Three-dimensional reconstruction detection method and system for brain tumor images

Through adaptive threshold segmentation, multi-scale feature integration and three-dimensional morphological constraint boundary extraction, combined with light projection and surface reconstruction technology, the computational complexity and insufficient visualization of brain tumor three-dimensional reconstruction in the existing technology is solved, and efficient tumors are achieved Boundary extraction and morphological reconstruction provide intuitive analytical means for brain tumor diagnosis.

CN118279302BActive Publication Date: 2024-11-08XIAN YIBINGFAN BIOTECHNOLOGY CO LTD
View PDF 8 Cites 0 Cited by

Patent Information

Application Number
CN202410695171.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-31
Publication Date
2024-11-08
Estimated Expiration
2044-05-31

AI Technical Summary

Technical Problem

The existing three-dimensional reconstruction technology has problems such as high computational complexity, difficulty in real-time interaction and visualization in brain tumor diagnosis, and it is difficult to accurately represent the boundaries and morphological characteristics of brain tumors.

Method used

Adaptive threshold segmentation algorithm and morphological filtering are used to extract brain tissue areas, combine multi-scale feature integration and adaptive fuzzy inference optimization strategies to determine tumor candidate areas, introduce a three-dimensional morphological constraint boundary extraction method, and realize three-dimensional through light projection and surface reconstruction. Visualization.

Benefits of technology

It improves the adaptability and flexibility of tumor area identification, realizes accurate three-dimensional tumor boundary extraction and realistic morphological reconstruction, and provides doctors with intuitive and three-dimensional diagnostic and treatment plan formulation methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118279302B_ABST
    Figure CN118279302B_ABST
Patent Text Reader

Abstract

The present invention provides a three-dimensional reconstruction detection method and system for brain tumor images, which relates to the field of detection technology, including registering and fusing brain medical images of different modalities, using a weighted fusion strategy to adaptively determine the fusion weight to obtain a fused brain image; extracting the brain tissue area of ​​the fused brain image; constructing a tumor candidate area extraction model with multi-scale feature integration to determine a tumor candidate feature set; dynamically adjusting the membership function and the inference rule according to an adaptive fuzzy reasoning optimization strategy, adaptively generating the optimal fuzzy reasoning parameters according to the tumor candidate feature set, and determining the tumor target area; tracking the tumor boundary of the tumor target area in three-dimensional space, combining local shape features and global shape prior knowledge to extract the three-dimensional tumor boundary; based on a three-dimensional visualization algorithm for ray casting and surface reconstruction, comprehensively utilizing multi-view projection and three-dimensional texture mapping technology to achieve three-dimensional morphological reconstruction of the tumor.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to detection technology, and in particular to a three-dimensional reconstruction detection method and system for brain tumor images. Background Art

[0002] Brain tumor is a disease that seriously threatens human health. Its early diagnosis and precise treatment are crucial to improving the survival rate of patients. Traditional brain tumor diagnosis mainly relies on doctors' observation and analysis of two-dimensional medical images, such as computed tomography (CT) and magnetic resonance imaging (MRI). However, due to the complex morphological structure of brain tumors, it is difficult to fully and intuitively display the spatial position, size and morphological characteristics of the tumor on a two-dimensional plane, which limits the accuracy of diagnosis and treatment plan formulation.

[0003] In order to overcome the limitations of two-dimensional medical images, researchers began to explore the use of three-dimensional reconstruction technology to assist in the diagnosis and treatment of brain tumors. Three-dimensional reconstruction technology generates a three-dimensional model of brain tumors by performing sequence registration, image segmentation and surface rendering on multiple layers of two-dimensional medical images, allowing doctors to observe the spatial structure and morphological characteristics of tumors from any perspective, thereby improving the accuracy of diagnosis and the visualization of treatment plans.

[0004] However, the existing 3D reconstruction technology still has some shortcomings when applied to brain tumor diagnosis. First, the brain tissue structure is complex and the contrast between soft tissues is low, which puts higher requirements on image segmentation and feature extraction algorithms. Secondly, brain tumors have diverse morphologies and irregular boundaries, which are difficult to represent and reconstruct using simple geometric models. In addition, traditional 3D reconstruction methods often have high computational complexity and are difficult to achieve real-time interaction and visualization, which limits their promotion in clinical applications. Summary of the invention

[0005] The embodiments of the present invention provide a three-dimensional reconstruction detection method and system for brain tumor images, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention,

[0007] Provides a 3D reconstruction detection method for brain tumor images, including:

[0008] The brain medical images of different modalities are registered and fused, and the fusion weights are adaptively determined using the weighted fusion strategy to obtain the fused brain image. The fused brain image is preprocessed using the adaptive threshold segmentation algorithm, and the optimal threshold is adaptively determined according to the grayscale distribution of the fused brain image. The brain tissue area of ​​the fused brain image is extracted by combining morphological filtering and connected domain analysis.

[0009] Construct a tumor candidate region extraction model with multi-scale feature integration, extract feature vectors of brain tissue regions at different scales, use feature selection algorithm to optimize feature subsets, and determine tumor candidate feature sets; according to the adaptive fuzzy reasoning optimization strategy, dynamically adjust the membership function and reasoning rules, adaptively generate the optimal fuzzy reasoning parameters based on the tumor candidate feature set, and determine the tumor target region;

[0010] A tumor boundary extraction method with three-dimensional morphological constraints is introduced to track the tumor boundary of the tumor target area in three-dimensional space, and the three-dimensional tumor boundary is extracted by combining local shape features and global shape prior knowledge; a three-dimensional visualization algorithm based on ray casting and surface reconstruction is used, which comprehensively utilizes multi-view projection and three-dimensional texture mapping technology to realize the three-dimensional morphological reconstruction of the tumor.

[0011] In an optional embodiment,

[0012] The fused brain image is preprocessed using an adaptive threshold segmentation algorithm. The optimal threshold is adaptively determined according to the grayscale distribution of the fused brain image. The brain tissue areas of the fused brain image are extracted by combining morphological filtering and connected domain analysis, including:

[0013] Calculating the grayscale histogram of the fused brain image, using the maximum inter-class variance method to adaptively determine the optimal segmentation threshold, and performing binary segmentation on the fused brain image based on the optimal segmentation threshold to obtain a foreground area and a background area;

[0014] Performing a morphological opening operation on the foreground area to remove noise and holes in the foreground area and smooth the boundary of the foreground area to obtain a modified foreground area;

[0015] Adopting a connected domain analysis method based on region growing, extracting the largest connected brain tissue region of the modified foreground region according to pixel grayscale similarity and spatial adjacency;

[0016] Performing a morphological closing operation on the maximum connected brain tissue region to fill the holes in the maximum connected brain tissue region to obtain an initial brain tissue region;

[0017] Combined with prior knowledge of brain anatomical structure, morphological filtering and boundary correction are performed on the initial brain tissue region to remove the residual non-brain tissue region, thereby obtaining the brain tissue region of the fused brain image.

[0018] In an optional embodiment,

[0019] The connected domain analysis method based on region growing is adopted to extract the largest connected brain tissue region of the modified foreground region according to the pixel grayscale similarity and spatial adjacency relationship, including:

[0020] In the modified foreground area, randomly selecting a pixel point as an initial seed point, and marking it as an original connected brain tissue area;

[0021] Calculating the grayscale difference between the initial seed point and all unmarked pixels in its neighborhood, incorporating pixels whose grayscale difference is less than a preset grayscale threshold into the original connected brain tissue region, and marking them as visited;

[0022] For the newly added pixels in the original connected brain tissue area, the grayscale difference is recursively calculated until no new pixels are included in the original connected brain tissue area;

[0023] Randomly selecting other pixel points as initial seed points until all pixel points in the modified foreground area are marked as visited, thereby obtaining multiple connected brain tissue areas;

[0024] The numbers of pixels in the multiple connected brain tissue regions are compared, and the connected region with the largest number of pixels is selected as the largest connected brain tissue region.

[0025] In an optional embodiment,

[0026] According to the adaptive fuzzy inference optimization strategy, the membership function and inference rules are dynamically adjusted, and the optimal fuzzy inference parameters are adaptively generated according to the tumor candidate feature set to determine the tumor target area including:

[0027] Cluster analysis is performed on the candidate tumor feature set to divide the brain tissue area into several clusters, each cluster representing a tumor feature pattern;

[0028] For each candidate tumor feature set corresponding to each cluster, the mean and standard deviation of the feature values ​​of the candidate tumor feature set are calculated to construct an initial membership function, where the initial membership function is used to represent the similarity between the clusters;

[0029] Based on expert knowledge and historical case data, an initial fuzzy inference rule base is constructed to associate the candidate tumor feature set with its corresponding tumor type and malignancy.

[0030] Optimizing the parameters of the initial membership function and the weights of the inference rules in the initial fuzzy inference rule base by using a genetic algorithm, evaluating the performance of multiple candidate solutions by using a fitness function, selecting the candidate solution with the largest fitness value corresponding to the fitness function as the optimal combination solution, and determining the optimized membership function and the optimized fuzzy inference rule by using the optimal combination solution;

[0031] For each brain tissue region, the membership is calculated using the optimized membership function, and the membership is used as the input of the optimized fuzzy inference rule to determine the inference result, and the brain tissue region with the largest inference value corresponding to the inference result is used as the tumor target region.

[0032] In an optional embodiment,

[0033] The parameters of the initial membership function and the weights of the inference rules in the initial fuzzy inference rule base are optimized by using a genetic algorithm, the performance of multiple candidate solutions is evaluated by using a fitness function, and the candidate solution with the largest fitness value corresponding to the fitness function is selected as the optimal combination solution. The optimized membership function and the optimized fuzzy inference rule are determined by using the optimal combination solution, including:

[0034] Using the parameters of the initial membership function and the weights of the inference rules in the initial fuzzy inference rule base as chromosome codes of the genetic algorithm;

[0035] An initial population is randomly generated, each individual corresponding to a combination of a set of parameters of an initial membership function and a weight of an inference rule in the initial fuzzy inference rule base;

[0036] The roulette wheel selection algorithm is used to select some individuals from the current initial population as parent individuals;

[0037] Perform a crossover operation on the parent individuals, randomly select two parent individuals, randomly select crossover points on the chromosomes of the two parent individuals, exchange the gene fragments after the crossover points, and generate new offspring individuals;

[0038] Perform mutation operations on new offspring individuals with random probability, randomly select the gene positions of new offspring individuals, iteratively merge parent individuals and new offspring individuals to form a new population, until the preset evolutionary generation is reached or the convergence condition is met;

[0039] Select the individual with the highest fitness value from the new population as the optimal combination solution.

[0040] In an optional embodiment,

[0041] The tumor boundary extraction method with 3D morphological constraints is introduced to track the tumor boundary of the tumor target area in 3D space. The 3D tumor boundary is extracted by combining local shape features and global shape prior knowledge, including:

[0042] Based on the determined tumor target area, an initial tumor boundary surface is generated by performing local Gaussian fitting on the three-dimensional morphology of the tumor target area;

[0043] Extracting local shape features of each vertex on the initial tumor boundary surface, wherein the local shape features include at least one of curvature, normal vector and shape index;

[0044] Based on the local shape features and the predetermined a priori global shape features, a three-dimensional morphological constraint energy function is constructed to evaluate the degree of matching between the current boundary surface and the actual tumor boundary;

[0045] By using variational inference and surface evolution technology, driven by the three-dimensional morphological constraint energy function, the initial tumor boundary surface is iteratively adjusted and optimized until the evolution of the initial tumor boundary surface converges or reaches a preset number of iterations, thereby obtaining the final three-dimensional tumor boundary.

[0046] In an optional embodiment,

[0047] Based on the local shape features and the predetermined prior global shape features, constructing a three-dimensional morphological constraint energy function includes:

[0048] ;

[0049] ;

[0050] ;

[0051] in, E represents the three-dimensional morphological constraint energy function, α and β represent the first weight value corresponding to the local shape feature function and the second weight value corresponding to the prior global shape feature function, respectively;

[0052] E local represents the local shape characteristic function, N represents the number of vertices of the initial tumor boundary surface, f i Indicates i The local shape features corresponding to the vertices are F The covariance matrix representing the local shape features;

[0053] E global represents the prior global shape feature function, p represents the shape parameter of the initial tumor boundary surface, represents the mean of the shape parameters.

[0054] In an optional embodiment,

[0055] The 3D visualization algorithm based on ray casting and surface reconstruction uses multi-view projection and 3D texture mapping technology to achieve 3D morphological reconstruction of tumors, including:

[0056] A virtual light source is set inside the tumor target area, and a ray projection algorithm is used to project the tumor target area from multiple perspectives to generate a set of two-dimensional projection images;

[0057] Perform image segmentation on each two-dimensional projection image, extract the tumor boundary contour, and obtain a set of two-dimensional contour curves;

[0058] Based on the geometric relationship of ray projection, the two-dimensional contour curve is back-projected into three-dimensional space to obtain a set of three-dimensional space curves, wherein the three-dimensional space curves represent the contour of the tumor surface at different viewing angles;

[0059] Using a 3D space curve interpolation algorithm, discrete 3D space curves are connected to generate a tumor surface mesh model;

[0060] Using multi-view texture mapping technology, the tumor texture images at different viewpoints are mapped to the tumor surface mesh model to obtain the three-dimensional morphology of the tumor.

[0061] In the second aspect of the embodiment of the present application,

[0062] Provides a 3D reconstruction detection system for brain tumor images, including:

[0063] The first unit is used to register and fuse brain medical images of different modalities, and use the weighted fusion strategy to adaptively determine the fusion weight to obtain a fused brain image; the fused brain image is preprocessed using an adaptive threshold segmentation algorithm, and the optimal threshold is adaptively determined according to the grayscale distribution of the fused brain image, and the brain tissue area of ​​the fused brain image is extracted by combining morphological filtering and connected domain analysis;

[0064] The second unit is used to build a tumor candidate region extraction model with multi-scale feature integration, extract feature vectors of brain tissue regions at different scales, use feature selection algorithms to optimize feature subsets, and determine tumor candidate feature sets; according to the adaptive fuzzy reasoning optimization strategy, dynamically adjust the membership function and reasoning rules, adaptively generate the optimal fuzzy reasoning parameters based on the tumor candidate feature set, and determine the tumor target region;

[0065] The third unit is used to introduce a tumor boundary extraction method with three-dimensional morphological constraints, track the tumor boundary of the tumor target area in three-dimensional space, and extract the three-dimensional tumor boundary by combining local shape features and global shape prior knowledge; a three-dimensional visualization algorithm based on ray casting and surface reconstruction, which comprehensively utilizes multi-view projection and three-dimensional texture mapping technology to achieve three-dimensional morphological reconstruction of the tumor.

[0066] According to a third aspect of the embodiments of the present invention,

[0067] An electronic device is provided, comprising:

[0068] processor;

[0069] a memory for storing processor-executable instructions;

[0070] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0071] By registering and fusing brain medical images of different modalities and using a weighted fusion strategy to adaptively determine the fusion weights, a fused brain image is obtained, which effectively integrates the complementary information of images of different modalities, improves the recognizability of brain tissue structure, and lays the foundation for subsequent tumor area extraction.

[0072] An adaptive fuzzy reasoning optimization strategy is introduced to dynamically adjust the membership function and reasoning rules according to the tumor candidate feature set, generate the optimal fuzzy reasoning parameters, and adaptively determine the tumor target area, which overcomes the limitations of traditional fixed fuzzy rules and improves the adaptability and flexibility of tumor area identification.

[0073] This application proposes a tumor boundary extraction method with three-dimensional morphology constraints, which tracks the boundaries of the tumor target area in three-dimensional space, combines local shape features and global shape prior knowledge, and realizes accurate three-dimensional tumor boundary extraction, overcomes the limitations of traditional two-dimensional segmentation methods, and improves the spatial accuracy of tumor boundary description.

[0074] By adopting a three-dimensional visualization algorithm based on ray projection and surface reconstruction, and comprehensively utilizing multi-view projection and three-dimensional texture mapping technology, realistic three-dimensional tumor morphology reconstruction is achieved, providing doctors with intuitive and three-dimensional means of observing and analyzing tumor morphology, and improving the visualization level of diagnosis and treatment plan formulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 A schematic diagram of the flow of a three-dimensional reconstruction detection method for brain tumor images according to an embodiment of the present invention;

[0076] Figure 2 It is a schematic structural diagram of a three-dimensional reconstruction detection system for brain tumor images according to an embodiment of the present invention. DETAILED DESCRIPTION

[0077] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0078] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0079] Figure 1 FIG. 1 is a flow chart of a three-dimensional reconstruction detection method for brain tumor images according to an embodiment of the present invention. Figure 1As shown, the method includes:

[0080] S101. Register and fuse brain medical images of different modalities, use weighted fusion strategy to adaptively determine fusion weights, and obtain fused brain images; use adaptive threshold segmentation algorithm to preprocess the fused brain images, adaptively determine the optimal threshold according to the grayscale distribution of the fused brain images, and combine morphological filtering and connected domain analysis to extract the brain tissue area of ​​the fused brain images;

[0081] In an optional embodiment,

[0082] The fused brain image is preprocessed using an adaptive threshold segmentation algorithm. The optimal threshold is adaptively determined according to the grayscale distribution of the fused brain image. The brain tissue areas of the fused brain image are extracted by combining morphological filtering and connected domain analysis, including:

[0083] Calculating the grayscale histogram of the fused brain image, using the maximum inter-class variance method to adaptively determine the optimal segmentation threshold, and performing binary segmentation on the fused brain image based on the optimal segmentation threshold to obtain a foreground area and a background area;

[0084] Performing a morphological opening operation on the foreground area to remove noise and holes in the foreground area and smooth the boundary of the foreground area to obtain a modified foreground area;

[0085] Adopting a connected domain analysis method based on region growing, extracting the largest connected brain tissue region of the modified foreground region according to pixel grayscale similarity and spatial adjacency;

[0086] Performing a morphological closing operation on the maximum connected brain tissue region to fill the holes in the maximum connected brain tissue region to obtain an initial brain tissue region;

[0087] Combined with prior knowledge of brain anatomical structure, morphological filtering and boundary correction are performed on the initial brain tissue region to remove the residual non-brain tissue region, thereby obtaining the brain tissue region of the fused brain image.

[0088] For example, in the field of medical imaging, accurate analysis of brain images is crucial for the diagnosis and treatment of various neurological diseases. Traditional image processing methods often rely on fixed thresholds to segment images, which may not be suitable for the particularities of brain images, such as grayscale inhomogeneity. Therefore, using an adaptive threshold segmentation algorithm to preprocess the fused brain image can more accurately extract the brain tissue area and improve the effectiveness and accuracy of subsequent analysis.

[0089] First, the grayscale histogram of the fused brain image is calculated. This step is to analyze the grayscale distribution of the entire image and provide a data basis for the determination of the adaptive threshold. The maximum inter-class variance method (Otsu's Method) is used to adaptively determine the optimal segmentation threshold of the image. This method automatically selects the segmentation threshold by maximizing the inter-class variance between the foreground and background to achieve the best segmentation effect. Based on the optimal segmentation threshold determined in the previous step, the fused brain image is binarized and the image is segmented into the foreground area (brain tissue) and the background area.

[0090] The binarized foreground region is subjected to morphological opening, including the process of erosion followed by dilation, which helps to remove small noise points and holes and smooth the boundaries of the foreground region, thereby obtaining the corrected foreground region. The connected domain analysis method based on region growing is used to extract the largest connected brain tissue region from the corrected foreground region according to the similarity of pixel grayscale and spatial adjacency. The morphological closing operation (dilation followed by erosion) is performed on the extracted largest connected brain tissue region to fill the holes in the region and obtain the initial brain tissue region. Finally, combined with the prior knowledge of brain anatomical structure, the initial brain tissue region is further subjected to morphological filtering and boundary correction to remove the residual non-brain tissue region and ensure that the brain tissue region in the obtained fused brain image is as accurate as possible.

[0091] In an optional embodiment,

[0092] The connected domain analysis method based on region growing is adopted to extract the largest connected brain tissue region of the modified foreground region according to the pixel grayscale similarity and spatial adjacency relationship, including:

[0093] In the modified foreground area, randomly selecting a pixel point as an initial seed point, and marking it as an original connected brain tissue area;

[0094] Calculating the grayscale difference between the initial seed point and all unmarked pixels in its neighborhood, incorporating pixels whose grayscale difference is less than a preset grayscale threshold into the original connected brain tissue region, and marking them as visited;

[0095] For the newly added pixels in the original connected brain tissue area, the grayscale difference is recursively calculated until no new pixels are included in the original connected brain tissue area;

[0096] Randomly selecting other pixel points as initial seed points until all pixel points in the modified foreground area are marked as visited, thereby obtaining multiple connected brain tissue areas;

[0097] The numbers of pixels in the multiple connected brain tissue regions are compared, and the connected region with the largest number of pixels is selected as the largest connected brain tissue region.

[0098] Exemplarily, in the corrected foreground area, a pixel is randomly selected as the initial seed point. The selection of this seed point can be based on a certain strategy, such as selecting a point with a grayscale value close to the regional average, or directly randomly selected. The selected seed point will be marked as part of the original connected brain tissue area and set to a visited state. The grayscale difference between the initial seed point and all unmarked pixels in its surrounding neighborhood (usually an 8-connected or 4-connected neighborhood) is calculated. If this grayscale difference is less than a preset grayscale threshold, these pixels are considered to belong to the same connected area as the seed point, so these pixels are included in the original connected brain tissue area and marked as visited.

[0099] For each pixel newly added to the original connected brain tissue region, the grayscale difference calculation and region growing process are repeated. This recursive process continues until no new pixels can be added to the current connected region. Once the current connected region cannot be further expanded, the growing process for this region is completed. At this time, another new seed point is randomly selected from the pixels in the revised foreground region that are not marked as visited until all pixels in the revised foreground region have been visited.

[0100] After growing all possible connected regions, compare the number of pixels in these regions. Select the connected region with the largest number of pixels, and this region is considered to be the largest connected brain tissue region. This step is based on the assumption that larger connected regions are more likely to represent major brain tissue structures.

[0101] Through the above steps, the largest connected brain tissue region can be effectively extracted from the corrected foreground region. This process utilizes the grayscale similarity and spatial adjacency between pixels, combined with the region growing method, to gradually construct an image region representing brain tissue. The maximum connected brain tissue region finally obtained can be used for further medical analysis and diagnosis.

[0102] S102. Construct a tumor candidate region extraction model with multi-scale feature integration, extract feature vectors of brain tissue regions at different scales, use feature selection algorithm to optimize feature subsets, and determine tumor candidate feature sets; dynamically adjust membership functions and inference rules based on adaptive fuzzy reasoning optimization strategy, adaptively generate optimal fuzzy reasoning parameters based on tumor candidate feature sets, and determine tumor target regions;

[0103] In an optional embodiment,

[0104] According to the adaptive fuzzy inference optimization strategy, the membership function and inference rules are dynamically adjusted, and the optimal fuzzy inference parameters are adaptively generated according to the tumor candidate feature set to determine the tumor target area including:

[0105] Cluster analysis is performed on the candidate tumor feature set to divide the brain tissue area into several clusters, each cluster representing a tumor feature pattern;

[0106] For each candidate tumor feature set corresponding to each cluster, the mean and standard deviation of the feature values ​​of the candidate tumor feature set are calculated to construct an initial membership function, where the initial membership function is used to represent the similarity between the clusters;

[0107] Based on expert knowledge and historical case data, an initial fuzzy inference rule base is constructed to associate the candidate tumor feature set with its corresponding tumor type;

[0108] Optimizing the parameters of the initial membership function and the weights of the inference rules in the initial fuzzy inference rule base by using a genetic algorithm, evaluating the performance of multiple candidate solutions by using a fitness function, selecting the candidate solution with the largest fitness value corresponding to the fitness function as the optimal combination solution, and determining the optimized membership function and the optimized fuzzy inference rule by using the optimal combination solution;

[0109] For each brain tissue region, the membership is calculated using the optimized membership function, and the membership is used as the input of the optimized fuzzy inference rule to determine the inference result, and the brain tissue region with the largest inference value corresponding to the inference result is used as the tumor target region.

[0110] Exemplarily, feature vectors of brain tissue are extracted at different scales. Typically, this involves extracting features such as texture, edges, shape, and intensity on the image using filters or windows of different sizes. This step is intended to capture the diverse local and global features that a tumor may exhibit. Local features: These include pixel intensity statistics within a small window, gradient direction histograms, etc. Global features: Features involving larger areas, such as texture analysis of the entire image or shape description of a region.

[0111] After extracting a large number of features, a feature selection algorithm is used to optimize the feature subset. This step is critical because it can remove redundant or irrelevant features and improve the efficiency and performance of the model. Among them, the feature selection algorithm can include filtering methods: such as variance analysis and correlation coefficient test, which are usually performed before model training.

[0112] Optionally, brain image data can be collected and a tumor candidate feature set can be extracted from it. The tumor candidate feature set is clustered using a clustering algorithm (such as K-means, GMM, etc.). Each cluster represents a tumor feature pattern, and the tumor can be grouped according to its shape, size, texture, and other characteristics.

[0113] For each cluster, calculate the mean and standard deviation of the feature values ​​in its tumor candidate feature set. Use the above statistics (mean and standard deviation) to construct initial membership functions, which can be Gaussian functions, to represent the similarity of samples in different clusters. Build an initial fuzzy inference rule base based on expert knowledge and historical case data. These rules will associate the tumor candidate feature set with the tumor type and its malignancy. Work with clinicians and oncologists to ensure that the set rule base can accurately reflect the clinical characteristics of the tumor.

[0114] Genetic algorithms are used to optimize the parameters of the membership function and the weights of the inference rules in the fuzzy inference rule base. This includes steps such as encoding, selection, crossover, and mutation. A fitness function is designed to evaluate the performance of multiple candidate solutions, and the candidate solution with the highest fitness value corresponding to the fitness function is selected as the optimal solution.

[0115] The membership of each brain tissue region is calculated using the optimized membership function. The calculated membership is used as input, and the inference result of each region is determined by the optimized fuzzy inference rule. The brain tissue region with the largest inference value in the inference result is marked as the tumor target region.

[0116] In particular, we collaborated with oncology experts to collect diagnostic criteria and characteristics of different types of tumors. These data include tumor growth rate, morphology, boundary characteristics, etc. Historical case data were collected from the hospital database, including patients' MRI or CT scan images, diagnosis results, tumor types, and related feature data. Based on expert advice and historical data analysis, we selected features that were highly correlated with tumor types, such as tumor size, shape, texture, enhancement pattern, etc. We performed necessary preprocessing on the selected features, such as normalization, denoising, and data standardization, to facilitate subsequent processing.

[0117] Rule form definition: Define the basic form of fuzzy rules, usually "if X is A, then Y is B", where X is the input variable (tumor feature set), A is the fuzzy set (fuzzy description of the feature), Y is the output variable (tumor type), and B is the fuzzy set (fuzzy description of the tumor type).

[0118] Based on the collected expert knowledge and historical data, preliminary fuzzy inference rules are formulated. For example, if the border of a tumor is fuzzy and it grows rapidly, it may be a malignant tumor. Fuzzy set definition: Define fuzzy sets for each feature and tumor type, such as dividing "boundary clarity" into fuzzy sets such as "clear", "fuzzy", and "very fuzzy". Construct membership functions for each fuzzy set, which describe the degree of membership of the fuzzy set corresponding to each eigenvalue. Commonly used membership functions include triangular, trapezoidal, Gaussian, etc. Preliminary testing: Apply the preliminary rule base to a portion of historical case data to observe the degree of match between its predicted results and the actual diagnosis. Optimization and adjustment: Adjust the parameters of fuzzy rules and membership functions based on test results and expert feedback. It may be necessary to add new rules or modify existing rules to improve the accuracy of diagnosis.

[0119] In an optional embodiment,

[0120] The parameters of the initial membership function and the weights of the inference rules in the initial fuzzy inference rule base are optimized by using a genetic algorithm, the performance of multiple candidate solutions is evaluated by using a fitness function, and the candidate solution with the largest fitness value corresponding to the fitness function is selected as the optimal combination solution. The optimized membership function and the optimized fuzzy inference rule are determined by using the optimal combination solution, including:

[0121] Using the parameters of the initial membership function and the weights of the inference rules in the initial fuzzy inference rule base as chromosome codes of the genetic algorithm;

[0122] An initial population is randomly generated, each individual corresponding to a combination of a set of parameters of an initial membership function and a weight of an inference rule in the initial fuzzy inference rule base;

[0123] The roulette wheel selection algorithm is used to select some individuals from the current initial population as parent individuals;

[0124] Perform a crossover operation on the parent individuals, randomly select two parent individuals, randomly select crossover points on the chromosomes of the two parent individuals, exchange the gene fragments after the crossover points, and generate new offspring individuals;

[0125] Perform mutation operations on new offspring individuals with random probability, randomly select the gene positions of new offspring individuals, iteratively merge parent individuals and new offspring individuals to form a new population, until the preset evolutionary generation is reached or the convergence condition is met;

[0126] Select the individual with the highest fitness value from the new population as the optimal combination solution.

[0127] Exemplarily, the parameters of the initial membership function and the weights of the inference rules in the fuzzy inference rule base are defined as chromosomes of the genetic algorithm. Each chromosome represents a possible solution, including all membership function parameters and rule weights.

[0128] The initial population is randomly generated, and each individual (chromosome) is a set of specific membership function parameters and inference rule weights. The population size is usually determined by the complexity of the problem and computing resources.

[0129] Based on the membership function parameters and the inference rule weights, a fitness function is constructed. The fitness function evaluates the performance of each individual and is usually related to the accuracy of the diagnosis. Evaluation criteria: including indicators such as diagnostic accuracy, misdiagnosis rate, and missed diagnosis rate.

[0130] The roulette wheel selection algorithm is used to select some individuals from the current population as parents. This selection method is based on the fitness value of the individual, and individuals with high fitness are more likely to be selected. Two parent individuals are randomly selected. The crossover points are randomly selected on the chromosomes of the two parent individuals, and the gene fragments after the crossover points are exchanged to generate new offspring individuals.

[0131] Perform mutation operations with a certain random probability to increase the diversity of the population. Randomly select a gene position of the new offspring individual and change its value (for example, adjust the parameters of the membership function or the weight of the rule).

[0132] The parent individuals and the new offspring individuals are merged to form a new population. The selection, crossover and mutation operations are repeated until the preset evolutionary generations are reached or a certain convergence condition is met (such as a small improvement in fitness). In each generation, the fitness of all individuals is evaluated. The individual with the highest fitness value is selected from the final population as the optimal solution.

[0133] Among them, constructing a fitness function usually involves synthesizing multiple performance indicators into a single score in order to evaluate different candidate solutions in a genetic algorithm. The fitness function of the present application is used to evaluate the classification accuracy and misclassification cost based on a fuzzy logic system. The goal of the fitness function is to maximize the accuracy while minimizing the misclassification cost. Therefore, a fitness function can be designed in which the accuracy is given priority, but the misclassification cost is also included in the calculation to adjust the accuracy score. Corresponding weight coefficients can be assigned to the classification accuracy and the misclassification cost, respectively, to adjust the relative importance of the accuracy and the misclassification cost in the fitness function.

[0134] S103. A tumor boundary extraction method with three-dimensional morphological constraints is introduced to track the tumor boundary of the tumor target area in three-dimensional space, and the three-dimensional tumor boundary is extracted by combining local shape features and global shape prior knowledge; a three-dimensional visualization algorithm based on ray casting and surface reconstruction is used, and multi-view projection and three-dimensional texture mapping technology are comprehensively utilized to achieve three-dimensional tumor morphological reconstruction.

[0135] In an optional embodiment,

[0136] The tumor boundary extraction method with 3D morphological constraints is introduced to track the tumor boundary of the tumor target area in 3D space. The 3D tumor boundary is extracted by combining local shape features and global shape prior knowledge, including:

[0137] Based on the determined tumor target area, an initial tumor boundary surface is generated by performing local Gaussian fitting on the three-dimensional morphology of the tumor target area;

[0138] Extracting local shape features of each vertex on the initial tumor boundary surface, wherein the local shape features include at least one of curvature, normal vector and shape index;

[0139] Based on the local shape features and the predetermined a priori global shape features, a three-dimensional morphological constraint energy function is constructed to evaluate the degree of matching between the current boundary surface and the actual tumor boundary;

[0140] By using variational inference and surface evolution technology, driven by the three-dimensional morphological constraint energy function, the initial tumor boundary surface is iteratively adjusted and optimized until the evolution of the initial tumor boundary surface converges or reaches a preset number of iterations, thereby obtaining the final three-dimensional tumor boundary.

[0141] In an optional embodiment,

[0142] Based on the local shape features and the predetermined prior global shape features, constructing a three-dimensional morphological constraint energy function includes:

[0143] ;

[0144] ;

[0145] ;

[0146] in, E represents the three-dimensional morphological constraint energy function, α and β represent the first weight value corresponding to the local shape feature function and the second weight value corresponding to the prior global shape feature function, respectively;

[0147] E local represents the local shape characteristic function, Nrepresents the number of vertices of the initial tumor boundary surface, f i Indicates i The local shape features corresponding to the vertices are F The covariance matrix representing the local shape features;

[0148] E global represents the prior global shape feature function, p represents the shape parameter of the initial tumor boundary surface, represents the mean of the shape parameters.

[0149] Exemplarily, first, three-dimensional image data of the tumor is acquired using medical imaging technology (such as CT or MRI), and a target area of ​​the tumor is determined by a clinical radiologist or using an automated algorithm.

[0150] Apply a local Gaussian fitting method to the tumor target area. This step involves mathematically modeling the morphology of the tumor area in three-dimensional space to generate an initial tumor boundary surface that preliminarily reflects the shape and size of the tumor. Calculate local shape features of vertices: At each vertex on the initial tumor boundary surface, calculate local shape features such as curvature, normal vector, and shape index. These features help describe the geometric and morphological properties of each point on the surface. Selection of shape features: Select appropriate shape features or feature combinations based on specific application requirements to best reflect local changes in tumor boundaries.

[0151] Based on the predetermined global shape features (which can be based on a tumor shape database obtained through statistical learning), an energy function is constructed. This function will evaluate the degree of match between the current boundary surface and the actual tumor boundary. The energy function is designed to minimize the deviation from the actual tumor boundary. This usually involves weighting the shape features and prior knowledge.

[0152] A variational inference method is applied to estimate the most likely tumor boundary, which seeks the optimal solution by minimizing an energy function. Driven by the energy function, the initial tumor boundary surface is iteratively adjusted through surface evolution techniques. This includes smoothing, stretching, or compressing the surface to better fit the actual shape of the tumor. The surface adjustment and energy evaluation process is repeated until the evolution of the tumor boundary surface converges or a preset number of iterations is reached. Appropriate stopping criteria are set, such as reaching a minimum energy value, reaching a maximum number of iterations, or the surface change is below a certain threshold.

[0153] In an optional embodiment,

[0154] The 3D visualization algorithm based on ray casting and surface reconstruction uses multi-view projection and 3D texture mapping technology to achieve 3D morphological reconstruction of tumors, including:

[0155] A virtual light source is set inside the tumor target area, and a ray projection algorithm is used to project the tumor target area from multiple perspectives to generate a set of two-dimensional projection images;

[0156] Perform image segmentation on each two-dimensional projection image, extract the tumor boundary contour, and obtain a set of two-dimensional contour curves;

[0157] Based on the geometric relationship of ray projection, the two-dimensional contour curve is back-projected into three-dimensional space to obtain a set of three-dimensional space curves, wherein the three-dimensional space curves represent the contour of the tumor surface at different viewing angles;

[0158] Using a 3D space curve interpolation algorithm, discrete 3D space curves are connected to generate a tumor surface mesh model;

[0159] Using multi-view texture mapping technology, the tumor texture images at different viewpoints are mapped to the tumor surface mesh model to obtain the three-dimensional morphology of the tumor.

[0160] Exemplarily, one or more virtual light sources are set inside or around the tumor target area. These light sources are used to illuminate the tumor and emit light from different angles. A ray casting algorithm is used to project the tumor target area from multiple perspectives. This involves calculating the path of light from the light source, passing through the tumor area, and hitting the virtual detector at different angles. The information of each intersection of light and the tumor is recorded to form a set of two-dimensional projection images. These images contain image data of the tumor observed from different angles.

[0161] Image segmentation techniques are applied to each projection image to extract the boundary contour of the tumor. This can be achieved through thresholding, edge detection, or other image segmentation algorithms. Clear two-dimensional contour curves are extracted from the segmentation results, which represent the contour of the tumor at each projection perspective. The two-dimensional contour curves are back-projected back into three-dimensional space using the geometric relationship when the light is projected. This step requires the precise calculation of the position of each contour point in three-dimensional space based on the path of the light and the projection angle. The set of points obtained by back-projection is connected into a set of curves in three-dimensional space, which describe the three-dimensional contour of the tumor at different perspectives.

[0162] Use curve interpolation algorithms (such as B-spline or Bezier curves) to connect discrete 3D space curves to form a continuous tumor surface mesh model. Map tumor texture images from different perspectives onto the tumor surface mesh model. This step requires processing texture coordinate alignment and texture synthesis to ensure a natural transition and display of the texture on the 3D model. After texture mapping is completed, the 3D model of the tumor can be rendered and viewed in 3D visualization software, showing the complete morphology and texture details of the tumor.

[0163] By registering and fusing brain medical images of different modalities and using a weighted fusion strategy to adaptively determine the fusion weights, a fused brain image is obtained, which effectively integrates the complementary information of images of different modalities, improves the recognizability of brain tissue structure, and lays the foundation for subsequent tumor area extraction.

[0164] An adaptive fuzzy reasoning optimization strategy is introduced to dynamically adjust the membership function and reasoning rules according to the tumor candidate feature set, generate the optimal fuzzy reasoning parameters, and adaptively determine the tumor target area, which overcomes the limitations of traditional fixed fuzzy rules and improves the adaptability and flexibility of tumor area identification.

[0165] This application proposes a tumor boundary extraction method with three-dimensional morphology constraints, which tracks the boundaries of the tumor target area in three-dimensional space, combines local shape features and global shape prior knowledge, and realizes accurate three-dimensional tumor boundary extraction, overcomes the limitations of traditional two-dimensional segmentation methods, and improves the spatial accuracy of tumor boundary description.

[0166] By adopting a three-dimensional visualization algorithm based on ray projection and surface reconstruction, and comprehensively utilizing multi-view projection and three-dimensional texture mapping technology, realistic three-dimensional tumor morphology reconstruction is achieved, providing doctors with intuitive and three-dimensional means of observing and analyzing tumor morphology, and improving the visualization level of diagnosis and treatment plan formulation.

[0167] Figure 2 FIG. 1 is a schematic diagram of the structure of a three-dimensional reconstruction detection system for brain tumor images according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0168] The first unit is used to register and fuse brain medical images of different modalities, and use the weighted fusion strategy to adaptively determine the fusion weight to obtain a fused brain image; the fused brain image is preprocessed using an adaptive threshold segmentation algorithm, and the optimal threshold is adaptively determined according to the grayscale distribution of the fused brain image, and the brain tissue area of ​​the fused brain image is extracted by combining morphological filtering and connected domain analysis;

[0169] The second unit is used to build a tumor candidate region extraction model with multi-scale feature integration, extract feature vectors of brain tissue regions at different scales, use feature selection algorithms to optimize feature subsets, and determine tumor candidate feature sets; according to the adaptive fuzzy reasoning optimization strategy, dynamically adjust the membership function and reasoning rules, adaptively generate the optimal fuzzy reasoning parameters based on the tumor candidate feature set, and determine the tumor target region;

[0170] The third unit is used to introduce a tumor boundary extraction method with three-dimensional morphological constraints, track the tumor boundary of the tumor target area in three-dimensional space, and extract the three-dimensional tumor boundary by combining local shape features and global shape prior knowledge; a three-dimensional visualization algorithm based on ray casting and surface reconstruction, which comprehensively utilizes multi-view projection and three-dimensional texture mapping technology to achieve three-dimensional morphological reconstruction of the tumor.

[0171] According to a third aspect of the embodiments of the present invention,

[0172] An electronic device is provided, comprising:

[0173] processor;

[0174] a memory for storing processor-executable instructions;

[0175] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0176] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0177] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A three-dimensional reconstruction detection method for brain tumor images, characterized in that: include: Register and fuse brain medical images of different modalities, use weighted fusion strategy to adaptively determine fusion weights, and obtain fused brain images; Adopting adaptive threshold segmentation algorithm to preprocess the fused brain image, the optimal threshold is adaptively determined according to the grayscale distribution of the fused brain image, and the brain tissue area of ​​the fused brain image is extracted by combining morphological filtering and connected domain analysis. Construct a tumor candidate region extraction model with multi-scale feature integration, extract feature vectors of brain tissue regions at different scales, use feature selection algorithm to optimize feature subsets, and determine tumor candidate feature sets; according to the adaptive fuzzy reasoning optimization strategy, dynamically adjust the membership function and reasoning rules, adaptively generate the optimal fuzzy reasoning parameters based on the tumor candidate feature set, and determine the tumor target region; A tumor boundary extraction method with 3D morphological constraints is introduced to track the tumor boundary of the tumor target area in 3D space, and the 3D tumor boundary is extracted by combining local shape features and global shape prior knowledge. A 3D visualization algorithm based on ray casting and surface reconstruction is used to realize 3D tumor morphological reconstruction by comprehensively utilizing multi-view projection and 3D texture mapping technology. According to the adaptive fuzzy inference optimization strategy, the membership function and inference rules are dynamically adjusted, and the optimal fuzzy inference parameters are adaptively generated according to the tumor candidate feature set to determine the tumor target area including: Cluster analysis is performed on the candidate tumor feature set to divide the brain tissue area into several clusters, each cluster representing a tumor feature pattern; For each candidate tumor feature set corresponding to each cluster, the mean and standard deviation of the feature values ​​of the candidate tumor feature set are calculated to construct an initial membership function, where the initial membership function is used to represent the similarity between the clusters; Based on expert knowledge and historical case data, an initial fuzzy inference rule base is constructed to associate the candidate tumor feature set with its corresponding tumor type; Optimizing the parameters of the initial membership function and the weights of the inference rules in the initial fuzzy inference rule base by using a genetic algorithm, evaluating the performance of multiple candidate solutions by using a fitness function, selecting the candidate solution with the largest fitness value corresponding to the fitness function as the optimal combination solution, and determining the optimized membership function and the optimized fuzzy inference rule by using the optimal combination solution; For each brain tissue region, the membership is calculated using the optimized membership function, and the membership is used as the input of the optimized fuzzy inference rule to determine the inference result, and the brain tissue region with the largest inference value corresponding to the inference result is used as the tumor target region; The parameters of the initial membership function and the weights of the inference rules in the initial fuzzy inference rule base are optimized by using a genetic algorithm, the performance of multiple candidate solutions is evaluated by using a fitness function, and the candidate solution with the largest fitness value corresponding to the fitness function is selected as the optimal combination solution. The optimized membership function and the optimized fuzzy inference rule are determined by using the optimal combination solution, including: Using the parameters of the initial membership function and the weights of the inference rules in the initial fuzzy inference rule base as chromosome codes of the genetic algorithm; An initial population is randomly generated, each individual corresponding to a combination of a set of parameters of an initial membership function and a weight of an inference rule in the initial fuzzy inference rule base; The roulette wheel selection algorithm is used to select some individuals from the current initial population as parent individuals; Perform a crossover operation on the parent individuals, randomly select two parent individuals, randomly select crossover points on the chromosomes of the two parent individuals, exchange the gene fragments after the crossover points, and generate new offspring individuals; Perform mutation operations on new offspring individuals with random probability, randomly select the gene positions of new offspring individuals, iteratively merge parent individuals and new offspring individuals to form a new population, until the preset evolutionary generation is reached or the convergence condition is met; Select the individual with the highest fitness value from the new population as the optimal combination solution; The 3D visualization algorithm based on ray casting and surface reconstruction uses multi-view projection and 3D texture mapping technology to achieve 3D morphological reconstruction of tumors, including: A virtual light source is set inside the tumor target area, and a ray projection algorithm is used to project the tumor target area from multiple perspectives to generate a set of two-dimensional projection images; Perform image segmentation on each two-dimensional projection image, extract the tumor boundary contour, and obtain a set of two-dimensional contour curves; Based on the geometric relationship of ray projection, the two-dimensional contour curve is back-projected into three-dimensional space to obtain a set of three-dimensional space curves, wherein the three-dimensional space curves represent the contour of the tumor surface at different viewing angles; Using a 3D space curve interpolation algorithm, discrete 3D space curves are connected to generate a tumor surface mesh model; Using multi-view texture mapping technology, the tumor texture images at different viewpoints are mapped to the tumor surface mesh model to obtain the three-dimensional morphology of the tumor.

2. The method according to claim 1, characterized in that The fused brain image is preprocessed using an adaptive threshold segmentation algorithm. The optimal threshold is adaptively determined according to the grayscale distribution of the fused brain image. The brain tissue areas of the fused brain image are extracted by combining morphological filtering and connected domain analysis, including: Calculating the grayscale histogram of the fused brain image, using the maximum inter-class variance method to adaptively determine the optimal segmentation threshold, and performing binary segmentation on the fused brain image based on the optimal segmentation threshold to obtain a foreground area and a background area; Performing a morphological opening operation on the foreground area to remove noise and holes in the foreground area and smooth the boundary of the foreground area to obtain a modified foreground area; Adopting a connected domain analysis method based on region growing, extracting the maximum connected brain tissue region of the modified foreground region according to pixel grayscale similarity and spatial adjacency; Performing a morphological closing operation on the maximum connected brain tissue region to fill the holes in the maximum connected brain tissue region to obtain an initial brain tissue region; Combined with prior knowledge of brain anatomical structure, morphological filtering and boundary correction are performed on the initial brain tissue region to remove the residual non-brain tissue region, thereby obtaining the brain tissue region of the fused brain image.

3. The method according to claim 2, characterized in that The connected domain analysis method based on region growing is adopted to extract the largest connected brain tissue region of the modified foreground region according to the pixel grayscale similarity and spatial adjacency relationship, including: In the modified foreground area, randomly selecting a pixel point as an initial seed point, and marking it as an original connected brain tissue area; Calculating the grayscale difference between the initial seed point and all unmarked pixels in its neighborhood, incorporating pixels whose grayscale difference is less than a preset grayscale threshold into the original connected brain tissue region, and marking them as visited; For the newly added pixels in the original connected brain tissue area, the grayscale difference is recursively calculated until no new pixels are included in the original connected brain tissue area; Randomly selecting other pixel points as initial seed points until all pixel points in the modified foreground area are marked as visited, thereby obtaining multiple connected brain tissue areas; The numbers of pixels in the multiple connected brain tissue regions are compared, and the connected region with the largest number of pixels is selected as the largest connected brain tissue region.

4. The method according to claim 1, characterized in that: The tumor boundary extraction method with 3D morphological constraints is introduced to track the tumor boundary of the tumor target area in 3D space. The 3D tumor boundary is extracted by combining local shape features and global shape prior knowledge, including: Based on the determined tumor target area, an initial tumor boundary surface is generated by performing local Gaussian fitting on the three-dimensional morphology of the tumor target area; Extracting local shape features of each vertex on the initial tumor boundary surface, wherein the local shape features include at least one of curvature, normal vector and shape index; Based on the local shape features and the predetermined a priori global shape features, a three-dimensional morphological constraint energy function is constructed to evaluate the degree of matching between the current boundary surface and the actual tumor boundary; By using variational inference and surface evolution technology, driven by the three-dimensional morphological constraint energy function, the initial tumor boundary surface is iteratively adjusted and optimized until the evolution of the initial tumor boundary surface converges or reaches a preset number of iterations, thereby obtaining the final three-dimensional tumor boundary.

5. The method according to claim 4, characterized in that Based on the local shape features and the predetermined prior global shape features, constructing a three-dimensional morphological constraint energy function includes: ; ; ; in, E represents the three-dimensional morphological constraint energy function, α and β represent the first weight value corresponding to the local shape feature function and the second weight value corresponding to the prior global shape feature function, respectively; E local represents the local shape characteristic function, N represents the number of vertices of the initial tumor boundary surface, f i Indicates i The local shape features corresponding to the vertices are F The covariance matrix representing the local shape features; E global represents the prior global shape feature function, p represents the shape parameter of the initial tumor boundary surface, represents the mean of the shape parameters.

6. A three-dimensional reconstruction detection system for brain tumor images, used to implement the method according to any one of claims 1 to 5, characterized in that: include: The first unit is used to register and fuse brain medical images of different modalities, and use the weighted fusion strategy to adaptively determine the fusion weight to obtain a fused brain image; Adopting adaptive threshold segmentation algorithm to preprocess the fused brain image, the optimal threshold is adaptively determined according to the grayscale distribution of the fused brain image, and the brain tissue area of ​​the fused brain image is extracted by combining morphological filtering and connected domain analysis. The second unit is used to build a tumor candidate region extraction model with multi-scale feature integration, extract feature vectors of brain tissue regions at different scales, use feature selection algorithms to optimize feature subsets, and determine tumor candidate feature sets; according to the adaptive fuzzy reasoning optimization strategy, dynamically adjust the membership function and reasoning rules, adaptively generate the optimal fuzzy reasoning parameters based on the tumor candidate feature set, and determine the tumor target region; The third unit is used to introduce a tumor boundary extraction method with three-dimensional morphological constraints, track the tumor boundary of the tumor target area in three-dimensional space, and extract the three-dimensional tumor boundary by combining local shape features and global shape prior knowledge; a three-dimensional visualization algorithm based on ray casting and surface reconstruction, which comprehensively utilizes multi-view projection and three-dimensional texture mapping technology to achieve three-dimensional morphological reconstruction of the tumor.

7. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 5.

Citation Information

Patent Citations

  • Level set medical image segmentation method based on heredity kernel fuzzy clustering

    CN103366379A

  • Volume data non-photorealistic rendering method based on GPU (graphic processing unit) acceleration

    CN103366395A

  • Quick volume data skeleton extraction method based on rendering

    CN104156997A

  • SAR image change detecting method based on genetic kernel fuzzy clustering

    CN104268574A

  • Prior boundary-fused brain tumor segmentation method

    CN113160142A