Image processing method and system based on fusion growing point optimization algorithm and Unet network

Abstract

The invention relates to the technical field of computer vision and deep learning algorithm optimization, in particular to an image processing method and system based on a fusion growing point optimization algorithm and a Unet network, and the method comprises the following steps: collecting a lumbar vertebra CT image, extracting the gray value and the HU value of each pixel in the image, marking a lumbar vertebra bone region, obtaining the density information of the lumbar vertebra bone region, and obtaining the density information of the lumbar vertebra bone region; analyzing the gradient of the lumbar vertebra, segmenting the initial region, obtaining information points of the basic region from the initial region, classifying, and generating initial structural features of the lumbar vertebra; according to the invention, by collecting the lumbar vertebra CT image and extracting the gray value and the HU value of the pixel, the density information of the lumbar vertebra bone region is obtained and preliminary segmentation is carried out, the gradient change of the region can be accurately analyzed, and a reliable basis is provided for the preliminary region of the image. In the process, the deviation of the lumbar vertebra bone form is successfully corrected by dynamically adjusting the image rotation angle and analyzing the angle difference of the boundary normal, so that the consistency of the form in the subsequent processing process is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and deep learning algorithm optimization, and in particular to a method and system for processing images based on a fusion growing point optimization algorithm and Unet network. Background Art

[0002] The field of computer vision and deep learning algorithm optimization technology includes technical content for analyzing and processing image and video data based on computer algorithms and artificial intelligence methods. The core content of this technical field includes image feature extraction, pattern recognition, target detection and segmentation, etc., aiming to extract meaningful information from large-scale image data in an automated way. The systematic research of computer vision technology covers multi-level algorithm development, including classic algorithms such as edge detection and texture analysis, as well as model optimization methods based on deep learning that have developed rapidly in recent years, such as convolutional neural networks and transfer learning methods, which are widely used in medical image analysis, autonomous driving, security monitoring and other fields.

[0003] Among them, image processing based on fusion growth point optimization algorithm and Unet network refers to a method for processing orthopedic medical images using fusion optimization algorithm and deep learning network. This patent subject aims at the problem of high-precision regional segmentation that needs to be achieved in orthopedic images. It uses the fusion growth point optimization algorithm to improve the adaptability of the initial parameter setting, and combines the multi-scale feature extraction capability of the Unet network model to complete the specific regional segmentation processing of orthopedic images. By introducing the algorithm-based weight initialization and learning rate adjustment mechanism, combined with the cascade feature extraction method in the network, the effective separation and enhancement processing of regional features of orthopedic images can be achieved. This method completes the structured processing of image data by combining optimization strategies with deep learning models.

[0004] Although the existing technology has achieved good results in some basic algorithms when processing orthopedic images, due to the complexity of large-scale image data and the inaccuracy of feature extraction, it is often unable to make effective adjustments and corrections when faced with morphological deviations or irregular structures, resulting in certain errors in the segmentation results. In terms of initial parameter setting, the existing technology has poor adaptability to different orthopedic images and often relies on fixed parameters, which may lead to overfitting or underfitting in practical applications, affecting the final processing effect. Especially for images with strong surface features, the existing technology is insufficient in the extraction and optimization of surface details, and it is easy to have blurred or incoherent boundaries, which makes the final image segmentation unable to provide accurate positioning or recognition in some key areas. In addition, traditional methods rarely use dynamic adjustment mechanisms during the processing process, and cannot flexibly respond to different image characteristics. Therefore, it is difficult to operate stably in various complex image scenes, affecting the versatility and practicality of the overall system. Summary of the invention

[0005] To solve the technical problems existing in the prior art, an embodiment of the present invention provides an image processing method and system based on a fusion growth point optimization algorithm and a Unet network. The technical solution is as follows:

[0006] An image processing method based on a fusion growth point optimization algorithm and a Unet network includes the following steps: S1: Collect lumbar CT images, extract the gray value and HU value of each pixel in the images, mark the lumbar bone region to obtain the density information of the lumbar bone region, analyze the lumbar bone gradient and segment the initial region, obtain the basic region information points from it and classify them to generate the preliminary structural features of the lumbar bone; S2: Extract the central axis and boundary normal line in the image from the preliminary structural features of the lumbar bone, obtain the angular difference between the central axis and the boundary normal line and dynamically adjust the rotation angle of the image to correct the morphological deviation of the lumbar bone and generate a lumbar bone region after angle correction; S3: Analyze the surface characteristics of the lumbar bone region after angle correction through a differential geometry optimization algorithm, calculate the local curvature change according to the surface characteristics, identify the surface features on the lumbar bone surface, map the surface features to the Unet network and extract the surface boundary information, and optimize the smoothness and connectivity of the surface boundary to generate an optimized result of the surface features; S4: Identify the gradient change of each pixel point according to the optimized result of the surface features, mark the key points in the lumbar bone region, cluster the key points to distinguish the lumbar bone region from other tissue regions outside the lumbar bone region, determine the boundary of the lumbar bone region and segment it to generate a lumbar bone region segmentation boundary map; S5: Enhance the lumbar bone edge according to the lumbar bone region segmentation boundary map, adjust the segmentation region with reference to the basic region information points, fill the boundary gaps and the same type of lumbar bone features based on the adjusted segmentation region, and perform iterative correction and smoothing processing on the boundary to generate an optimized lumbar bone segmentation map.

[0007] The improvement of the present invention is that the preliminary structural features of the lumbar bone include corner points, connection points, bending points, the central axis of the lumbar bone region, and the boundary normal line of the lumbar bone region; the lumbar bone region after angle correction includes the lumbar bone region after rotation angle adjustment; the optimized result of the surface features includes the surface characteristics of the lumbar bone region and the surface boundary information; the lumbar bone region segmentation boundary map includes the refined lumbar bone region and the segmentation boundary; the optimized lumbar bone segmentation map includes clear edges, adjusted segmentation regions, extended segmentation regions, segmentation regions after noise removal, and boundaries after correction and smoothing processing.

[0008] The improvement of the present invention is that the specific steps of S1 are as follows: S101: Collect the lumbar CT images of the patient, extract the grayscale value and HU value of each pixel in the CT image, compare the pixel grayscale value with a preset grayscale threshold, mark the lumbar bone area where the grayscale threshold is higher than the preset grayscale threshold, screen out the area where the HU value exceeds the preset HU threshold from it and extract density data, and generate the density information of the marked lumbar bone area; S102: Analyze the gradient change rate of the marked lumbar bone area according to the density information of the marked lumbar bone area, refer to the lumbar bone area gradient information to segment the initial area of the marked grayscale threshold, obtain the basic area information points of the lumbar bone according to the segmented grayscale area, and generate the basic information points of the lumbar bone area; S103: Identify the corresponding corner points, connection points, and bending point positions in the basic information points of the lumbar bone area, determine the preliminary structure of the lumbar bone according to the position information of each point, and generate the preliminary structure characteristics of the lumbar bone.

[0009] The improvement of the present invention is that the specific steps of S2 are as follows: S201: Extract the central axis and boundary normal of the lumbar bone area from the preliminary structure characteristics of the lumbar bone, locate the contour of the lumbar bone, analyze the direction and position differences between the central axis and the boundary normal, and generate the central axis and boundary normal information; S202: Obtain the angle difference value between the central axis and the boundary normal from the central axis and boundary normal information, refer to the influence of the difference of each angle on the morphological change of the lumbar bone area, and generate the angle difference result; S203: Dynamically adjust the rotation angle of the lumbar bone image according to the angle difference result, use image processing technology to correct the rotation of the image, adjust the morphological deviation of the original shooting angle, mark the asymmetric area of the lumbar CT image, and generate the lumbar bone area after angle correction.

[0010] The improvement of the present invention is that the specific steps of S3 are as follows: S301: Conduct geometric analysis on the lumbar bone area after angle correction, obtain the surface curved surface information of the lumbar bone area, analyze the curved surface characteristics of the lumbar bone area through a differential geometry optimization algorithm, and the curved surface characteristics include curvature and curved surface shape, and generate the curved surface characteristic information of the lumbar bone area; S302: Calculate the local curvature value of the lumbar bone surface area according to the curved surface characteristic information of the lumbar bone area, mark the lumbar bone area that exceeds the preset curvature threshold, and generate the local curvature change information of the marked lumbar bone area; S303: Identify the curved surface characteristics on the lumbar bone surface based on the local curvature change information of the lumbar bone area, map the curved surface characteristics to the Unet network and extract the curved surface boundary information, and optimize the smoothness and connectivity of the curved surface boundary to generate the optimized result of the curved surface characteristics.

[0011] The improvement of the present invention is that for the local curvature value of the surface area of the lumbar vertebra, the formula is used: ; Calculate the local curvature value of the surface area of the lumbar vertebra ; Among them, , Are respectively the first curvature radius and the second curvature radius of the key points in the main curvature direction of the lumbar vertebra region, , Is the correction coefficient for adjusting the first curvature radius And the second curvature radius , Is the morphological flexibility factor, Is the curvature change amount, Is the change amount of the curvature angle.

[0012] The improvement of the present invention is that the specific steps of S4 are as follows: S401: According to the surface feature optimization result, use the Shi-Tomasi corner detection algorithm to identify the gradient change of each pixel point, analyze the change amplitude of adjacent pixels, compare the gray difference value of each adjacent pixel, and mark the key point area of the lumbar vertebra region according to the comparison result; S402: Use the K-means clustering algorithm to perform spatial clustering calculation on the key point area, identify the key points and group them according to the position and gradient difference, distinguish the lumbar vertebra region and other tissue regions outside the lumbar vertebra region according to the grouped key points, and generate the lumbar vertebra region classification result; S403: According to the lumbar vertebra region classification result, compare the gray difference and gradient information, divide the boundary of each lumbar vertebra region through morphological gradient, use the watershed algorithm to segment the boundary of the lumbar vertebra region, and refine the segmented lumbar vertebra region to generate the lumbar vertebra region segmentation boundary map.

[0013] The improvement of the present invention is that for the lumbar vertebra region classification result, the formula is used: ; Among them, Represents the weighted Euclidean distance between the key point And the clustering center , And Are respectively the coordinate values of the key point and the clustering center in the Dimension, And Are respectively the gradient information of the key point and the clustering center, Is the gradient weighting factor, Is the number of dimensions.

[0014] The improvement of the present invention is as follows. The specific steps of S5 are as follows: S501: According to the lumbar vertebra region segmentation boundary map, use the region growing method to optimize the clarity of the lumbar vertebra edge in the segmentation region, analyze the contour information of the segmentation region and identify the changes of adjacent pixels, determine the difference in the clarity of the lumbar vertebra edge, and smooth the edge according to the gradient change of adjacent pixels to generate an optimized edge clarity result; S502: According to the optimized edge clarity result, refer to the basic region information points, analyze the position and distribution of the basic region information points, and adjust the range of the lumbar vertebra segmentation region to fill the original boundary vacancy to generate an adjusted segmentation region; S503: According to the adjusted segmentation region, analyze the gray level difference of adjacent pixels and merge similar lumbar vertebra features, remove the noise of similar lumbar vertebra features, and use the morphological algorithm to perform iterative correction and smoothing processing on the boundary after expansion and merging to generate an optimized lumbar vertebra segmentation map.

[0015] Based on the fusion growth point optimization algorithm and the Unet network image processing system, the system includes: A data preprocessing module, which is used to collect lumbar CT images, extract the gray value and HU value of each pixel, mark the lumbar vertebra region and obtain the density information of the lumbar vertebra region, analyze the gray value to calculate the regional gradient, segment and mark the initial region, extract the basic region information points including corner points, connection points and bending points and classify them to generate the preliminary structural features of the lumbar vertebra; A feature extraction module, which is used to extract the central axis and boundary normal line in the image based on the preliminary structural features of the lumbar vertebra, obtain the angle difference between the central axis and the boundary normal line and dynamically adjust the image rotation angle to correct the morphological deviation of the original shooting angle to generate a lumbar vertebra region after angle correction; An angle correction module, which is used to analyze the surface characteristics of the region through the differential geometry optimization algorithm based on the lumbar vertebra region after angle correction, calculate the change of local curvature, identify the surface features and extract the surface boundary information of the lumbar vertebra region, and optimize the smoothness and connectivity of the surface boundary to generate an optimized surface feature result; A segmentation optimization module, which is used to identify the gradient change and mark the key points of the lumbar vertebra region based on the optimized surface feature result, classify the key points using the spatial clustering method, distinguish the lumbar vertebra from other tissue regions, identify the region boundary and perform segmentation to generate a lumbar vertebra region segmentation boundary map; A boundary correction module is used to optimize the edges of the segmented region based on the lumbar vertebra region segmentation boundary map, analyze the basic region information points and adjust their positions and distributions, expand and fill the boundary gaps in the segmented region, merge similar features and remove noise by judging the gray - level differences of adjacent pixels, and perform boundary correction and smoothing to generate an optimized lumbar vertebra segmentation map.

[0016] The beneficial effects brought by the technical solutions provided in the embodiments of the present invention at least include: In the present invention, by collecting lumbar CT images and extracting the gray - level values and HU values of pixels, the density information of the lumbar vertebra region is obtained and preliminary segmentation is performed. It can accurately analyze the gradient changes in this region, providing a reliable basis for the preliminary region of the image. During this process, by dynamically adjusting the image rotation angle and analyzing the difference in the boundary normal angle, the deviation of the lumbar vertebra morphology is successfully corrected, thus ensuring the consistency of the morphology in the subsequent processing. Applying the differential geometry optimization algorithm to further analyze the surface characteristics, identify and optimize the surface features of the lumbar vertebra, enabling the surface information to be more accurately mapped into the network, and then optimizing the surface boundary to ensure the boundary smoothness and connectivity. Through the analysis of gradient changes and key - point clustering, the segmentation of the lumbar vertebra region and other tissues can be refined, clearly defining the bone region, and thus enhancing the accuracy of the segmentation. Finally, through the iterative correction and smoothing of the segmented region, the optimized segmentation map can better reflect the characteristics of the lumbar vertebra, reducing the errors caused by inaccurate initial parameters. Compared with the prior art, this segmentation processing method, through the integration of optimization adjustment and deep learning, makes the segmentation result more accurate and stable. Especially in complex orthopedic images, it can effectively improve the adaptability and robustness of the processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0018] Figure 1 is the flowchart of the method of the present invention; Figure 2 is the schematic diagram of the refinement process of step S1 of the present invention; Figure 3 is the schematic diagram of the refinement process of step S2 of the present invention; Figure 4 is the schematic diagram of the refinement process of step S3 of the present invention; Figure 5 is the schematic diagram of the refinement process of step S4 of the present invention; Figure 6Schematic diagram of the refinement process of step S5 of the present invention; Figure 7 System module diagram of the present invention. Detailed implementation manners

[0019] The technical solutions in the present invention will be described below with reference to the accompanying drawings.

[0020] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as an "example" in the present invention should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of the word "example" is intended to present concepts in a specific manner. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one of the two.

[0021] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference is not emphasized, the meanings they express are the same. "(of)", "corresponding", and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference is not emphasized, the meanings they express are the same.

[0022] In the embodiments of the present invention, sometimes subscripts such as W 1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meanings they express are the same.

[0023] To make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0024] Please refer to Figure 1 , the embodiments of the present invention provide an image processing method based on a fusion growth point optimization algorithm and a Unet network, including the following steps: S1: Collect lumbar CT images, extract the gray value and HU value of each pixel in the images, mark the lumbar bone area to obtain the density information of the lumbar bone area, analyze the lumbar bone gradient and segment the initial area, obtain the basic area information points from it and classify them to generate the preliminary structural features of the lumbar bone; S2: Extract the central axis and boundary normal line in the image from the preliminary structural features of the lumbar bone, obtain the angular difference between the central axis and the boundary normal line and dynamically adjust the rotation angle of the image to correct the morphological deviation of the lumbar bone and generate the lumbar bone area after angle correction; S3: Analyze the surface characteristics of the lumbar bone region after angle correction through a differential geometry optimization algorithm, calculate the local curvature change according to the surface characteristics, identify the surface features on the lumbar bone surface, map the surface features to the Unet network and extract the surface boundary information, optimize the smoothness and connectivity of the surface boundary, and generate the optimized result of the surface features; S4: Identify the gradient change of each pixel point according to the optimized result of the surface features, mark the key points in the lumbar bone region, cluster the key points to distinguish the lumbar bone region from other tissue regions outside the lumbar bone region, determine the boundary of the lumbar bone region and segment it to generate the lumbar bone region segmentation boundary map; S5: Enhance the lumbar bone edge according to the lumbar bone region segmentation boundary map, adjust the segmentation region with reference to the basic region information points, fill the boundary gaps and the same type of lumbar bone features based on the adjusted segmentation region, perform iterative correction and smoothing processing on the boundary, and generate the optimized lumbar bone segmentation map; The preliminary structural features of the lumbar bone include corner points, connection points, bending points, the central axis of the lumbar bone region, and the boundary normal of the lumbar bone region. The lumbar bone region after angle correction includes the lumbar bone region after rotation angle adjustment; the optimized result of the surface features includes the surface characteristics and surface boundary information of the lumbar bone region. The lumbar bone region segmentation boundary map includes the refined lumbar bone region and the segmentation boundary. The optimized lumbar bone segmentation map includes clear edges, adjusted segmentation regions, extended segmentation regions, segmentation regions after noise removal, and boundaries after correction and smoothing processing.

[0025] Please refer to Figure 2 , and the specific steps of S1 are as follows:

[0026] S101: Collect the lumbar CT images of the patient, extract the gray value and HU value of each pixel in the CT image, compare the pixel gray value with the preset gray threshold, mark the lumbar bone region with a gray threshold higher than the preset gray threshold, screen the regions with HU values exceeding the preset HU threshold from them and extract the density data to generate the density information of the marked lumbar bone region; Obtain the lumbar CT images of the patient. The gray value and HU value of each pixel in the image need to be extracted through image processing software. At this time, a gray threshold needs to be set, usually setting the threshold range according to the CT device and the patient's body type. For example, if the gray threshold is set to 200, it means that only the pixels with gray values greater than 200 will be marked as the lumbar bone region. If the proportion of pixels with gray values greater than 200 in the image is relatively high, these regions may be the lumbar bone regions. Next, screen the regions with higher gray values and further screen them according to the HU value. The HU value generally ranges from 100 to 1000, and the HU value of the lumbar bone usually exceeds 250. Through this process, it can be ensured that only the regions with higher density (i.e., the lumbar bone regions) will be marked and their density data will be extracted.

[0027] S102: Analyze the gradient change rate of the marked lumbar vertebra region based on the density information of the marked lumbar vertebra region, refer to the gradient information of the lumbar vertebra region to segment the initial region of the marked grayscale threshold, obtain the basic region information points of the lumbar vertebra according to the segmented grayscale region, and generate the basic information points of the lumbar vertebra region; Perform a gradient change rate analysis on the density information of the marked lumbar vertebra region. The gradient change rate can be obtained by calculating the change rate of the grayscale value of each region. The positions with larger gradients are usually where the boundaries or structures change significantly. For example, when the grayscale value of a part of the region jumps from a lower value to a higher value, it indicates that there is a significant change in the structure in this region. Assume that the set grayscale threshold is 200. If the grayscale value of adjacent regions jumps from below 200 to above 200, it means that there are significant density changes in these regions, so they can be determined as the boundaries of the lumbar vertebra. According to these gradient change rates, further segment the marked region to accurately delimit the range of the lumbar vertebra. Then, extract the basic information points from the segmented region, for example, determine the position of the basic points through geometric features such as angles, edges, and central positions.

[0028] S103: Identify the corresponding corner points, connection points, and bending point positions in the basic information points of the lumbar vertebra region, determine the preliminary structure of the lumbar vertebra according to the position information of each point, and generate the preliminary structure features of the lumbar vertebra; Identify the accurate positions of the corner points, connection points, and bending points based on the basic information points. Corner points usually appear at the bends, turns, or joints of the bones, and the positions of these points are calibrated by detecting regions with large grayscale changes or edge intensities. For example, if the grayscale value of a certain part of the bone changes rapidly, it indicates that this region is a corner point. The connection point is the junction of two different regions, usually with a small and smooth grayscale change, indicating that this point may be a connection point. The bending point refers to the turning point of the bone in the curved part, usually with a large grayscale value change and a significant change in direction. In actual operation, through image processing techniques such as edge detection, corner detection, and region segmentation, these key points are extracted, and the position of each point in the lumbar vertebra is determined by calculating its geometric features (such as coordinate position, relative distance), and finally the preliminary structure features of the lumbar vertebra are generated.

[0029] Please refer to Figure 3 , the specific steps of S2 are as follows:

[0030] S201: Extract the central axis and boundary normal of the lumbar vertebra region from the preliminary structure features of the lumbar vertebra, locate the contour of the lumbar vertebra, analyze the direction and position differences between the central axis and the boundary normal, and generate the central axis and boundary normal information; Analyze the imaging data of lumbar vertebrae. Use skeleton extraction or image segmentation techniques to identify the contour of the lumbar vertebrae from the image. Extract the central axis of the lumbar vertebrae from the image. The central axis is generally a straight line or curve connecting key regions in the lumbar vertebrae, ensuring the capture of the longitudinal structure of the lumbar spine. For the boundary normal, it is necessary to calculate the direction line perpendicular to the boundary based on the edge part of the lumbar vertebrae. These normals can be obtained by calculating the gradient direction of each pixel point. A large gradient value usually corresponds to the edge of the structure, and the boundary normal is determined through this information. When analyzing the direction difference between the central axis and the boundary normal, by comparing the angular difference between the two, the geometric changes of the lumbar vertebrae can be judged.

[0031] S202: Obtain the angular difference value between the central axis and the boundary normal from the central axis and boundary normal information. Refer to the morphological change impact on the lumbar vertebrae region for each angular difference analysis, and generate the angular difference result; Determine its angular difference by calculating the angle between the central axis and the boundary normal. For this purpose, it is necessary to extract the normal direction between each pair of key points and use the vector dot product formula to calculate the angular difference. Each pair of angular difference values reflects the morphological changes of the lumbar vertebrae around that point. If the angular difference is large, it indicates that there may be significant morphological variations or asymmetries in this area. Then, by setting an angular threshold to judge whether these differences are significant. For example, set an angular difference threshold of 10 degrees. When the angular difference between the two normals is greater than this value, it is considered that the morphology of the lumbar vertebrae at this part has changed significantly. By analyzing the angular difference result, the possible diseased areas, bending degree, or abnormally posed parts in the lumbar vertebrae can be further identified.

[0032] S203: Dynamically adjust the rotation angle of the lumbar vertebrae image according to the angular difference result. Use image processing techniques to perform rotation correction on the image, adjust the morphological deviation of the original shooting angle, mark the asymmetric regions of the lumbar CT image, and generate the lumbar vertebrae region after angle correction; According to the angular difference result, calculate the rotation angle that needs to be performed. For example, if the angular difference is greater than the set threshold (such as 10 degrees), then image rotation adjustment is required. The calculation of the rotation angle is usually based on the coordinate axes of the image data, and a rotation matrix is used to perform rotation correction on the image. In this process, first determine the rotation center of the image and calculate the rotation angle. After the rotation adjustment, the asymmetric regions in the image can be detected by morphological analysis methods. Through image processing techniques such as threshold segmentation and edge detection, mark the regions that are inconsistent with the standard morphology after rotation correction. These asymmetric regions may be due to the angle problem during image acquisition or the irregular natural morphology of the lumbar spine.

[0033] Please refer to Figure 4 , the specific steps of S3 are as follows:

[0034] S301: Perform geometric analysis on the lumbar bone region after angle correction to obtain the surface curvature information of the lumbar bone region. Analyze the surface curvature characteristics of the lumbar bone region through a differential geometry optimization algorithm. The surface curvature characteristics include curvature and surface shape, and generate the surface curvature characteristic information of the lumbar bone region; Performing geometric analysis on the lumbar bone region after angle correction usually involves using 3D modeling techniques to extract the surface curvature information of the lumbar vertebra from the imaging data. These curvature information helps to determine the three-dimensional morphology of the lumbar bone for subsequent differential geometry analysis. Through a differential geometry optimization algorithm, various characteristics of the surface can be extracted, including curvature and surface shape. The calculation method of curvature is usually based on the local characteristics of the surface, such as the normal direction, tangent plane, and changes in adjacent points on the surface, and is calculated through the gradient descent method. The surface curvature characteristics can provide important information about the morphological changes of the lumbar bone, especially in cases of lesion areas, degenerative diseases, or abnormal postures. Through calculation and analysis, the surface curvature characteristic information of the lumbar bone region can be generated.

[0035] S302: Calculate the local curvature value of the lumbar bone surface region according to the surface curvature characteristic information of the lumbar bone region, mark the lumbar bone regions that exceed the preset curvature threshold, and generate the local curvature change information of the marked lumbar bone regions; Based on the surface curvature characteristic information of the lumbar bone region, according to the formula: ; Calculate the local curvature value of the lumbar bone surface region ; where, and are the first curvature radius and the second curvature radius of the key points in the main curvature direction of the lumbar bone region respectively, and are the correction coefficients for adjusting the first curvature radius and the second curvature radius respectively. Usually, the setting ranges of and are [0, 1]. = = 1 is the default setting. When there are no special requirements, this value is used to indicate that the curvature radius is not corrected and the original curvature value is directly used. When and are adjusted respectively, in some regions with complex structures, such as the joint part of the lumbar vertebra, the curvature may be significantly different in two directions. At this time, different correction coefficients can be set for the two main curvature radii, is the morphological flexibility factor, The setting range is from 0 to 1. Larger values (e.g., 0.5 to 1) are usually applied to the boundary analysis of the lumbar region to ensure that the curvature can obtain sufficient weight in classification. By experimenting with the curvature values of different regions, the value can be selected according to the characteristics of the lumbar vertebra. For example, if the curvature change in the L2 region is large, setting = 0.7 can help accurately delimit the boundary and optimize the segmentation. is the amount of curvature change, is the amount of change in the curvature angle, (local curvature value) is the degree of curvature of the surface at a certain point, obtained by calculating the curvature of the point in two directions, usually using geometric analysis techniques or surface analysis methods based on differential geometry. The curvature value can be obtained through image analysis tools. For example, in 3D image processing software, it can be obtained by surface fitting or directly extracting surface features. (principal curvature radius 1) is the curvature radius of the point in the first principal curvature direction. The curvature radius is a measure of the surface curvature. The smaller the curvature radius, the greater the curvature of the surface at that point. It can be obtained by fitting the local surface or using the Gaussian curvature calculation method, and calculated by observing the curvature change around the point. For example, it can be obtained through the surface fitting function of image processing or scanning equipment. (principal curvature radius 2) is the curvature radius of the point in the second principal curvature direction, similar to , but in a different direction from and is usually obtained by quadratic curve fitting of the local surface. represents the curvature in the direction perpendicular to . and (correction coefficient) is used to correct the curvature radius to improve the accuracy of the model, determined by experimenting with the actual data of the local area. In specific cases, such as when the shape of the lumbar vertebra is relatively complex, these coefficients can be appropriately adjusted to reflect more accurate surface features. (morphological flexibility factor) is a factor used to describe the morphological change of the surface, mainly adjusted by analyzing the degree of curvature and shape change of the surface. It can be obtained through statistical analysis. For example, in the study of the lumbar vertebra region, this value can be set according to the statistical results of various patient data to reflect the overall trend. (amount of curvature change) is the change amount of curvature in a certain direction, usually obtained by calculating the curvature difference between adjacent points or adjacent regions. For example, it can be obtained through differential calculation of a specific region or through the change of image data. (Change in curvature angle) refers to the change in angle in the direction of curvature, which can be obtained by measuring the angle change on a curve or surface. It reflects the turning changes on the surface of the lumbar bone region at different positions. This value is usually obtained through angle calculation in image analysis, especially during the process of surface reconstruction.

[0036] For example, in the analysis of lumbar spine CT images, the extracted parameter values are as follows: - mm, mm, , , , mm, . Based on these parameters, the local curvature value is calculated by the formula as follows: ; The results show that the surface change at this point is relatively large, and by introducing a correction coefficient and a morphological flexibility factor, it can better adapt to the complex surface characteristics of the lumbar bone, providing a more accurate reference for further morphological analysis and treatment.

[0037] S303: Identify the surface features of the lumbar bone based on the local curvature change information of the lumbar bone region, map the surface features to the Unet network and extract the surface boundary information, optimize the smoothness and connectivity of the surface boundary, and generate the optimized result of the surface features; Extract the surface features of the lumbar bone based on the local curvature change information. Through the previously calculated local curvature values, identify the surface features of different regions on the lumbar bone surface, and these regions usually represent the key structural parts of the lumbar bone. For example, in the L2 region of the lumbar bone, the local curvature value is , showing a significant curvature change in this region, suggesting a possible risk of fracture or deformation. Then, map these surface features to the Unet network, and use the powerful image segmentation and feature extraction capabilities of this network to extract accurate surface boundary information. Through the Unet network, the boundary of the lumbar bone region can be accurately identified in the image, and further segment the regions with different curvature changes. After extracting the surface boundary, further optimize the smoothness and connectivity of the boundary, eliminate unnecessary noise and breaks, and ensure the smoothness and integrity of the boundary. For example, for the boundary of the L3 region, after optimization, the smoothness of the surface is improved from the initial curvature value to , and the connectivity is improved, avoiding data loss or discontinuity.

[0038] Please refer to Figure 5 , and the specific steps of S4 are as follows:

[0039] S401: Optimize the results according to the surface features, use the Shi-Tomasi corner detection algorithm to identify the gradient changes of each pixel point, analyze the change amplitude of adjacent pixels, compare the gray difference values of each adjacent pixel, and mark the key point area of the lumbar bone region according to the comparison results; Based on the optimized results of the surface features, it is processed by the Shi-Tomasi corner detection algorithm, which uses the gradient changes between pixel points to identify the corners in the image. During the corner detection process, the gradient of each pixel will be calculated, usually by calculating the gray difference in the surrounding area of the pixel to measure the change amplitude of the gradient. After detecting the area with large gradient changes, further analyze the gray difference of adjacent pixels and compare it with the set threshold. Through the comparison of the gray differences, the lumbar bone region with significant morphological features can be marked, and then the key point area can be extracted.

[0040] S402: Use the K-means clustering algorithm to perform spatial clustering calculations on the key point area, identify the key points and group them according to the position and gradient differences, distinguish the lumbar bone region and other tissue regions outside the lumbar bone region according to the grouped key points, and generate the classification result of the lumbar bone region; Referring to the classification result of the lumbar bone region, according to the formula: ; where, represents the weighted Euclidean distance between the key point and the clustering center , , are the coordinate values of the key point and the clustering center in the th dimension respectively, , are the gradient information of the key point and the clustering center respectively, is the gradient weighting factor, usually ranges from 0 to 100. Larger values (such as 10 to 50) are usually used to process areas with complex morphology and large gradient changes, while smaller values (such as 0.1 to 1) are suitable for areas with relatively smooth morphology and less gradient changes. According to the complexity and regional characteristics of the lumbar spine image, value can be determined through experiments. For example, in the L3 region of the lumbar bone, it is observed that in the lumbar region within the normal curvature range, when = 10, the surface boundary can be well separated from other soft tissues, so this value is selected, is the number of dimensions, represents the key point and the clustering center The distance between them is used to assign key points to the nearest cluster center. The smaller the distance, the more likely the key point belongs to that cluster center. This calculation is based on an extension of the Euclidean distance, including position differences and gradient differences. and represent the key point and the cluster center in the -dimensional space. For lumbar spine CT images in 3D space, usually represents one of the three coordinate axes (X, Y, Z). For example, is the coordinate of the key point on the X-axis, is the coordinate on the Y-axis, is the coordinate on the Z-axis. These coordinates are usually directly obtained through image processing or scanning devices. and represent the gradient information of the key point and the cluster center respectively. The gradient information is reflected by the gray-level change of the image, indicating the intensity change at a certain point. The larger the gradient value, the more obvious the change around that point, usually meaning the edge or the turning point of the structure. These gradient information can be calculated by image processing software (such as OpenCV) and the gray-level change of the local area can be extracted using gradient calculation algorithms (such as Sobel or Prewitt operators). is the gradient weighting factor, which is used to control the influence of gradient information in calculating the distance. A larger value will enhance the role of gradient information in the clustering process, making the clustering pay more attention to the areas with large gray-level changes. This value is determined through experiments, for example, by testing different values, evaluating the effect of the clustering results and selecting the best value. For example, in lumbar CT images, the selected may better emphasize the boundary area. represents the number of dimensions of the space. For 3D image data, , that is, the coordinates of each point are composed of three-dimensional coordinate values of X, Y, and Z. In 2D images, ; in 3D images, .

[0041] Through the K-means algorithm, several cluster centers need to be initialized. By calculating the weighted Euclidean distance between each key point and the cluster center, the cluster membership of each key point is determined. For example, there is a key point with coordinates (5, 4, 3) on the CT image, and its gradient value is , while the coordinates of the cluster center are (4, 3, 2), and the gradient value is If is set, then calculate the weighted distance between this point and the cluster center: ; Through this calculation method, the K-means algorithm will assign this key point to the cluster center with the minimum distance. This process will be repeated multiple times until the update of the cluster center converges.

[0042] The results show that the clustering results calculated by the K-means clustering algorithm can accurately distinguish the lumbar vertebra region from other regions according to the position and gradient information of the key points.

[0043] S403: According to the lumbar vertebra region classification results, compare the gray level difference and gradient information, divide the boundaries of each lumbar vertebra region through morphological gradient, use the watershed algorithm to segment the boundaries of the lumbar vertebra regions, refine the segmented lumbar vertebra regions, and generate a lumbar vertebra region segmentation boundary map; Based on the classification results, compare the gray level differences and gradient information between different regions. By comparing these differences, the boundaries of the lumbar vertebra regions can be clearly identified. Then, use the morphological gradient operator to extract the boundaries of the lumbar vertebrae. Morphological gradient helps to enhance the edge information, making the boundaries more obvious. Through morphological operations, the boundary positions of the lumbar vertebra regions are further clarified. Then, use the watershed algorithm to segment the extracted boundaries. The watershed algorithm divides different regions in the image by analyzing the changes in gray level and gradient information, so as to distinguish the lumbar vertebra regions from the surrounding tissue regions. The key of the watershed algorithm lies in its ability to process complex images, which can effectively perform region division in complex shapes and variable boundaries. After segmentation, refine the segmentation results to eliminate small noise and breaks, ensuring the continuity and accuracy of the boundaries. Finally, the generated lumbar vertebra region segmentation boundary map can provide high-quality data support for subsequent lumbar vertebra analysis, disease detection and treatment.

[0044] Please refer to Figure 6 for the specific steps of S5 as follows:

[0045] S501: According to the lumbar vertebra region segmentation boundary map, use the region growing method to optimize the clarity of the lumbar vertebra edges in the segmented region, analyze the contour information of the segmented region and identify the changes of adjacent pixels, determine the differences in the clarity of the lumbar vertebra edges, and smooth the edges according to the gradient changes of adjacent pixels to generate an optimized edge clarity result; Combining the fusion growth point optimization algorithm with Unet network image processing, select appropriate seed points through the region growing method and gradually expand the region, thereby optimizing the segmentation boundary of the lumbar vertebra region. The growth point algorithm can accurately detect the edges of the lumbar vertebrae with higher gray levels and eliminate the boundary blurring caused by segmentation by continuously expanding the coverage area. The sharpness of the edge is related to the gray level difference of adjacent pixels. Therefore, by analyzing the gray level changes of adjacent pixels, it can be determined which region edges are clearer and which region edges are blurred. After identifying the differences, the algorithm will perform smoothing processing on the blurred boundaries to remove noise and enhance the contour features of the region.

[0046] S502: According to the optimized edge sharpness result, refer to the basic region information points, analyze the position and distribution of the basic region information points, and adjust the range of the lumbar vertebra segmentation region, fill the original boundary gaps, and generate the adjusted segmentation region; Based on the optimized edge sharpness, combined with the fusion growth point optimization algorithm, analyze the distribution of the basic region information points to clarify the key regions of the lumbar vertebrae. By identifying the positions of these basic information points, it can help determine the main structural boundaries of the lumbar vertebrae and further adjust the range of the segmentation region based on this. For example, if it is found that some parts of the original segmentation region are not fully covered or there are gaps, these gaps can be filled through the adjustment algorithm to ensure that all regions of the lumbar vertebrae are accurately identified and segmented. This adjustment process will continuously correct and improve the segmentation result, making the lumbar vertebra region more complete and coherent.

[0047] S503: Analyze the gray level differences of adjacent pixels according to the adjusted segmentation region and merge similar lumbar vertebra features, remove the noise of similar lumbar vertebra features, and use morphological algorithms for iterative correction and smoothing processing of the boundary after expansion and merging to generate an optimized lumbar vertebra segmentation map; Analyze the gray level differences of adjacent pixels in the adjusted lumbar vertebra segmentation region. Through this analysis, the obvious differences between the lumbar vertebra region and the non-bone region can be identified, and at the same time, some noise features that do not belong to the lumbar vertebrae can be removed. Merge similar lumbar vertebra features to ensure the consistency of the classification results. Next, use morphological algorithms to further expand and merge these regions, smooth the boundaries between regions, and remove unnecessary cracks or noise points. On this basis, through iterative correction and smoothing processing, the segmentation of the lumbar vertebra region is refined to ensure the high quality of the segmentation result. Finally, the generated optimized segmentation map can more accurately reflect the actual shape of the lumbar vertebrae and provide clearer image data for subsequent analysis, diagnosis, and treatment.

[0048] Please refer to Figure 7 , based on the fusion growth point optimization algorithm and Unet network image processing system, the system includes: A data preprocessing module, which is used to collect lumbar CT images, extract the gray value and HU value of each pixel, mark the lumbar bone area and obtain the density information of the lumbar bone area, analyze the gray value to calculate the regional gradient, segment and mark the initial area, extract the basic area information points including corner points, connection points and bending points and classify them, and generate the preliminary structural features of the lumbar bone; A feature extraction module, which is used to extract the central axis and boundary normal line in the image based on the preliminary structural features of the lumbar bone, obtain the angular difference between the central axis and the boundary normal line and dynamically adjust the image rotation angle, correct the morphological deviation of the original shooting angle, and generate the lumbar bone area after angle correction; An angle correction module, which is used to analyze the surface characteristics of the area through a differential geometry optimization algorithm based on the lumbar bone area after angle correction, calculate the change of local curvature, identify the surface features and extract the surface boundary information of the lumbar bone area, optimize the smoothness and connectivity of the surface boundary, and generate the optimized result of the surface features; A segmentation optimization module, which is used to identify the gradient change and mark the key points of the lumbar bone area based on the optimized result of the surface features, classify the key points using a spatial clustering method, distinguish the lumbar bone from other tissue areas, identify the regional boundary and segment it, and generate the segmentation boundary map of the lumbar bone area; A boundary correction module, which is used to optimize the edge of the segmented area based on the segmentation boundary map of the lumbar bone area, analyze the basic area information points and adjust their positions and distributions, expand and fill the boundary gaps in the segmented area, merge similar features and remove noise by judging the gray difference of adjacent pixels, and perform boundary correction and smoothing to generate the optimized lumbar bone segmentation map.

[0049] As mentioned above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. Based on the fusion growth point optimization algorithm and Unet network image processing method, it is characterized by: The following steps are involved: S1: Collect lumbar CT images, extract the grayscale value and HU value of each pixel in the image, mark the lumbar vertebrae area to obtain the lumbar vertebrae area density information, analyze the lumbar vertebrae gradient and segment the initial area, obtain the basic area information points and classify them, and generate the preliminary structural features of the lumbar vertebrae; S2: extracting the central axis and boundary normal in the image from the preliminary structural features of the lumbar vertebrae, obtaining the angle difference between the central axis and the boundary normal, and dynamically adjusting the rotation angle of the image to correct the lumbar vertebrae morphological deviation, and generating an angle-corrected lumbar vertebrae region; S3: Analyzing the surface characteristics of the angle-corrected lumbar vertebrae region by a differential geometry optimization algorithm, calculating the local curvature change according to the surface characteristics, identifying the surface features of the lumbar vertebrae surface, mapping the surface features to a Unet network and extracting the surface boundary information, optimizing the smoothness and connectivity of the surface boundary, and generating a surface feature optimization result; S4: identifying the gradient change of each pixel point according to the surface feature optimization result, marking the key points of the lumbar vertebrae region, clustering the key points to distinguish the lumbar vertebrae region from other tissue regions outside the lumbar vertebrae region, determining the boundary of the lumbar vertebrae region and segmenting it, and generating a segmentation boundary map of the lumbar vertebrae region; S5: enhancing the edge of the lumbar vertebrae according to the lumbar vertebrae region segmentation boundary map, adjusting the segmentation region with reference to the basic region information points, filling the boundary gaps and similar lumbar vertebrae features based on the adjusted segmentation region, iteratively correcting and smoothing the boundary, and generating an optimized lumbar vertebrae segmentation map.

2. The method for processing network images based on the fusion growth point optimization algorithm and Unet according to claim 1 is characterized in that: The preliminary structural features of the lumbar vertebrae include corner points, connection points, bending points, the central axis of the lumbar vertebrae area, and the boundary normal of the lumbar vertebrae area. The lumbar vertebrae area after angle correction includes the lumbar vertebrae area after rotation angle adjustment. The surface feature optimization results include surface characteristics and surface boundary information of the lumbar vertebrae area. The lumbar vertebrae area segmentation boundary map includes the refined lumbar vertebrae area and the segmentation boundary. The optimized lumbar vertebrae segmentation map includes clear edges, adjusted segmentation areas, expanded segmentation areas, segmentation areas after noise removal, and corrected and smoothed boundaries.

3. The method for processing network images based on the fusion growth point optimization algorithm and Unet according to claim 1 is characterized in that: The specific steps of S1 are as follows: S101: collecting lumbar CT images of the patient, extracting the grayscale value and HU value of each pixel in the CT image, comparing the pixel grayscale value with a preset grayscale threshold, marking the lumbar vertebrae bone area whose grayscale threshold is higher than the preset grayscale threshold, screening the area whose HU value exceeds the preset HU threshold and extracting density data, and generating density information of the marked lumbar vertebrae bone area; S102: analyzing the gradient change rate of the marked lumbar vertebrae region according to the density information of the marked lumbar vertebrae region, segmenting the initial region of the marked grayscale threshold with reference to the gradient information of the lumbar vertebrae region, obtaining basic region information points of the lumbar vertebrae according to the segmented grayscale region, and generating basic region information points of the lumbar vertebrae; S103: Identify the positions of the corresponding corner points, connection points, and bending points in the basic information points of the lumbar vertebrae region, determine the preliminary structure of the lumbar vertebrae according to the position information of each point, and generate preliminary structural features of the lumbar vertebrae.

4. The method for processing network images based on the fusion growth point optimization algorithm and Unet according to claim 1 is characterized in that: The specific steps of S2 are as follows: S201: extracting the central axis and boundary normal of the lumbar vertebra region from the preliminary structural features of the lumbar vertebra, locating the contour of the lumbar vertebra, analyzing the direction and position difference of the central axis and the boundary normal, and generating the central axis and boundary normal information; S202: Obtaining the angle difference between the central axis and the boundary normal from the central axis and boundary normal information, and generating an angle difference result by referring to the difference analysis of each angle and the influence of the morphological change of the lumbar vertebrae region; S203: Dynamically adjust the rotation angle of the lumbar vertebrae image according to the angle difference result, use image processing technology to perform rotation correction on the image, adjust the morphological deviation of the original shooting angle, mark the asymmetric area of ​​the lumbar vertebrae CT image, and generate an angle-corrected lumbar vertebrae area.

5. The method for processing network images based on the fusion growth point optimization algorithm and Unet according to claim 1 is characterized in that: The specific steps of S3 are as follows: S301: performing geometric analysis on the angle-corrected lumbar vertebrae region to obtain surface information of the lumbar vertebrae region, analyzing surface features of the lumbar vertebrae region using a differential geometry optimization algorithm, the surface features including curvature and surface shape, and generating surface feature information of the lumbar vertebrae region; S302: Calculating the local curvature value of the lumbar vertebrae surface area according to the surface feature information of the lumbar vertebrae area, marking the lumbar vertebrae area exceeding a preset curvature threshold, and generating local curvature change information of the marked lumbar vertebrae area; S303: Identify the surface features of the lumbar vertebrae surface based on the local curvature change information of the lumbar vertebrae region, map the surface features onto the Unet network and extract the surface boundary information, optimize the smoothness and connectivity of the surface boundary, and generate a surface feature optimization result.

6. The method for processing network images based on the fusion growth point optimization algorithm and Unet according to claim 5 is characterized in that: For the local curvature value of the lumbar vertebral surface area, the formula is used: ； Calculate the local curvature value of the lumbar vertebral surface area ; in, , are the first and second curvature radii of the key points in the lumbar vertebrae region in the direction of the main curvature, , To adjust the first curvature radius and the second radius of curvature The correction factor of is the morphological flexibility factor, is the curvature change, is the change in curvature angle.

7. The method for processing network images based on the fusion growth point optimization algorithm and Unet according to claim 1 is characterized in that: The specific steps of S4 are as follows: S401: using the Shi-Tomasi corner detection algorithm according to the surface feature optimization result to identify the gradient change of each pixel, analyzing the change amplitude of adjacent pixels, comparing the grayscale difference value of each adjacent pixel, and marking the key point area of ​​the lumbar vertebra area according to the comparison result; S402: performing spatial clustering calculation on the key point area using a K-means clustering algorithm, identifying key points and grouping them according to position and gradient differences, distinguishing the lumbar vertebrae area and other tissue areas outside the lumbar vertebrae area according to the grouped key points, and generating a lumbar vertebrae area classification result; S403: Compare the grayscale difference and gradient information according to the classification result of the lumbar vertebrae region, divide the boundary of each lumbar vertebrae region by morphological gradient, segment the boundary of the lumbar vertebrae region using the watershed algorithm, refine the segmented lumbar vertebrae region, and generate a lumbar vertebrae region segmentation boundary map.

8. The method for processing network images based on the fusion growth point optimization algorithm and Unet according to claim 7 is characterized in that: For the lumbar vertebrae area classification results, the formula is used: ； in, Indicates key points With cluster center The weighted Euclidean distance between and are the key points and cluster centers in the The coordinate value of the dimension, and are the gradient information of key points and cluster centers respectively, is the gradient weighting factor, is the number of dimensions.

9. The method for processing network images based on the fusion growth point optimization algorithm and Unet according to claim 1, characterized in that: The specific steps of S5 are as follows: S501: optimizing the clarity of the lumbar vertebrae edge in the segmented region using a region growing method according to the lumbar vertebrae region segmentation boundary map, analyzing the contour information of the segmented region and identifying the changes of adjacent pixels, determining the difference in the clarity of the lumbar vertebrae edge, smoothing the edge according to the gradient changes of adjacent pixels, and generating an optimized edge clarity result; S502: Analyze the location and distribution of the basic region information points according to the edge definition optimization result and refer to the basic region information points, and adjust the range of the lumbar vertebra segmentation region to fill the original boundary gaps and generate an adjusted segmentation region; S503: Analyze the grayscale difference of adjacent pixels according to the adjusted segmented area and merge similar lumbar vertebrae features, remove similar lumbar vertebrae feature noise, use morphological algorithm to expand and merge, perform iterative correction and smoothing on the boundary, and generate an optimized lumbar vertebrae segmentation map.

10. Based on the fusion growth point optimization algorithm and Unet network image processing system, it is characterized by: According to any one of claims 1 to 9, the system is executed based on the fusion growth point optimization algorithm and the Unet network image processing method, comprising: The data preprocessing module is used to collect lumbar CT images, extract the gray value and HU value of each pixel, mark the lumbar vertebrae area and obtain the lumbar vertebrae area density information, analyze the gray value to calculate the regional gradient, segment and mark the initial area, extract and classify the basic area information points including corner points, connection points, and bending points, and generate the preliminary structural features of the lumbar vertebrae; A feature extraction module is used to extract the central axis and the boundary normal in the image based on the preliminary structural features of the lumbar vertebrae, obtain the angle difference between the central axis and the boundary normal and dynamically adjust the image rotation angle, correct the morphological deviation of the original shooting angle, and generate a lumbar vertebrae area after angle correction; An angle correction module is used to analyze the surface characteristics of the area based on the angle-corrected lumbar vertebra area through a differential geometry optimization algorithm, calculate the change of local curvature, identify surface features and extract surface boundary information of the lumbar vertebra area, optimize the smoothness and connectivity of the surface boundary, and generate a surface feature optimization result; A segmentation optimization module, which is used to identify gradient changes and mark key points of the lumbar vertebrae region based on the surface feature optimization result, classify the key points using a spatial clustering method, distinguish the lumbar vertebrae from other tissue regions, identify region boundaries and perform segmentation, and generate a lumbar vertebrae region segmentation boundary map; The boundary correction module is used to optimize the edge of the segmented area based on the lumbar vertebrae area segmentation boundary map, analyze the basic area information points and adjust their position and distribution, expand and fill the boundary gaps in the segmented area, merge similar features and remove noise by judging the grayscale differences of adjacent pixels, correct and smooth the boundaries, and generate an optimized lumbar vertebrae segmentation map.

Images