcus edge and the more defined the trend. This is the foundational operation, quantifying the state of the brain sulcus trend (including normal brain sulcus trends and truncated abnormal brain sulcus trends), serving as a pre-operation for subsequent steps. This yields the gray-level gradient direction vector for each pixel in the enhanced image.

[0043] Step S2: The brain sulcus region in the enhanced image is used as the foreground region and other regions are used as the background region for binarization to obtain a binary image; the Zhang-Suen algorithm is used to process the foreground region in the binary image to obtain the skeleton line in each foreground region.

[0044] The above obtained enhanced images after enhancing the direction of the sulci. Although some tumor features were identified in the general area, the sulci and grayscale characteristics of the meninges themselves made precise segmentation impossible. Now it is necessary to analyze the damage of the tumor to the features of the sulci region in order to obtain the edge information of the tumor and locate the tumor area.

[0045] In the enhanced image obtained from the previous steps, which enhances the direction of the sulci, we abandon the traditional approach that relies on grayscale features and instead analyze the connectivity features of the sulci on the surface of the cerebral cortex. Although tumors in the sulci can make segmentation difficult, the normal morphological features of the sulci are disrupted, and the edges of the tumor can be effectively extracted at the location of the disruption. This allows us to effectively locate the tumor lesion area and significantly reduce the coarse segmentation connectivity of the approximate area, thereby achieving more accurate lesion segmentation.

[0046] Thus, the orientation vector of each pixel in the enhanced image was obtained in the above steps. Then, skeleton processing was performed to highlight the trend. All connected sulci were acquired and located, followed by orientation disorder analysis to obtain the actual tumor truncation edge. The obtained optimized edge can be used as the basis for re-segmentation of the original region, ultimately achieving fine segmentation of the lesion.

[0047] Specifically, the enhanced image obtained after preprocessing is first subjected to skeleton extraction to obtain the brain sulcus center line with a single pixel width as the skeleton line, and then post-analysis is performed.

[0048] First, in the enhanced image, the grayscale difference between the brain sulci and the tumor and normal tissue is significant. Therefore, binarization is performed using an appropriate threshold, treating the brain sulci region in the enhanced image as the foreground region and the other regions as the background region to obtain a binary image. In the binary image, the highlighted areas represent the brain sulci region (foreground region).

[0049] Furthermore, an iterative thinning algorithm is applied to the foreground region in the binary image. Preferably, the Zhang-Suen algorithm is used in this application to process the foreground region in the binary image separately, gradually stripping away boundary pixels while preserving the centerline structure to obtain a skeleton of single-pixel width as the skeleton line in each foreground region. Before obtaining the final skeleton line, post-processing is required, including removing short branches (isolated skeleton segments with a length less than a threshold) to connect minor breaks caused by noise, thus obtaining the skeleton line in each foreground region. After obtaining the skeleton line, the location of the actual truncation can be analyzed based on the continuity of the skeleton line. It should be noted that binarization processing and skeleton extraction are well-known techniques and will not be described in detail here.

[0050] Step S3: Obtain the tangent of an endpoint by performing curve fitting based on the pixel point adjacent to an endpoint of a skeleton line and the coordinates of the endpoint; obtain candidate endpoints of an endpoint based on the distance between an endpoint of a skeleton line and the endpoints of other skeleton lines; obtain suspected truncated endpoints of an endpoint based on the angle between an endpoint and the tangents of the candidate endpoints of the endpoint.

[0051] The skeleton lines obtained through the above steps are 1 pixel wide. By simplifying the sulci into single-pixel-wide skeletons, the influence of variations in sulcus width can be eliminated, allowing for direct analysis of their connectivity. For sulci truncated by a tumor, the extensions of the two skeleton lines should ideally connect naturally, but due to tumor occupation, this connection is physically blocked. By analyzing the consistency of the extension directions of the endpoints of the skeleton lines, regions with highly consistent directions but spatially interrupted can be identified, thus providing a preliminary location of suspected truncated areas. These regions may contain actual tumor truncations, or they may include termination points of normal sulci or noise.

[0052] Furthermore, the two endpoints of each skeleton line are obtained. For each endpoint of a skeleton line, a predetermined number of pixels on the skeleton line are sequentially taken from that endpoint towards the skeleton line containing that endpoint, and these are recorded as the neighboring pixels of that endpoint. A curve fitting is performed using the coordinates of the endpoint and each of its neighboring pixels to obtain a fitted curve. The tangent line to the fitted curve is then drawn through the endpoint to obtain the tangent line of that endpoint. In this way, the tangent lines of each endpoint of each skeleton line can be obtained.

[0053] Curve fitting is a well-known technique and will not be elaborated on here. The preset number is 5. The more numbers there are, the more accurate the fitted curve will be. The preset number can be adjusted according to the balance between computational efficiency and accuracy.

[0054] Furthermore, it is necessary to calculate the distance between an endpoint on one skeletal line and all endpoints on other skeletal lines, and select endpoints whose distance is less than a distance threshold as candidate endpoints. The distance threshold is the diameter of the largest tumor in the patient's history of tumors of the same type and in the early to mid-stages. The distance between two segments of the same sulcus separated by a tumor usually does not exceed the diameter of the tumor itself, and the largest diameter of early to mid-stage tumors in history has a common statistical range. This allows for the exclusion of distal endpoints that are far from the current endpoint, obtaining candidate endpoints for each endpoint. Additionally, endpoints that have been analyzed in pairs are not re-analyzed in subsequent analyses. For example, if there are three endpoints a, b, and c, and the candidate endpoints for a are b and c, then when obtaining candidate endpoints for b and c, a need not be considered again, thus improving computational efficiency.

[0055] Furthermore, it is necessary to analyze the changes in the direction of the skeleton lines to determine which endpoints among the candidate endpoints are actually truncated. Thus, the suspected truncated endpoints are obtained based on the angles between an endpoint and the tangents of each of its candidate endpoints.

[0056] Specifically, the angle between the tangent of an endpoint and the tangent of a candidate endpoint is obtained as the tangent angle; the ratio of the square of the tangent angle to the angle decay parameter is negatively correlated using an exponential function with the natural constant as the base to obtain the directional consistency index of a candidate endpoint; if the directional consistency index of a candidate endpoint is greater than the directional consistency threshold, then the candidate endpoint is a suspected truncated endpoint of the endpoint.

[0057] The specific calculation model for the directional consistency index is as follows:

[0058] ,

[0059] Where C represents the directional consistency index between an endpoint and a candidate endpoint of that endpoint; θ represents the angle between the tangent of the endpoint and the tangent of a candidate endpoint of that endpoint, i.e., the angle of the tangents, and the range of this angle is mapped to [the relevant information] during the calculation. This is used to measure the directional consistency between tangents. The denominator is the parameter that controls the angle decay, i.e., the angle decay parameter. To prevent the index from changing too quickly, Take half the angle of the largest tangent, that is Overall, the smaller the angle and the more consistent the direction of the extended lines, the more likely they originally belonged to the same sulcus and remained continuous in direction after being truncated by the tumor. The larger the value of C, the better.

[0060] The directional consistency threshold is the 90th percentile of all C values ​​obtained through the above steps in brain sulci regions without tumors. This allows for the identification of potential truncation endpoints for each endpoint.

[0061] This step first identifies the endpoints of the skeleton line and analyzes their virtual connection relationship. By calculating the consistency of the endpoints with the direction of the nearby tangents, unnatural endpoints formed due to tumor truncation are initially screened out, which are suspected truncation endpoints, and then further analysis is carried out.

[0062] Step S4: Based on the gray-level gradient direction vector of each pixel in the foreground region where each suspected truncated endpoint is located, the pixels in the region where each suspected truncated endpoint is located are filtered to obtain the disordered pixels of the endpoint; based on all the disordered pixels of an endpoint, each tumor part region of the endpoint is obtained; the center point of each tumor part region of each endpoint is used as the seed point for region growth, and region growth is performed in each tumor part region to obtain the tumor region.

[0063] The above process yielded suspected truncation endpoints for each endpoint. These suspected truncation endpoints included actual truncations, but also noise from normal brain sulci. Therefore, noise interference still needs to be eliminated.

[0064] First, the foreground region (brain sulcus region) in the obtained binary image needs to be mapped onto the enhanced image, and then further analysis can be performed on the enhanced image.

[0065] Next, based on the grayscale gradient direction vector of each pixel in the foreground region where each suspected truncated endpoint is located, the pixels in the region where each suspected truncated endpoint is located are filtered to obtain the disordered pixels of that endpoint.

[0066] Specifically, for an endpoint, each pixel in the foreground region where each suspected truncated endpoint is located is recorded as a pixel to be analyzed; a pixel to be analyzed and all pixels within its eight-neighborhood are combined to form a pixel set corresponding to the pixel to be analyzed; the arithmetic mean of the gray-level gradient direction vectors of all pixels in the pixel set corresponding to the pixel to be analyzed is calculated to obtain the average gray-level gradient direction vector; the average cosine of the angle between the gray-level gradient direction vector of each pixel in the pixel set corresponding to the pixel to be analyzed and the average gray-level gradient direction vector is calculated and recorded as the average angle cosine value; the disorder level of the pixel to be analyzed is obtained by subtracting the average angle cosine value from a first preset value; if the disorder level of a pixel to be analyzed is greater than the disorder level threshold, then the pixel to be analyzed is a disordered pixel.

[0067] The specific model for calculating the degree of disorder is as follows:

[0068] ,

[0069] Wherein, HL represents the disorder level of a pixel to be analyzed corresponding to an endpoint, that is, a pixel in the foreground region where each of the suspected truncated endpoints of the endpoint is located; N represents the number of pixels in the set of pixels corresponding to the pixel to be analyzed, which consists of the pixel to be analyzed and all pixels in the eight neighborhoods of the pixel to be analyzed. Since it is an eight neighborhood, the value of N is 9. The grayscale gradient direction vector of the q-th pixel in the pixel set; This represents the average gray-level gradient direction vector obtained by calculating the arithmetic mean of the gray-level gradient direction vectors of all pixels in the set of pixels corresponding to the pixel to be analyzed. The denominator is the product of the magnitude of the gray-level gradient direction vector of the q-th pixel and the magnitude of the average gray-level gradient direction vector. Let be the cosine of the angle between the gray-level gradient direction vector of the q-th pixel in the pixel set and the average gray-level gradient direction vector. The larger the value, the more consistent the directions. HL is the cosine of the average angle. The larger the HL value, the more disordered the pixel is, the more likely it is to be located at the truncation point of a tumor, and the more likely it is to be a pixel in the tumor region.

[0070] In addition, the disorder level threshold is obtained by obtaining the HL of each pixel in other brain sulci regions (foreground regions) without tumors, and then taking the maximum value as the disorder level threshold.

[0071] Next, morphological processing is performed on all the disordered pixels of an endpoint to obtain different mask regions; if a mask region contains a suspected truncated endpoint of the endpoint, then the mask region is a tumor region.

[0072] It should be noted that morphological processing is performed on all disordered pixels at an endpoint, including morphological closing operations, merging neighboring regions, filling holes, smoothing boundaries, etc. This technique is well-known and will not be elaborated on here. In addition, the mask region obtained here is only a part of the tumor region, so subsequent region growing is required to obtain the complete tumor region.

[0073] Finally, the above yields the actual tumor region, which is the area where the tumor covers the brain sulci. These regions provide the convergence outline of the actual tumor region for the final fine segmentation. Fine segmentation can be achieved using existing segmentation techniques (here, the seed growth segmentation method is used). The general steps are as follows:

[0074] Region merging: Align the obtained tumor regions with the enhanced image, that is, map each tumor region onto the enhanced image to obtain an enhanced image that marks the tumor region.

[0075] Seed point selection: The above results are the actual tumor region. Selecting the seed point within it ensures that the initial growth location is exactly inside the tumor. Therefore, the center point of the actual tumor region is used as the seed point.

[0076] Segmentation: Region growing is performed in each tumor region using existing region growing algorithms.

[0077] Boundary processing and result generation: The segmentation boundaries after region growing are smoothed to eliminate unreasonable jagged edges. Morphological opening operations are applied to remove small holes and isolated regions, ensuring that the final boundaries are continuous and closed, thus achieving refined segmentation and obtaining the tumor region. Region growing and post-processing are common image processing techniques and will not be elaborated on in detail here.

[0078] In summary, this application adopts a hierarchical verification strategy. First, it locates the suspected truncation endpoints through brain sulcus morphology analysis, and then verifies and confirms the actual tumor region through gradient features (grayscale gradient direction vector), thereby improving the accuracy of tumor region segmentation.

[0079] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0080] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. The above descriptions are merely preferred embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for segmenting brain tumor lesion regions based on CT image analysis, characterized in that, The method includes: The meningeal CT image is segmented to obtain the initial segmentation region, and the initial segmentation region is mapped to a separate image for preprocessing to obtain the enhanced image; the gray-level gradient direction vector of each pixel is obtained based on the gray-level changes of each pixel in the enhanced image. The brain sulcus region in the enhanced image is used as the foreground region, and other regions are used as the background region. The images are then binarized to obtain a binary image. The Zhang-Suen algorithm is used to process the foreground region in the binary image to obtain the skeleton lines in each foreground region. Get the tangent line of a skeleton line by performing curve fitting based on the pixel point adjacent to an endpoint of the skeleton line and the coordinates of the endpoint; get the candidate endpoints of an endpoint based on the distance between an endpoint of a skeleton line and the endpoints of other skeleton lines; get the suspected truncated endpoints of an endpoint based on the angle between an endpoint and the tangent lines of the candidate endpoints of the endpoint. The pixels in the foreground region of each suspected truncated endpoint are filtered to obtain the disordered pixels of the endpoint. The tumor regions of the endpoint are obtained based on all the disordered pixels of the endpoint. The center point of each tumor region of each endpoint is used as the seed point for region growth. Region growth is performed in each tumor region to obtain the tumor region. The step of filtering pixels in the foreground region where each suspected truncated endpoint is located based on the grayscale gradient direction vector of each pixel in the region where each suspected truncated endpoint is located to obtain the disordered pixels of that endpoint includes: For an endpoint, each pixel in the foreground region where each suspected truncated endpoint is located is recorded as a pixel to be analyzed. A set of pixels corresponding to the pixel to be analyzed and all pixels in its eight-neighborhood are formed. The arithmetic mean of the gray-level gradient direction vectors of all pixels in the set of pixels corresponding to the pixel to be analyzed is calculated to obtain the average gray-level gradient direction vector. The average cosine of the angle between the gray-level gradient direction vector of each pixel in the set of pixels corresponding to the pixel to be analyzed and the average gray-level gradient direction vector is calculated and recorded as the average angle cosine value. The disorder level of the pixel to be analyzed is obtained by subtracting the average angle cosine value from a first preset value. If the disorder level of a pixel to be analyzed is greater than the disorder level threshold, then the pixel to be analyzed is a disordered pixel.

2. The method for segmenting brain tumor lesion regions based on CT image analysis according to claim 1, characterized in that, The step of obtaining the gray-level gradient direction vector of each pixel based on the gray-level changes of each pixel in the enhanced image includes: Obtain the gradient components of a pixel in the enhanced image in the x and y directions, and the gradient components in the x and y directions constitute the direction vector of the pixel.

3. The method for segmenting brain tumor lesion regions based on CT image analysis according to claim 1, characterized in that, The step of obtaining the tangent line at a skeleton line by performing curve fitting based on the pixel point adjacent to one endpoint of the skeleton line and the coordinates of that endpoint includes: For an endpoint of a skeleton line, a preset number of pixels on the skeleton line are sequentially taken from the endpoint to the skeleton line where the endpoint is located, and recorded as the neighboring pixels of the endpoint; curve fitting is performed using the coordinates of the endpoint and each neighboring pixel of the endpoint to obtain a fitted curve, and the tangent of the fitted curve is obtained through the endpoint.

4. The method for segmenting brain tumor lesion regions based on CT image analysis according to claim 1, characterized in that, The step of obtaining candidate endpoints for an endpoint based on the distance between an endpoint on a skeleton line and all endpoints on other skeleton lines includes: Calculate the distance between an endpoint on a skeleton line and all endpoints on other skeleton lines, and select endpoints whose distance is less than a distance threshold as candidate endpoints for that endpoint.

5. The method for segmenting brain tumor lesion regions based on CT image analysis according to claim 1, characterized in that, The step of obtaining the suspected truncated endpoint of an endpoint based on the angle between an endpoint and the tangents of each candidate endpoint of that endpoint includes: Obtain the angle between the tangent of an endpoint and the tangent of a candidate endpoint of that endpoint, and use it as the tangent angle; use an exponential function with the natural constant as the base to perform a negative correlation mapping on the ratio of the square of the tangent angle to the angle decay parameter to obtain the directional consistency index of a candidate endpoint of that endpoint; if the directional consistency index of a candidate endpoint of that endpoint is greater than the directional consistency threshold, then the candidate endpoint is a suspected truncated endpoint of that endpoint.

6. The method for segmenting brain tumor lesion regions based on CT image analysis according to claim 1, characterized in that, The step of obtaining each tumor region of an endpoint based on all disordered pixels of that endpoint includes: Morphological processing is performed on all the disordered pixels of an endpoint to obtain different mask regions; if a mask region contains a suspected truncated endpoint of the endpoint, then the mask region is a tumor part region.

Images