Medical image cerebral vessel segmentation and reconstruction system and optimization method
By optimizing cerebral blood vessel segmentation and reconstruction through an improved U-Net model, adaptive filtering, and reconstruction algorithm, the problems of time-consuming, labor-intensive, and insufficiently accurate methods in traditional methods are solved, achieving efficient and accurate cerebral blood vessel image processing and improving the reliability and safety of diagnosis and treatment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANKAI UNIV
- Filing Date
- 2026-03-10
- Publication Date
- 2026-05-29
AI Technical Summary
Traditional medical imaging methods for segmenting cerebrovascular vessels are time-consuming, labor-intensive, and produce inconsistent results. The three-dimensional reconstruction models are coarse and have incomplete topological structures. They also lack standardized preprocessing procedures and closed-loop accuracy assessments, resulting in insufficient diagnostic efficiency and treatment safety for cerebrovascular diseases.
An improved U-Net model combined with an attention mechanism is used for segmentation, adaptive median filtering and histogram equalization are used for preprocessing, the moving cube algorithm is used for 3D reconstruction, and a high-quality 3D model is generated through Laplacian smoothing and topology optimization. The model is trained and evaluated by combining Dice coefficient and cross-entropy loss function.
It significantly improves the accuracy of cerebral blood vessel segmentation and the quality of 3D models, reduces the risk of misdiagnosis and missed diagnosis, provides reliable diagnostic evidence, improves diagnostic efficiency and surgical safety, and supports applications in multiple scenarios.
Smart Images
Figure CN122115478A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to cerebral blood vessels, specifically to a medical imaging system for cerebral blood vessel segmentation and reconstruction, and its optimization method. Background Technology
[0002] In the clinical diagnosis and treatment and medical research of cerebrovascular diseases, medical imaging (such as CT, MRI, DSA) is the core basis for observing the morphology of cerebral blood vessels and judging lesions. However, cerebral blood vessels are delicate, have complex branches, and have little difference in gray scale from surrounding tissues, which limits the effectiveness of traditional image processing methods.
[0003] Traditional cerebral vascular segmentation relies heavily on manual or semi-automated tools. Manual segmentation requires doctors to annotate images frame by frame, which is time-consuming and laborious, and the results are greatly affected by experience, with poor consistency among different doctors. Semi-automated methods often fail to detect small branches or misjudge lesion areas due to noise interference and blurred vascular boundaries, making it difficult to meet the needs of precise diagnosis and treatment.
[0004] In the 3D reconstruction process, traditional techniques are prone to problems such as rough model surfaces and incomplete topological structures. For example, the reconstructed model may have jagged protrusions or mesh gaps, making it impossible to clearly present the spatial course and branching relationship of blood vessels. This makes it difficult for doctors to formulate precise surgical plans based on the model, increasing surgical risks.
[0005] Furthermore, existing methods lack standardized preprocessing procedures, and the image data formats and quality vary greatly among different devices and patients, directly affecting the accuracy of subsequent segmentation and reconstruction. Moreover, most methods lack closed-loop accuracy assessment mechanisms, failing to correct errors in a timely manner, resulting in insufficient reliability of the output results. These problems restrict the efficiency of cerebrovascular disease diagnosis and the safety of treatment, urgently requiring breakthroughs in more efficient and precise image processing technologies. Summary of the Invention
[0006] The technical problem to be solved by the present invention is to overcome the defects of the above-mentioned technologies and provide a medical imaging cerebral blood vessel segmentation and reconstruction system and an optimized method.
[0007] To solve the above-mentioned technical problems, the present invention provides a medical imaging cerebral blood vessel segmentation and reconstruction system and an optimization method:
[0008] Medical imaging cerebral vascular segmentation and reconstruction system, including:
[0009] The medical image acquisition module is used to acquire medical image data including cerebral blood vessels;
[0010] A preprocessing module, connected to the medical image acquisition module, is used to preprocess the medical image data, including noise reduction, contrast enhancement, and image standardization.
[0011] A cerebral blood vessel segmentation module, connected to the preprocessing module, is used to segment cerebral blood vessels in the preprocessed medical image data based on a deep learning model to obtain cerebral blood vessel segmentation results.
[0012] A 3D reconstruction module, connected to the cerebral blood vessel segmentation module, is used to perform 3D modeling based on the cerebral blood vessel segmentation results and generate a 3D model of the cerebral blood vessels.
[0013] The model optimization module, connected to the three-dimensional reconstruction module, is used to smooth the cerebral blood vessel three-dimensional model and optimize its topology.
[0014] The result output module is connected to the model optimization module and is used to output the optimized 3D model of cerebral blood vessels.
[0015] As an improvement, the deep learning model is an improved U-Net model. This improved U-Net model adds an attention mechanism module between the encoding and decoding paths. This attention mechanism module is used to strengthen the feature weights of the cerebral vascular region, and the feature weights are calculated as follows: ,in Indicates that it is located at Location feature weights This represents the feature saliency score at that location, with the denominator being the exponential sum of the feature saliency scores at all locations. This formula gives higher weight to important feature regions, thereby improving segmentation accuracy.
[0016] As an improvement, the noise reduction process in the preprocessing module employs an adaptive median filtering algorithm. This algorithm dynamically adjusts the filtering window size based on the image noise density. The window size adjustment formula is as follows: ,in This is the actual size of the filtering window. This is the initial window size. To adjust the coefficient, The noise density (ranging from 0 to 1) is used to increase the window size as the noise density increases, thus balancing the noise reduction effect with detail preservation.
[0017] As an improvement, the 3D reconstruction module includes:
[0018] A voxelization unit is used to convert the brain blood vessel segmentation results into three-dimensional voxel data.
[0019] The surface extraction unit, connected to the voxelization unit, is used to extract the cerebral blood vessel surface mesh from the three-dimensional voxel data using the moving cube algorithm, wherein the isosurface calculation satisfies: , Indicates voxels in The grayscale value of the location, The formula is used to determine the boundary surface between cerebral blood vessels and surrounding tissues, with a preset threshold.
[0020] The model building unit, connected to the surface extraction unit, is used to build an initial three-dimensional model of cerebral blood vessels based on the surface mesh.
[0021] As an improvement, the smoothing process of the model optimization module adopts the Laplacian smoothing algorithm. The Laplacian smoothing algorithm adjusts the vertex coordinates to make the curvature of the mesh surface tend to be uniform. The vertex coordinate update formula is: ,in These are the smoothed vertex coordinates. This represents the number of adjacent vertices of that vertex. The coordinates of adjacent vertices are used to shift the vertex coordinates towards the average position of adjacent vertices using this formula, thus achieving surface smoothing; the topology optimization includes removing redundant triangular faces and repairing mesh defects.
[0022] Optimization method for segmentation and reconstruction of cerebral blood vessels in medical imaging, including the following steps:
[0023] S1. Acquire medical imaging data including cerebral blood vessels;
[0024] S2. Preprocess the medical image data, the preprocessing including denoising, contrast enhancement and image standardization;
[0025] S3. Based on a deep learning model, cerebral blood vessel segmentation is performed on the preprocessed medical image data to obtain cerebral blood vessel segmentation results;
[0026] S4. Based on the cerebral blood vessel segmentation results, perform three-dimensional modeling to generate a three-dimensional model of the cerebral blood vessels;
[0027] S5. Smooth the 3D model of the cerebral blood vessels and optimize its topology;
[0028] S6. Output the optimized 3D model of cerebral blood vessels.
[0029] As an improvement, the training process of the deep learning model in step S3 includes:
[0030] Obtain a training dataset of medical images labeled with cerebral vascular regions;
[0031] The training dataset is augmented with data including random rotation, scaling, and elastic deformation.
[0032] The enhanced training dataset is input into the initial U-Net model for training. The model parameters are adjusted using a loss function, which is a weighted sum of Dice coefficient loss and cross-entropy loss, as shown in the formula: ,in This is the total loss value. The weighting coefficient (with a value range of 0-1). For Dice coefficient loss, Cross-entropy loss is used to balance the overlap of segmented regions with the classification probability error, thereby improving the model training effect.
[0033] When the model converges, training stops, and the trained deep learning model is obtained.
[0034] As an improvement, the contrast enhancement in step S2 employs an adaptive histogram equalization algorithm. This algorithm divides the image into multiple sub-regions and performs histogram equalization on each sub-region separately. The contrast gain of each sub-region is calculated as follows: , This represents the maximum grayscale value of the pixels within the sub-region. This represents the minimum grayscale value of a pixel within the sub-region. The contrast gain (ranging from 0 to 1) is used to quantify the contrast of a sub-region, guiding the adjustment of equalization intensity.
[0035] As an improvement, the specific process of topology optimization in step S5 is as follows:
[0036] Traverse the triangular faces of the 3D model of cerebral blood vessels, calculate the area and normal vector of each face, and the formula for calculating the face area is: ,in Let the area of the triangular facet be... and For the two edge vectors that make up the patch, The magnitude of the vector is used to accurately calculate the area of the patch;
[0037] Triangular facets with an area smaller than a preset threshold are identified as redundant facets and deleted.
[0038] Detect defective areas on the mesh surface and fill and repair these areas by generating new triangular facets.
[0039] As an improvement, after generating the 3D model of the cerebral blood vessels in step S4, a step of accuracy evaluation of the 3D model is also included. The accuracy evaluation uses the Hausdorff distance and the average surface distance as evaluation indicators, wherein the Hausdorff distance is calculated using the following formula: , For the set of points on the model surface, For reference standard surface point set, The Euclidean distance between two points. This represents the maximum and minimum distance between two point sets, used to measure the overall deviation of the model from the standard; when the evaluation index does not meet the preset requirements, return to step S3 to re-perform cerebral blood vessel segmentation.
[0040] The advantages of this invention compared to existing technologies are as follows: This technical solution provides multifaceted practical value for the processing of cerebrovascular features in medical images, significantly assisting clinical diagnosis and related work. Regarding segmentation accuracy, by improving the U-Net model and combining it with an attention mechanism, the feature weights of cerebrovascular regions are strengthened. Furthermore, a specific weighted loss function is used to balance segmentation and classification errors, accurately identifying small blood vessels, stenosis, or malformations, reducing misdiagnosis and missed diagnosis, and providing a reliable basis for disease diagnosis.
[0041] In the preprocessing stage, adaptive median filtering adjusts the window size according to noise density, both reducing noise and preserving details; adaptive histogram equalization enhances the contrast between blood vessels and surrounding tissues; and standardized data format ensures greater stability in subsequent processing. For 3D reconstruction and optimization, a moving cube algorithm is used to generate the model, and Laplacian smoothing optimizes the surface, removing redundant faces and repairing defects to present a clear spatial morphology of blood vessels. This assists surgeons in developing surgical plans and simulating procedures, improving surgical safety.
[0042] The system is fully automated, significantly reducing processing time and outputting standardized models and parameter reports, thus alleviating the burden on medical staff. Simultaneously, the accurate results support statistical analysis in medical research and teaching demonstrations, and can also generate universally formatted models for 3D printing, promoting applications across multiple scenarios and comprehensively improving the efficiency and quality of cerebrovascular disease diagnosis and treatment and related work. Attached Figure Description
[0043] Figure 1 This is a system structure framework diagram of the medical imaging cerebral blood vessel segmentation and reconstruction system and optimization method of the present invention.
[0044] Figure 2 This is a flowchart illustrating the execution of the medical imaging cerebral blood vessel segmentation and reconstruction system and its optimization method of the present invention.
[0045] Figure 3 This is a flowchart of the preprocessing module of the medical imaging cerebral blood vessel segmentation and reconstruction system and optimization method of the present invention.
[0046] Figure 4 This is a flowchart of the model training process for the medical imaging cerebral blood vessel segmentation and reconstruction system and optimization method of the present invention. Detailed Implementation
[0047] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings, which illustrate embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of this application will be thorough and complete.
[0048] 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 application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.
[0049] It is understood that spatial relation terms such as "below," "under," "below," "below," "above," "over," etc., can be used here to describe the relationship between one element or feature shown in the figure and other elements or features. It should be understood that, in addition to the orientation shown in the figure, spatial relation terms also include different orientations of the device in use and operation. For example, if the device in the figure is flipped, the element or feature described as "below" or "under" or "below" of the other element or feature will be oriented "over" the other element or feature. Therefore, the exemplary terms "below" and "under" can include both upper and lower orientations. Furthermore, the device may also include other orientations, such as being rotated 90 degrees or other orientations, and the spatial descriptive terms used herein will be interpreted accordingly.
[0050] It should be noted that when one element is considered to be "connected" to another element, it can be directly connected to the other element or connected to the other element through an intermediary element. In the following embodiments, "connection" should be understood as "electrical connection," "communication connection," etc., if the connected circuits, modules, units, etc., have the transmission of electrical signals or data between them.
[0051] When used herein, the singular forms of “a,” “an,” and “the” may also include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “comprising,” “including,” or “having,” etc., specify the presence of the stated feature, whole, step, operation, component, part, or combination thereof, but do not preclude the possibility of the presence or addition of one or more other features, wholes, steps, operations, components, parts, or combinations thereof.
[0052] Referring to the attached figures, the medical imaging cerebral vessel segmentation and reconstruction system and its optimization method are described. The medical imaging cerebral vessel segmentation and reconstruction system includes:
[0053] The medical image acquisition module is used to acquire medical image data including cerebral blood vessels;
[0054] A preprocessing module, connected to the medical image acquisition module, is used to preprocess the medical image data, including noise reduction, contrast enhancement, and image standardization.
[0055] A cerebral blood vessel segmentation module, connected to the preprocessing module, is used to segment cerebral blood vessels in the preprocessed medical image data based on a deep learning model to obtain cerebral blood vessel segmentation results.
[0056] A 3D reconstruction module, connected to the cerebral blood vessel segmentation module, is used to perform 3D modeling based on the cerebral blood vessel segmentation results and generate a 3D model of the cerebral blood vessels.
[0057] The model optimization module, connected to the three-dimensional reconstruction module, is used to smooth the cerebral blood vessel three-dimensional model and optimize its topology.
[0058] The result output module is connected to the model optimization module and is used to output the optimized 3D model of cerebral blood vessels.
[0059] As an improvement, the deep learning model is an improved U-Net model. This improved U-Net model adds an attention mechanism module between the encoding and decoding paths. This attention mechanism module is used to strengthen the feature weights of the cerebral vascular region, and the feature weights are calculated as follows: ,in Indicates that it is located at Location feature weights This represents the feature saliency score at that location, with the denominator being the exponential sum of the feature saliency scores at all locations. This formula gives higher weight to important feature regions, thereby improving segmentation accuracy.
[0060] As an improvement, the noise reduction process in the preprocessing module employs an adaptive median filtering algorithm. This algorithm dynamically adjusts the filtering window size based on the image noise density. The window size adjustment formula is as follows: ,in This is the actual size of the filtering window. This is the initial window size. To adjust the coefficient, The noise density (ranging from 0 to 1) is used to increase the window size as the noise density increases, thus balancing the noise reduction effect with detail preservation.
[0061] As an improvement, the 3D reconstruction module includes:
[0062] A voxelization unit is used to convert the brain blood vessel segmentation results into three-dimensional voxel data.
[0063] The surface extraction unit, connected to the voxelization unit, is used to extract the cerebral blood vessel surface mesh from the three-dimensional voxel data using the moving cube algorithm, wherein the isosurface calculation satisfies: , Indicates voxels in The grayscale value of the location, The formula is used to determine the boundary surface between cerebral blood vessels and surrounding tissues, with a preset threshold.
[0064] The model building unit, connected to the surface extraction unit, is used to build an initial three-dimensional model of cerebral blood vessels based on the surface mesh.
[0065] As an improvement, the smoothing process of the model optimization module adopts the Laplacian smoothing algorithm. The Laplacian smoothing algorithm adjusts the vertex coordinates to make the curvature of the mesh surface tend to be uniform. The vertex coordinate update formula is: ,in These are the smoothed vertex coordinates. This represents the number of adjacent vertices of that vertex. The coordinates of adjacent vertices are used to shift the vertex coordinates towards the average position of adjacent vertices using this formula, thus achieving surface smoothing; the topology optimization includes removing redundant triangular faces and repairing mesh defects.
[0066] Optimization method for segmentation and reconstruction of cerebral blood vessels in medical imaging, including the following steps:
[0067] S1. Acquire medical imaging data including cerebral blood vessels;
[0068] S2. Preprocess the medical image data, the preprocessing including denoising, contrast enhancement and image standardization;
[0069] S3. Based on a deep learning model, cerebral blood vessel segmentation is performed on the preprocessed medical image data to obtain cerebral blood vessel segmentation results;
[0070] S4. Based on the cerebral blood vessel segmentation results, perform three-dimensional modeling to generate a three-dimensional model of the cerebral blood vessels;
[0071] S5. Smooth the 3D model of the cerebral blood vessels and optimize its topology;
[0072] S6. Output the optimized 3D model of cerebral blood vessels.
[0073] As an improvement, the training process of the deep learning model in step S3 includes:
[0074] Obtain a training dataset of medical images labeled with cerebral vascular regions;
[0075] The training dataset is augmented with data including random rotation, scaling, and elastic deformation.
[0076] The enhanced training dataset is input into the initial U-Net model for training. The model parameters are adjusted using a loss function, which is a weighted sum of Dice coefficient loss and cross-entropy loss, as shown in the formula: ,in This is the total loss value. The weighting coefficient (with a value range of 0-1). For Dice coefficient loss, Cross-entropy loss is used to balance the overlap of segmented regions with the classification probability error, thereby improving the model training effect.
[0077] When the model converges, training stops, and the trained deep learning model is obtained.
[0078] As an improvement, the contrast enhancement in step S2 employs an adaptive histogram equalization algorithm. This algorithm divides the image into multiple sub-regions and performs histogram equalization on each sub-region separately. The contrast gain of each sub-region is calculated as follows: , This represents the maximum grayscale value of the pixels within the sub-region. This represents the minimum grayscale value of a pixel within the sub-region. The contrast gain (ranging from 0 to 1) is used to quantify the contrast of a sub-region, guiding the adjustment of equalization intensity.
[0079] As an improvement, the specific process of topology optimization in step S5 is as follows:
[0080] Traverse the triangular faces of the 3D model of cerebral blood vessels, calculate the area and normal vector of each face, and the formula for calculating the face area is: ,in Let the area of the triangular facet be... and For the two edge vectors that make up the patch, The magnitude of the vector is used to accurately calculate the area of the patch;
[0081] Triangular facets with an area smaller than a preset threshold are identified as redundant facets and deleted.
[0082] Detect defective areas on the mesh surface and fill and repair these areas by generating new triangular facets.
[0083] As an improvement, after generating the 3D model of the cerebral blood vessels in step S4, a step of accuracy evaluation of the 3D model is also included. The accuracy evaluation uses the Hausdorff distance and the average surface distance as evaluation indicators, wherein the Hausdorff distance is calculated using the following formula: , For the set of points on the model surface, For reference standard surface point set, The Euclidean distance between two points. It represents the maximum and minimum distance between two point sets, used to measure the overall deviation of the model from the standard;
[0084] I. Specific Implementation Methods of the System Structure:
[0085] like Figure 1 As shown (this is a text description; a system architecture diagram can be used in actual applications), the medical imaging cerebral blood vessel segmentation and reconstruction system of the present invention includes a medical image acquisition module, a preprocessing module, a cerebral blood vessel segmentation module, a three-dimensional reconstruction module, a model optimization module, and a result output module. These modules are connected in sequence and work together to complete the segmentation and reconstruction process of cerebral blood vessels.
[0086] Medical image acquisition module:
[0087] This module is primarily used to acquire medical imaging data including cerebral vascular images. This medical imaging data can be of various types, such as computed tomography (CT) images, magnetic resonance imaging (MRI) images, and digital subtraction angiography (DSA) images. In practical applications, this module can establish a data interface with the hospital's Picture Archiving and Communication System (PACS) to directly read the patient's cerebral vascular imaging data; alternatively, it can acquire image data in real time using dedicated image acquisition equipment and convert it into a digital format that the system can process, such as DICOM format. The acquired image data will then be transmitted to the preprocessing module for further processing.
[0088] Preprocessing module:
[0089] The preprocessing module is connected to the medical image acquisition module, receives the medical image data transmitted by the medical image acquisition module, and performs preprocessing operations such as noise reduction, contrast enhancement and image standardization on it to improve image quality and provide a better data foundation for subsequent cerebral blood vessel segmentation.
[0090] For noise reduction, this module employs an adaptive median filtering algorithm. The adaptive median filtering algorithm dynamically adjusts the filtering window size based on the image noise density; the window size adjustment formula is as follows: .in, This is the actual size of the filtering window. The initial window size is used in this embodiment. It can be set to 3×3 pixels; In order to adjust the coefficients, in this embodiment The value is 2; Let be the noise density, ranging from 0 to 1, representing the proportion of noisy pixels to total pixels in the image. Using this formula, when the image noise density is low, the filter window size is small, which can better preserve the detailed information of cerebral blood vessels while removing noise; when the noise density is high, the window size automatically increases to enhance the denoising effect, thus balancing denoising performance and detail preservation. For example, when the noise density... When the value is 0.1, the actual filter window size Since window size is usually an integer, it can be 3; when When it is 0.5, At this point, the window size is adjusted to 4×4 pixels.
[0091] Contrast enhancement employs an adaptive histogram equalization algorithm. This algorithm divides the image into multiple sub-regions and performs histogram equalization on each sub-region separately. The sub-region contrast gain is calculated as follows: ,in This represents the maximum grayscale value of the pixels within the sub-region. This represents the minimum grayscale value of a pixel within the sub-region. This represents the contrast gain, with a value ranging from 0 to 1. This formula can be used to quantify the contrast of a sub-region. A smaller value indicates lower contrast in the sub-region, requiring enhanced equalization processing to improve contrast; when... When the value is large, the sub-region itself has high contrast, so the equalization intensity can be appropriately reduced to avoid excessive enhancement that could amplify noise. For example, if a certain sub-region... , ,but This indicates that the sub-region has moderate contrast and can be appropriately equalized.
[0092] Image standardization involves unifying preprocessed image data to a preset grayscale range and size to facilitate subsequent processing by deep learning models. For example, image grayscale values can be normalized to the range of 0-255, and the image size can be adjusted to 512×512 pixels.
[0093] Cerebral blood vessel segmentation module:
[0094] The cerebral blood vessel segmentation module is connected to the preprocessing module, receives the preprocessed medical image data, and performs cerebral blood vessel segmentation based on a deep learning model to obtain the cerebral blood vessel segmentation results.
[0095] In this embodiment, the deep learning model employs an improved U-Net model, which adds an attention mechanism module between the encoding and decoding paths of the traditional U-Net model. The attention mechanism module is used to enhance the feature weights of the cerebral vascular region, and its feature weights are calculated as follows: .in, Indicates that it is located at Location feature weights This represents the feature saliency score at that location, obtained by performing convolution operations and activation function processing on the feature map output from the encoding path; the denominator is the exponential sum of the feature saliency scores at all locations. This formula ensures that the feature saliency score of the cerebral vascular region in the feature map is... The corresponding feature weights are relatively high. The size is also relatively large, allowing for greater focus on the characteristics of cerebral vascular regions during decoding, suppressing interference from background regions, and improving segmentation accuracy. For example, if a certain location... of Other locations The sum of the exponents of the sum is (Assuming the sum is 100), then The features at this location will receive higher weight in subsequent processing.
[0096] 3D Reconstruction Module:
[0097] The 3D reconstruction module is connected to the cerebral blood vessel segmentation module, receives the cerebral blood vessel segmentation results, and performs 3D modeling based on these results to generate a 3D model of the cerebral blood vessels. This module includes voxelization units, surface extraction units, and model building units.
[0098] Voxelization units are used to convert two-dimensional cerebral blood vessel segmentation results into three-dimensional voxel data. For multi-layer continuous two-dimensional images, a three-dimensional voxel matrix is constructed by mapping the segmented cerebral blood vessel regions in each layer of the image to their corresponding positions in three-dimensional space, where the value of each voxel indicates whether the location is a cerebral blood vessel region.
[0099] The surface extraction unit is connected to the voxelization unit, and the moving cubes algorithm is used to extract the surface mesh of brain blood vessels from the 3D voxel data. The core of the moving cubes algorithm is to extract the voxel surface by determining the isosurfaces in the voxels whose gray values are equal to a preset threshold, where the isosurface calculation satisfies: . Indicates voxels in The grayscale value of the location, In this embodiment, a preset threshold is used as the threshold. The settings can be adjusted based on the grayscale difference between cerebral blood vessels and surrounding tissues. For example, when the grayscale value of the cerebral blood vessel area is high, the settings can be adjusted accordingly. Set to a mid-to-high range of grayscale values. This formula can accurately determine the boundary surface between cerebral blood vessels and surrounding tissues, laying the foundation for subsequent mesh construction.
[0100] The model building unit is connected to the surface extraction unit, and an initial 3D model of the cerebral blood vessels is built based on the extracted surface mesh. This unit connects and integrates the surface mesh to form a complete 3D mesh model, which includes geometric information such as vertices, edges, and faces.
[0101] Model optimization module:
[0102] The model optimization module is connected to the 3D reconstruction module. It receives the initial 3D model of cerebral blood vessels and performs smoothing and topology optimization on it to improve the quality and accuracy of the model.
[0103] The smoothing process employs the Laplacian smoothing algorithm, which adjusts vertex coordinates to make the surface curvature of the mesh more uniform. The vertex coordinate update formula is as follows: .in, These are the smoothed vertex coordinates. This represents the number of adjacent vertices of that vertex. These are the coordinates of adjacent vertices. This formula shifts the coordinates of each vertex towards the average position of its neighboring vertices, thereby reducing sharp bumps and depressions on the mesh surface and making the surface smoother. For example, if a vertex... There are 4 adjacent vertices, with coordinates as follows: , , , Then the smoothed vertex coordinates .
[0104] Topology optimization includes removing redundant triangular faces and fixing mesh defects. When removing redundant triangular faces, the triangular faces of the 3D model of the cerebral blood vessels are traversed, and the area of each facet is calculated. The formula for calculating the facet area is: .in, Let the area of the triangular facet be... and For the two edge vectors that make up the patch, This represents the magnitude of the vector. This formula allows for the precise calculation of facet areas. Triangular faces with areas smaller than a preset threshold (e.g., 0.5 square millimeters) are identified as redundant and deleted to simplify the model structure without affecting its overall shape. When repairing mesh defects, new triangular faces are generated to fill and repair the defective areas by detecting unclosed regions on the mesh surface, ensuring the integrity of the 3D model's topology.
[0105] Result output module:
[0106] The output module connects to the model optimization module, receiving the optimized 3D model of cerebral blood vessels and outputting the model in various formats. These output formats include, but are not limited to: 3D visualization on a display device, allowing doctors to visually observe the morphology and structure of cerebral blood vessels; storing the model data in a common 3D model format, such as STL, for subsequent 3D printing or further analysis and processing; and generating reports on relevant model parameters, such as the length, diameter, and number of branches of cerebral blood vessels, providing quantitative indicators for clinical diagnosis.
[0107] II. Specific Implementation Methods of the Optimization Method:
[0108] The medical imaging cerebral blood vessel segmentation and reconstruction optimization method of the present invention includes the following steps:
[0109] Step S1: Acquire medical imaging data including cerebral blood vessels:
[0110] Medical imaging data, including cerebrovascular images, are acquired from patients using medical imaging acquisition equipment or the hospital's image storage and transmission system. This imaging data can be of various types, such as CT, MRI, and DSA. After acquisition, it is converted into a digital format that the system can process, such as DICOM, to prepare for subsequent processing. During the acquisition process, it is crucial to ensure the integrity and accuracy of the image data to avoid affecting subsequent segmentation and reconstruction results due to missing or corrupted data.
[0111] Step S2: Preprocess the medical image data:
[0112] Preprocessing operations such as denoising, contrast enhancement, and image standardization are performed on the acquired medical image data.
[0113] The denoising process employs an adaptive median filtering algorithm, dynamically adjusting the filtering window size based on the image noise density. The window size adjustment formula is as follows: The specific processing procedure is as described in the denoising process of the preprocessing module in the above system. Denoising reduces noise interference in the image, making the outline of cerebral blood vessels clearer.
[0114] Contrast enhancement employs an adaptive histogram equalization algorithm, dividing the image into multiple sub-regions and calculating the contrast gain for each sub-region separately. and according to The equalization intensity is adjusted, and the specific process is as described in the contrast enhancement section of the preprocessing module in the above system. Contrast enhancement improves the grayscale difference between cerebral blood vessels and surrounding tissues, making cerebral blood vessels easier to identify and segment.
[0115] Image standardization unifies image data to a preset grayscale range and size, such as normalizing grayscale values to 0-255 and adjusting the size to 512×512 pixels, to ensure consistency and stability in subsequent processing.
[0116] Step S3: Perform cerebral blood vessel segmentation on the preprocessed medical image data based on a deep learning model:
[0117] Preprocessed medical image data is input into an improved U-Net model for cerebral blood vessel segmentation. This model adds an attention mechanism module between the encoding and decoding paths, using feature weight calculation methods. The feature weights of the cerebral blood vessel region are enhanced, and the specific model structure and segmentation process are as described in the cerebral blood vessel segmentation module of the above system.
[0118] During model training, a training dataset of medical images annotated with cerebral vascular regions is first acquired. This dataset can be derived from clinical cases and annotated by professional physicians to ensure the accuracy and reliability of the data. Then, data augmentation is performed on the training dataset, including random rotation (rotation angle range of -30° to 30°), scaling (scaling ratio range of 0.8 to 1.2), and elastic deformation, to increase the diversity of the dataset and improve the model's generalization ability.
[0119] The enhanced training dataset is input into the initial U-Net model for training. The model parameters are adjusted using a loss function, which is a weighted sum of Dice coefficient loss and cross-entropy loss, as shown in the formula: .in, This is the total loss value. The weighting coefficient has a value range of 0-1. In this embodiment... It can be set to 0.6; The Dice coefficient loss is used to measure the degree of overlap between the model's segmentation results and the labeled results. , For the model segmentation results, The annotation results; Cross-entropy loss is used to measure classification probability error. This formula balances the overlap of segmented regions with classification probability error, enabling the model to accurately segment cerebral blood vessel regions while reducing the classification error rate during training, thus improving model training performance. Training stops when the model's loss value stabilizes and meets the preset convergence condition, resulting in a trained deep learning model. This model is then used to segment cerebral blood vessels from preprocessed medical image data, yielding the segmentation results.
[0120] Step S4: Perform 3D modeling based on the cerebral blood vessel segmentation results:
[0121] Based on the cerebral blood vessel segmentation results obtained in step S3, a 3D model of the cerebral blood vessels is generated. First, the 2D segmentation results are converted into 3D voxel data using voxelization units. Then, the surface extraction unit uses the moving cube algorithm, based on the isosurface calculation formula... The surface mesh of cerebral blood vessels is extracted, and finally the model building unit constructs an initial three-dimensional model of cerebral blood vessels based on the surface mesh. The specific process is as described in the three-dimensional reconstruction module of the above system.
[0122] After generating the initial 3D model of the cerebral blood vessels, the process also includes an accuracy evaluation step. The accuracy evaluation uses the Hausdorff distance and the average surface distance as evaluation metrics, where the Hausdorff distance is calculated using the following formula: . For the set of points on the model surface, For reference standard surface point sets (such as model surface point sets manually segmented and reconstructed by experts). The Euclidean distance between two points. The maximum and minimum distances between two point sets are used to measure the overall deviation of the model from the standard. The average surface distance is calculated as the average of the distances between all corresponding point pairs in the two point sets, and is used to measure the overall approximation of the model. When the evaluation metric does not meet the preset requirements (e.g., the Hausdorff distance is greater than the preset threshold of 5 mm), the process returns to step S3 to re-segment the cerebral blood vessels to improve the model's accuracy.
[0123] Step S5: Smooth the 3D model of cerebral blood vessels and optimize its topology:
[0124] The initial 3D cerebral vascular model obtained in step S4 is then smoothed and its topology optimized. The smoothing process employs the Laplacian smoothing algorithm, updated via vertex coordinates. Adjust the vertex coordinates to make the curvature of the mesh surface more uniform, as described in the smoothing process of the model optimization module in the above system.
[0125] The specific process of topology optimization is as follows: First, traverse the triangular patches of the 3D model of cerebral blood vessels, and then calculate the area of each patch according to the formula. The area and normal vector of each facet are calculated; then, triangular faces with an area less than a preset threshold are identified as redundant and deleted; finally, gaps in the mesh surface are detected, and new triangular faces are generated to fill and repair these gaps, as described in the topology optimization section of the model optimization module in the above system. These optimization operations improve the quality and accuracy of the 3D model, making it more consistent with the actual morphology of cerebral blood vessels.
[0126] Step S6: Output the optimized 3D model of cerebral blood vessels:
[0127] The optimized 3D cerebrovascular model obtained in step S5 is output in various forms, such as 3D visualization on a display device, storage in common formats like STL, and generation of relevant parameter reports. Specific output formats are as described in the results output module of the aforementioned system. The output model and reports provide doctors with clear and accurate cerebrovascular imaging information, assisting them in disease diagnosis, treatment planning, and surgical planning.
[0128] Improving the accuracy of cerebral blood vessel segmentation and providing a reliable basis for clinical diagnosis: This technical solution adopts an improved U-Net model combined with an attention mechanism module, and uses a feature weight calculation formula... By enhancing the feature weights of cerebral vascular regions, background noise interference is effectively suppressed, significantly improving segmentation accuracy. For cerebral blood vessels—structures with slender diameters, complex branching, and minimal grayscale differences from surrounding tissues—traditional segmentation methods often result in missed or false detections. This approach, however, utilizes an attention mechanism to precisely focus on key areas, clearly identifying minute vascular branches, stenotic sites, and malformed vascular clusters, making the segmentation results more closely resemble the actual anatomical structure. This provides physicians with a precise imaging basis for diagnosing cerebrovascular lesions (such as aneurysms, arteriovenous malformations, and cerebral infarction), reducing the risk of misdiagnosis or missed diagnosis due to segmentation errors, and contributing to improved early diagnosis rates and accuracy.
[0129] Meanwhile, during model training, a weighted sum of Dice coefficient loss and cross-entropy loss is used as the loss function. This approach balances the overlap of segmented regions with classification probability error, enabling the model to maintain stable segmentation performance even in complex imaging environments. It achieves high-precision segmentation for various types of medical images, including CT, MRI, and DSA, enhancing the system's versatility and reliability.
[0130] Optimize the preprocessing workflow to improve image quality and data consistency: The preprocessing module dynamically adjusts the filter window size through an adaptive median filtering algorithm. This approach achieves a precise balance between noise removal and detail preservation. For image areas with high noise density, the window size automatically increases to enhance denoising; for areas with less noise but rich detail (such as small blood vessel branches), the window size remains small to avoid detail loss due to over-filtering. This adaptive processing method solves the problem of "noise removal and detail preservation being mutually exclusive" in traditional fixed-window filtering, enabling subsequent segmentation processes to be based on clearer image data and further improving the overall processing effect.
[0131] Furthermore, the adaptive histogram equalization algorithm calculates the contrast gain of sub-regions ( This method specifically enhances the grayscale difference between cerebral blood vessels and surrounding tissues, effectively improving the recognition of blurred blood vessel boundaries, especially in low-contrast images. Image standardization processing unifies image data from different sources and of different specifications into a consistent grayscale range and size, eliminating interference from data format differences in subsequent processing. This ensures the system's stability and consistency when faced with diverse input data, providing a data foundation for multi-center clinical research or cross-device image analysis.
[0132] Generate high-quality 3D models to aid in clinical treatment planning and surgical guidance: The 3D reconstruction module extracts the surface mesh of cerebral blood vessels using the moving cube algorithm (isosurface calculation satisfies...). By combining voxelization and model building processes, two-dimensional segmentation results can be transformed into accurate three-dimensional models, fully presenting the spatial morphology, branching structure, and anatomical relationship with surrounding tissues of cerebral blood vessels. Compared with traditional two-dimensional images, three-dimensional models can provide doctors with more intuitive and comprehensive spatial information, helping them to more clearly understand the spatial characteristics of blood vessel course, branch distribution, and lesion location.
[0133] Laplace smoothing in the model optimization module (vertex coordinate update formula) This makes the surface curvature of the mesh more uniform, eliminating jagged protrusions or irregular depressions that may occur during reconstruction, resulting in a smoother model surface that more closely resembles the anatomical morphology of real blood vessels; topology optimization removes redundant triangular patches (area calculation formula) By repairing mesh vulnerabilities, the integrity and topological correctness of the model were ensured, thus avoiding clinical judgment bias caused by model defects.
[0134] These high-quality 3D models can be used directly for preoperative planning. For example, in cerebral vascular bypass surgery, doctors can use 3D models to accurately measure the diameter, length, and branch angle of blood vessels and formulate more reasonable surgical plans. In interventional treatment, they can simulate catheter pathways to improve the accuracy and safety of the surgery and reduce surgical risks and patient trauma.
[0135] Through accuracy evaluation and iterative optimization, the reliability and robustness of the results are ensured: This technical solution introduces the Hausdorff distance after the 3D model is generated. The average surface distance is used as an evaluation metric to quantify the deviation between the model and the standard reference data. When the evaluation metric does not meet the preset requirements, the system automatically returns to the segmentation step for reprocessing, forming a closed-loop iterative mechanism of "segmentation-reconstruction-evaluation-optimization" to ensure that the output 3D model always maintains high accuracy.
[0136] This iterative optimization mechanism effectively solves the error accumulation problem that may exist in the "one-time processing" of traditional methods. Especially for complex cases (such as cerebral vascular malformations, severe calcification, etc.), it can gradually correct segmentation and reconstruction errors through multiple iterations, thereby improving the reliability of the results. At the same time, this mechanism enables the system to have a certain degree of self-calibration capability, reduces the over-reliance on the quality of input images, and enhances the robustness of the system in real clinical environments.
[0137] Improving the efficiency of medical image processing and reducing the workload of medical staff: Traditional cerebral vascular segmentation and reconstruction work largely relies on manual operation or semi-automated tools, which is not only time-consuming and labor-intensive, but also prone to inconsistent results due to differences in operator experience. This technical solution significantly shortens processing time through fully automated processing (from image acquisition, preprocessing, segmentation, reconstruction to optimized output). For example, for a set of MRI images containing dozens of slices, traditional manual segmentation may take several hours, while this system can complete the entire process and output a high-quality 3D model in minutes, significantly improving work efficiency.
[0138] In addition, the standardized 3D models and quantitative parameter reports (such as blood vessel length, diameter, number of branches, etc.) output by the system reduce the workload of doctors in manual measurement and analysis, enabling medical staff to devote more energy to clinical diagnosis and patient treatment, thereby improving the overall efficiency and quality of medical services.
[0139] Advancing medical research and teaching, and expanding the application scenarios of the technology: Precise cerebral vascular segmentation and reconstruction results provide reliable quantitative data support for medical research. For example, in epidemiological studies of cerebrovascular diseases, statistical analysis of vascular morphology parameters from a large number of cases can be used to explore the patterns of disease occurrence and development; in new drug development, it can be used to assess the impact of drugs on cerebral vascular morphology, providing objective evidence for drug efficacy evaluation.
[0140] In the field of medical education, high-quality 3D models can intuitively display the anatomical structure of cerebral blood vessels, helping medical students better understand the complex relationships and spatial distribution of blood vessel branches, making up for the limitations of two-dimensional images or physical specimens in traditional anatomy teaching, and improving teaching effectiveness.
[0141] Meanwhile, the system outputs 3D models that can be stored in a universal format (such as STL), supporting the application of 3D printing technology to create physical cerebral blood vessel models for use in surgical simulations, doctor-patient communication, and other scenarios, enabling patients to more intuitively understand their condition and treatment plans, and improving the efficiency of doctor-patient communication.
[0142] In summary, through multi-module collaborative work and the application of innovative algorithms, significant improvements have been achieved in segmentation accuracy, model quality, processing efficiency, and clinical applicability. This not only provides strong technical support for clinical diagnosis and treatment but also has broad application value in medical research, teaching, and doctor-patient communication, and is of great significance to promoting the development of cerebrovascular disease diagnosis and treatment technology.
Claims
1. A medical imaging system for cerebral blood vessel segmentation and reconstruction, characterized in that: include: The medical image acquisition module is used to acquire medical image data including cerebral blood vessels; A preprocessing module, connected to the medical image acquisition module, is used to preprocess the medical image data, including noise reduction, contrast enhancement, and image standardization. A cerebral blood vessel segmentation module, connected to the preprocessing module, is used to segment cerebral blood vessels in the preprocessed medical image data based on a deep learning model to obtain cerebral blood vessel segmentation results. A 3D reconstruction module, connected to the cerebral blood vessel segmentation module, is used to perform 3D modeling based on the cerebral blood vessel segmentation results and generate a 3D model of the cerebral blood vessels. The model optimization module, connected to the three-dimensional reconstruction module, is used to smooth the cerebral blood vessel three-dimensional model and optimize its topology. The result output module is connected to the model optimization module and is used to output the optimized 3D model of cerebral blood vessels.
2. The medical imaging cerebral blood vessel segmentation and reconstruction system according to claim 1, characterized in that: The deep learning model is an improved U-Net model, which adds an attention mechanism module between the encoding and decoding paths. This attention mechanism module is used to strengthen the feature weights of cerebral vascular regions. The feature weights are calculated as follows: ,in Indicates that it is located at Location feature weights This represents the feature saliency score at that location, with the denominator being the exponential sum of the feature saliency scores at all locations. This formula gives higher weight to important feature regions, thereby improving segmentation accuracy.
3. The medical imaging cerebral blood vessel segmentation and reconstruction system according to claim 1, characterized in that: The noise reduction process in the preprocessing module employs an adaptive median filtering algorithm. This algorithm dynamically adjusts the filtering window size based on the image noise density. The window size adjustment formula is as follows: ,in This is the actual size of the filtering window. This is the initial window size. To adjust the coefficient, The noise density (ranging from 0 to 1) is used to increase the window size as the noise density increases, thus balancing the noise reduction effect with detail preservation.
4. The medical imaging cerebral blood vessel segmentation and reconstruction system according to claim 1, characterized in that: The three-dimensional reconstruction module includes: A voxelization unit is used to convert the brain blood vessel segmentation results into three-dimensional voxel data. The surface extraction unit, connected to the voxelization unit, is used to extract the cerebral blood vessel surface mesh from the three-dimensional voxel data using the moving cube algorithm, wherein the isosurface calculation satisfies: , Indicates voxels in The grayscale value of the location, The formula is used to determine the boundary surface between cerebral blood vessels and surrounding tissues, with a preset threshold. The model building unit, connected to the surface extraction unit, is used to build an initial three-dimensional model of cerebral blood vessels based on the surface mesh.
5. The medical imaging cerebral blood vessel segmentation and reconstruction system according to claim 1, characterized in that: The smoothing process in the model optimization module employs the Laplacian smoothing algorithm. This algorithm adjusts the vertex coordinates to make the surface curvature of the mesh more uniform. The vertex coordinate update formula is as follows: ,in These are the smoothed vertex coordinates. This represents the number of adjacent vertices of that vertex. The coordinates of adjacent vertices are used to shift the vertex coordinates towards the average position of adjacent vertices using this formula, thus achieving surface smoothing; the topology optimization includes removing redundant triangular faces and repairing mesh defects.
6. An optimized method for segmentation and reconstruction of cerebral blood vessels in medical imaging, characterized by: Includes the following steps: S1. Acquire medical imaging data including cerebral blood vessels; S2. Preprocess the medical image data, the preprocessing including denoising, contrast enhancement and image standardization; S3. Based on a deep learning model, cerebral blood vessel segmentation is performed on the preprocessed medical image data to obtain cerebral blood vessel segmentation results; S4. Based on the cerebral blood vessel segmentation results, perform three-dimensional modeling to generate a three-dimensional model of the cerebral blood vessels; S5. Smooth the 3D model of the cerebral blood vessels and optimize its topology; S6. Output the optimized 3D model of cerebral blood vessels.
7. The optimized method for segmentation and reconstruction of cerebral blood vessels in medical imaging according to claim 6, characterized in that: The training process of the deep learning model in step S3 includes: Obtain a training dataset of medical images labeled with cerebral vascular regions; The training dataset is augmented with data including random rotation, scaling, and elastic deformation. The enhanced training dataset is input into the initial U-Net model for training. The model parameters are adjusted using a loss function, which is a weighted sum of Dice coefficient loss and cross-entropy loss, as shown in the formula: ,in This is the total loss value. The weighting coefficient (with a value range of 0-1). For Dice coefficient loss, Cross-entropy loss is used to balance the overlap of segmented regions with the classification probability error, thereby improving the model training effect. When the model converges, training stops, and the trained deep learning model is obtained.
8. The optimized method for segmentation and reconstruction of cerebral blood vessels in medical imaging according to claim 6, characterized in that: The contrast enhancement in step S2 employs an adaptive histogram equalization algorithm. This algorithm divides the image into multiple sub-regions and performs histogram equalization on each sub-region separately. The contrast gain of each sub-region is calculated as follows: , This represents the maximum grayscale value of the pixels within the sub-region. This represents the minimum grayscale value of a pixel within the sub-region. The contrast gain (ranging from 0 to 1) is used to quantify the contrast of a sub-region, guiding the adjustment of equalization intensity.
9. The optimized method for segmentation and reconstruction of cerebral blood vessels in medical imaging according to claim 6, characterized in that: The specific process of topology optimization in step S5 is as follows: Traverse the triangular faces of the 3D model of cerebral blood vessels, calculate the area and normal vector of each face, and the formula for calculating the face area is: ,in Let the area be the triangular facet. and Let the two edge vectors that make up the patch be... The magnitude of the vector is used to accurately calculate the area of the patch; Triangular facets with an area smaller than a preset threshold are identified as redundant facets and deleted. Detect defective areas on the mesh surface and fill and repair these areas by generating new triangular facets.
10. The optimized method for segmentation and reconstruction of cerebral blood vessels in medical imaging according to claim 6, characterized in that: After generating the 3D model of the cerebral blood vessels in step S4, the method further includes an accuracy evaluation step for the 3D model. The accuracy evaluation uses the Hausdorff distance and the average surface distance as evaluation indicators, wherein the Hausdorff distance is calculated using the following formula: For the set of points on the model surface, For reference standard surface point set, The Euclidean distance between two points. This represents the maximum and minimum distance between two point sets, used to measure the overall deviation of the model from the standard; when the evaluation index does not meet the preset requirements, return to step S3 to re-perform cerebral blood vessel segmentation.