Cerebrovascular unsupervised multi-view automatic segmentation method, equipment, medium and product

Abstract

The invention discloses a brain blood vessel unsupervised multi-view automatic segmentation method and device, a medium and a product, and relates to the field of image segmentation, and the method comprises the steps: carrying out the segmentation of a brain blood vessel based on a multi-mode magnetic resonance three-dimensional image (such as T1WI, TOF-MRA and the like); a multi-scale Frangi filtering method is adopted to construct a Frangi filtering view angle group of the enhanced image, a multi-scale Sto filtering method is adopted to construct a Sto filtering view angle group, a multi-scale vascular diffusion enhancement method is adopted to construct a diffusion enhancement view angle group, and a Dirichlet process hybrid model is adopted to generate a vascular classification result view angle group; and carrying out view angle image fusion on the obtained view angle groups by adopting a multi-view angle fusion method based on map segmentation to obtain a cerebrovascular segmentation image, and optimizing a segmentation result through post-processing. According to the invention, accurate and stable cerebrovascular segmentation images can be obtained in an unsupervised manner, and the method can be applied to magnetic resonance angiography images with different modals and different scanning parameters, and has high robustness.

Description

Technical Field

[0001] The present application relates to the field of image segmentation, and in particular to a method, device, medium and product for unsupervised multi-view automatic segmentation of cerebral blood vessels. Background Art

[0002] As the energy source of the central nervous system, cerebral blood vessels play an important role in the homeostasis of brain function. Cerebrovascular aging is a key link in brain aging and an important factor causing neurodegenerative diseases. The development of magnetic resonance vascular imaging technology has enabled non-invasive and repeated observation of individual cerebrovascular structures in clinical research. However, due to the complexity of the cerebrovascular network structure, in actual work, radiologists usually make subjective qualitative judgments on magnetic resonance vascular images based on professional knowledge and experience, resulting in limited group-level research and clinical precision diagnosis of cerebrovascular structure at this stage.

[0003] At present, the three-dimensional segmentation and reconstruction methods of cerebral blood vessels can be roughly divided into two categories. One is the unsupervised machine learning method, whose representative methods are Frangi filtering and Sato filtering. These methods are based on the understanding of the single characteristics of cerebral vascular signals, and often cannot accurately and stably identify cerebral vascular structures under multi-scale and multi-modal conditions; the other is the method based on deep neural networks. Due to the complexity of the cerebral vascular network structure and the lack of multi-modal cerebral vascular MRI data, there is still a lack of high-quality standardized cerebral vascular multi-modal MRI annotation datasets. In addition, due to the limitations of model architecture design and the principle of magnetic resonance vascular imaging, the existing methods generally have poor results in the segmentation and recognition of small blood vessels. In addition, due to the semi-quantitative nature of magnetic resonance imaging (i.e., signal intensity is affected by equipment and parameters rather than absolute values), the scanning equipment and scanning parameter settings of different centers will cause significant changes in image contrast and quality, resulting in artifacts of magnetic resonance vascular imaging signals and inconsistency of vascular signals of different scales. Therefore, many existing method models are limited in their application and promotion due to the lack of robustness in migration and generalization in different scenarios. Summary of the invention

[0004] In order to solve the above problems, the present application provides a method, device, medium and product for unsupervised multi-view automatic segmentation of cerebral blood vessels.

[0005] To achieve the above objectives, this application provides the following solutions: In a first aspect, the present application provides an unsupervised multi-view automatic segmentation method for cerebral blood vessels, comprising: Obtain multimodal 3D MRI images of the brain; multimodal 3D MRI images include T1 weighted imaging (T1 WI) and time-of-flight Magnetic Resonance Angiography (TOF-MRA); A Frangi filter viewing angle group is constructed based on the multi-modal magnetic resonance three-dimensional image using a multi-scale Frangi filter method; A multi-scale Sato filtering method is used to construct a Sato filtering viewing angle group based on the multi-modal magnetic resonance three-dimensional image; A multi-scale vascular diffusion enhancement method is used to construct a multi-scale vascular diffusion enhancement viewing angle group based on the multi-modal magnetic resonance three-dimensional image; Using a Dirichlet process mixture model to construct a blood vessel classification result perspective group based on the multimodal magnetic resonance three-dimensional image; A multi-view fusion method based on atlas segmentation is used to fuse the Frangi filter view group, the Sato filter view group, the multi-scale vascular diffusion enhancement view group and the vascular classification result view group, and a cerebral vascular segmentation image is obtained through post-processing optimization.

[0006] Optionally, a multi-scale Frangi filtering method is used to construct a Frangi filtering viewing angle group based on the multi-modal magnetic resonance three-dimensional image, including: In the multimodal magnetic resonance three-dimensional image space, voxel points are obtained, and a voxel-level Hessian matrix is ​​constructed based on the voxel points; Performing eigenvalue decomposition on the Hessian matrix and determining the absolute value of the eigenvalue obtained by the decomposition; Arrange the eigenvalues ​​in descending order according to the absolute values ​​to obtain an absolute value sequence of the eigenvalues; constructing a vascular enhancement response value based on the absolute values ​​of all eigenvalues ​​in the absolute value sequence of the eigenvalues, and determining probability information of the existence of a tubular continuum structure at different scales at each voxel point in the multimodal magnetic resonance three-dimensional image space based on the vascular enhancement response value; Based on the probability information, a three-dimensional blood vessel probability distribution matrix having the same dimension as the multimodal magnetic resonance three-dimensional image is generated at each scale, and the three-dimensional blood vessel probability distribution matrix is ​​used as a Frangi filter viewing angle at the corresponding scale; The Frangi filter view group is formed based on the Frangi filter view at all scales.

[0007] Optionally, a multi-scale Sato filtering method is used to construct a Sato filtering viewing angle group based on the multi-modal magnetic resonance three-dimensional image, including: In the multimodal magnetic resonance three-dimensional image space, voxel points are obtained, and a voxel-level Hessian matrix is ​​constructed based on the voxel points; Performing eigenvalue decomposition on the Hessian matrix, and sorting the eigenvalues ​​to form an eigenvalue sequence; Constructing a Sato vascular enhancement response value based on all eigenvalues ​​in the eigenvalue sequence, and determining a Sato probability value of the existence of a tubular continuum structure at different scales at each voxel point in the multimodal magnetic resonance three-dimensional image space based on the Sato vascular enhancement response value; Based on the Sato probability value, a three-dimensional blood vessel probability distribution matrix having the same dimension as the multimodal magnetic resonance three-dimensional image is generated at each scale, and the three-dimensional blood vessel probability distribution matrix is ​​used as the Sato filter perspective at the corresponding scale; The Sato filter perspective group is formed based on the Sato filter perspectives at all scales.

[0008] Optionally, a multi-scale vascular diffusion enhancement method is used to construct a multi-scale vascular diffusion enhancement viewing angle group based on the multi-modal magnetic resonance three-dimensional image, including: In the multimodal magnetic resonance three-dimensional image space, voxel points are obtained, and a voxel-level Hessian matrix is ​​constructed based on the voxel points; According to the Hessian matrix, different scales are selected, and smoothing and enhancement processing is performed based on the Frangi filter results to establish the vascular response function at different scales; constructing a three-dimensional diffusion tensor using the vascular response function and the Hessian matrix; The three-dimensional diffusion tensor is used to perform specific diffusion enhancement along the axial direction of the blood vessel, and after a preset diffusion iteration time, a three-dimensional blood vessel probability distribution matrix with the same dimension as the multi-modal magnetic resonance three-dimensional image is generated at each scale, and the three-dimensional blood vessel probability distribution matrix is ​​used as a multi-scale blood vessel diffusion enhancement perspective at a corresponding scale; The multi-scale vascular diffusion enhancement viewing angle group is formed based on the multi-scale vascular diffusion enhancement viewing angles at all scales.

[0009] Optionally, the vascular response function at different scales is expressed as: ; Where exp( ) represents the exponential function, , , and are all hyperparameters, , and are the eigenvalues ​​of the Hessian matrix, e is the natural logarithm, Used to distinguish between disc-like and linear structures, For identification of globular structures, = Indicates structural strength.

[0010] Optionally, a Dirichlet process mixture model is used to construct a blood vessel classification result perspective group based on the multimodal magnetic resonance three-dimensional image, including: Random sampling from the Dirichlet process yields a Dirichlet distribution and generates a probability density function for the mixed distribution; Sampling a categorical variable from the mixed distribution, and sampling a voxel signal value from a corresponding multivariate Gaussian distribution based on the categorical variable; using the categorical variable as a cluster label; Using Gibbs sampling method, a blood vessel classification result is obtained based on the probability density function and the voxel signal value; The blood vessel classification result viewing angle group is constructed based on the blood vessel classification result.

[0011] Optionally, a multi-view fusion method based on atlas segmentation is used to fuse the Frangi filter view group, the Sato filter view group, the multi-scale vascular diffusion enhancement view group and the vascular classification result view group to obtain a cerebral vascular segmentation image, including: For each perspective blood vessel segmentation image in the Frangi filter perspective group, the Sato filter perspective group, the multi-scale blood vessel diffusion enhancement perspective group, and the blood vessel classification result perspective group, a similarity matrix between voxels is constructed based on a Gaussian kernel function using a sparse representation method; Through the weighted average method, a unified similarity matrix is ​​obtained based on the similarity matrix ; According to the unified similarity matrix Degree of Certainty Matrix , and based on the unified similarity matrix Sum degree matrix Determine the graph Laplacian matrix; the graph Laplacian matrix is ; , where is the degree matrix element, To unify the similarity matrix elements, and represents a variable. In this formula, and Represents the row and column numbers of the matrix; Perform eigenvalue decomposition on the graph Laplace matrix, and arrange the decomposed eigenvalues ​​in descending order, retaining the top eigenvalues ​​and the previous The eigenvector corresponding to the eigenvalue is used to adjust the sensitivity of blood vessel recognition to obtain the cerebral blood vessel segmentation image; is the number of eigenvalues, as a parameter.

[0012] In a second aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-mentioned unsupervised multi-view automatic segmentation method for cerebral blood vessels.

[0013] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-mentioned unsupervised multi-view automatic segmentation method for cerebral blood vessels.

[0014] In a fourth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned unsupervised multi-view automatic segmentation method for cerebral blood vessels.

[0015] According to the specific embodiments provided in this application, this application has the following technical effects: The present application provides a method, device, medium and product for unsupervised multi-view automatic segmentation of cerebrovascular vessels, which comprehensively considers multiple essential common characteristics of cerebrovascular tissue in multimodal magnetic resonance images, and selects and constructs different feature capture methods (multi-scale Frangi filtering method, multi-scale Sato filtering method, multi-scale vascular diffusion enhancement method and Dirichlet process mixed model) from the perspectives of continuity of cerebrovascular signals at different scales, signal distribution differences, and tubular geometric characteristics. A multi-view fusion method is used to integrate cross-scale segmentation information of multiple feature perspectives, and the information is merged into a unified segmentation result without relying on large-scale annotated data. The information between each perspective can be interactively verified, so that the final cerebrovascular segmentation image has the advantages of high accuracy and strong generalization, and effectively solves the problems of inaccurate recognition of multi-scale and multi-modal cerebrovascular structures, poor segmentation of small blood vessels, and insufficient robustness due to signal artifacts and inconsistency in existing methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1A schematic diagram of a flow chart of an unsupervised multi-view automatic segmentation method for cerebral blood vessels provided in one embodiment of the present application; Figure 2 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0019] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0020] In an exemplary embodiment, the present application provides a method for unsupervised multi-view automatic segmentation of cerebral blood vessels, which is executed by a computer device, specifically, a computer device such as a terminal or a server, or a terminal and a server. In the present embodiment, the server is used as an example for explanation. Figure 1 As shown, the method includes: Step 100: Acquire a multi-modal 3D magnetic resonance image of the brain. The multi-modal 3D magnetic resonance image includes a T1-weighted imaging image and a TOF-MRA magnetic resonance image.

[0021] Step 101: construct a Frangi filter viewing angle group based on a multi-modal magnetic resonance three-dimensional image using a multi-scale Frangi filter method.

[0022] Step 102: construct a Sato filtering view group based on the multi-modal magnetic resonance three-dimensional image using a multi-scale Sato filtering method.

[0023] Step 103: construct a multi-scale vascular diffusion enhancement viewing angle group based on the multi-modal magnetic resonance three-dimensional image using a multi-scale vascular diffusion enhancement method.

[0024] Step 104: construct a blood vessel classification result perspective group based on the multimodal magnetic resonance three-dimensional image using a Dirichlet process mixture model.

[0025] Step 105: Use a multi-view fusion method based on atlas segmentation to fuse the Frangi filter view group, the Sato filter view group, the multi-scale vascular diffusion enhancement view group and the vascular classification result view group, and obtain a cerebral vascular segmentation image through post-processing optimization.

[0026] In another exemplary embodiment of the present application, the cerebrovascular signals in the three-dimensional images of multimodal magnetic resonance (T1-weighted imaging images and TOF-MRA magnetic resonance images) are all shown as tubular continuum structures in three-dimensional space. According to the essential characteristics of the cerebrovascular vessels, the present application adopts a multiscale Frangi filtering method to enhance and extract the cerebrovascular signals to form a segmented Frangi filtering perspective group. Based on this, in this embodiment, the implementation process of step 101 includes: 11. In the multimodal MRI three-dimensional image space, obtain voxel points and construct the voxel-level Hessian matrix based on the voxel points. For example: Calculate the second-order partial derivatives of the voxel points in the multimodal magnetic resonance three-dimensional image space and construct the Hessian matrix at the voxel level, which is expressed as: .

[0027] In the formula, is the Hessian matrix, is the spatial coordinate of a given voxel point, is defined at coordinates A multivariate function in three-dimensional space, is the symbol for the partial derivative operation.

[0028] 12. Perform eigenvalue decomposition on the Hessian matrix and determine the absolute value of the decomposed eigenvalue. The eigenvalue decomposition formula is: .

[0029] Where Λ is a diagonal matrix, and its diagonal elements are the eigenvalues ​​of the Hessian matrix, which are arranged in descending order of absolute value as follows: , , ....... is an eigenvector matrix, where each column vector represents the eigenvector corresponding to the corresponding eigenvalue. Represents the transpose of the eigenvector.

[0030] Take the absolute value of the eigenvalue of the Hessian matrix and mark it as , , ....... When taking the three eigenvalues ​​with the largest absolute values, we have: .

[0031] 13. Arrange the eigenvalues ​​in descending order according to their absolute values ​​to obtain the absolute value sequence of the eigenvalues.

[0032] 14. The first three in the absolute value sequence based on the eigenvalue ( , , ) The absolute value corresponding to the eigenvalue is used to calculate the vascular enhancement response value, and based on the response value, the probability information of the existence of tubular continuum structure at different scales at each voxel point in the multimodal MRI three-dimensional image space is determined. Among them, the statistic of structural strength The calculation formula is: , , .

[0033] In the formula, Represents the Hessian matrix The Euclidean norm of , is a statistic used to distinguish between disk-like and linear structures, is a statistic used to identify globular structures.

[0034] The probability information calculation formula of the tubular continuum structure is: .

[0035] Among them, the three-dimensional vector = represents a voxel in three-dimensional space, Represents the vessel probability value on the voxel. , , and These are all hyperparameters that need to be preset to control the sensitivity of the Frangi filter to spherical, tubular, and disk-like structures and background noise. = Indicates structural strength.

[0036] 15. Based on the probability information, a three-dimensional vascular probability distribution matrix with the same dimension as the multimodal magnetic resonance three-dimensional image is generated at each scale, and the three-dimensional vascular probability distribution matrix is ​​used as the Frangi filter perspective at the corresponding scale.

[0037] The classic Frangi filter segmentation method uses a simple strategy of taking the maximum probability when processing cross-scale blood vessel integration and identification, which will lead to obvious information loss. In order to solve this problem, this embodiment adopts a two-step strategy of cross-scale filtering and fusion separation, retains the probability information at each scale, and forms a Frangi filter perspective group according to different filter scales. The specific steps of generating the Frangi filter perspective group image are as follows: (1) The original image (i.e., multimodal MRI 3D image) is compared with the sThe three-dimensional Gaussian convolution kernel is used for convolution operation, and the specific operation process is as follows: .

[0038] in, is the result of the convolution operation, The scale is The result of the convolution operation of the three-dimensional Gaussian convolution kernel function. To act on voxels The scale on The three-dimensional Gaussian convolution kernel function is in the form of: .

[0039] in, is the dimension of the Gaussian function, for example =3, it represents three-dimensional space.

[0040] (2) Change the scale , at a given scale List ( 1, 2.… ) under the condition of the probability calculation of tubular continuum structure. Generate a three-dimensional blood vessel probability distribution matrix with the same dimension as the original image as the Frangi filter perspective at this scale.

[0041] 16. A Frangi filter perspective group is formed based on the Frangi filter perspectives at all scales for use in multi-perspective fusion.

[0042] In another exemplary embodiment of the present application, the above step 102 is mainly to construct a second-view image of blood vessel segmentation using a multi-scale Sato filtering method. After the original image is convolved using a three-dimensional Gaussian convolution kernel function with a scale of s [see the above step (1)], the implementation steps of step 102 include: 21. In the multimodal magnetic resonance three-dimensional image (T1-weighted imaging image and TOF-MRA magnetic resonance image) space, voxel points are obtained, and a voxel-level Hessian matrix is ​​constructed based on the voxel points. The specific process is the same as step 11 above.

[0043] 22. Perform eigenvalue decomposition on the Hessian matrix and sort the eigenvalues ​​to form an eigenvalue sequence. Before sorting the eigenvalues, the absolute value is not taken, and the positive and negative signs are considered for direct sorting, and we have: .

[0044] 23. Based on the eigenvalue sequence ( , , ) eigenvalues ​​to calculate the Sato probability value, and based on the probability value, determine the probability information of the existence of tubular continuum structure at different scales for each voxel point in the multimodal magnetic resonance three-dimensional image space. The calculation formula of the Sato probability value of the tubular continuum structure is: .

[0045] In the formula, The representative coordinates are The probability value of the blood vessel of the voxel point. The hyperparameter in the formula here is After preset, it is used to control the sensitivity of Sato filtering to background noise.

[0046] 24. Based on the Sato probability value, a three-dimensional vascular probability distribution matrix with the same dimension as the three-dimensional image of the multimodal magnetic resonance (T1-weighted imaging image and TOF-MRA magnetic resonance image) is generated at each scale, and the three-dimensional vascular probability distribution matrix is ​​used as the Sato filtering perspective at the corresponding scale.

[0047] The classic Sato filtering (Multiscale Sato Filtering) blood vessel segmentation method uses a simple strategy of taking the maximum probability when processing cross-scale blood vessel integration and identification, which will also result in obvious information loss. In order to solve this problem, in the Sato perspective construction process, this embodiment can also adopt a two-step strategy of cross-scale filtering and fusion separation, retain the probability information at each scale, and form a Sato filter perspective group according to different filtering scales, and then merge them uniformly in the later stage. Based on this, the specific workflow is as follows: Select a series of different Gaussian kernel function scales ( 1, 2.… ), according to the above-mentioned Sato probability value calculation formula of tubular continuum structure, each scale A three-dimensional vascular probability distribution matrix with the same dimension as the multimodal MRI three-dimensional image is generated as the Sato filter perspective at this scale.

[0048] 25. A Sato filter perspective group is formed based on the Sato filter perspectives at all scales.

[0049] In another exemplary embodiment of the present application, the cerebral vascular structure has an obvious continuity feature. Based on this feature, the present application adopts a multiscale vessel enhancement diffusion method and constructs a three-dimensional diffusion tensor using a Hessian matrix, so that the vessel segmentation label diffuses and grows along the axial direction of the vessel, suppresses the vertical extension of the label, and ensures the continuity of the label during vessel segmentation. Based on this, the implementation process of step 103 includes: 31. In the three-dimensional image space of multimodal magnetic resonance (T1-weighted imaging image and TOF-MRA magnetic resonance image), voxel points are obtained, and a voxel-level Hessian matrix is ​​constructed based on the voxel points. The specific process is the same as step 11 above.

[0050] 32. According to the Hessian matrix, different scales are selected, and the vascular response functions at different scales are established by smoothing and enhancing the Frangi filter provided above, thereby greatly reducing the interference of noise and non-vascular structures. Among them, the vascular response functions at different scales are expressed as: .

[0051] In the formula, exp( ) represents the exponential function. Here, the hyperparameter , , and After preset, it is used to control the smoothness of the vascular response function. e is the natural logarithm.

[0052] 33. Use the vascular response function and the Hessian matrix to construct a three-dimensional diffusion tensor. Among them: (1) Perform eigenvalue decomposition on the Hessian matrix to obtain the eigenvector matrix .

[0053] (2) Constructing the diffusion eigenvalue matrix based on the vascular response function , the matrix The elements on the diagonal are: .

[0054] .

[0055] In the formula, is the vascular response function, , and Both represent matrices Elements on the diagonal line. and It is a hyperparameter that needs to be preset to control the intensity and speed of diffusion in different directions, as well as the sensitivity to vascular tissue.

[0056] 34. Use the three-dimensional diffusion tensor to perform axial-specific diffusion of the labeled blood vessels. After a preset time t, a distribution matrix with the same dimension as the multi-modal magnetic resonance three-dimensional image is generated at each scale as the multi-scale blood vessel diffusion enhancement perspective at the corresponding scale. The diffusion formula is expressed as: .

[0057] In the formula, is the Nabla operator, and the output is a vector of partial derivatives of each variable of the multivariate function. Represents the diffusion time. is the result of the convolution operation mentioned above, which is the original image after being convolved with the scale The processed image obtained by convolution operation with the three-dimensional Gaussian convolution kernel is represented as a multivariate function in three-dimensional space; the dimension of the Gaussian function is In the above formula, represents a three-dimensional diffusion tensor, which is determined by the vascular response function and the Hessian matrix. Represents the divergence operation, and the output is the average rate of change of the local signal of the convolved image in three-dimensional space.

[0058] Select a series of different Gaussian kernel function scales ( 1, 2.… ), according to the above diffusion formula at their respective scales, after After time diffusion, each scale A three-dimensional vascular probability distribution matrix of the same dimension as the multi-modal MRI three-dimensional image is generated as a multi-scale vascular diffusion enhancement perspective at this scale.

[0059] 35. Based on the multi-scale vascular diffusion enhancement perspectives at all scales, a multi-scale vascular diffusion enhancement perspective group is formed.

[0060] In another exemplary embodiment of the present application, another basic feature of cerebrovascular tissue is that it has a relatively stronger signal on the magnetic resonance vascular imaging image relative to non-vascular tissue, forming a signal contrast. According to this characteristic of cerebrovascular signals, the present application performs unsupervised clustering analysis based on the signal intensity values ​​of all voxels to identify vascular tissue. However, the hyperparameters in the unsupervised clustering analysis (such as the number of categories) will significantly affect the clustering effect. In order to solve this problem, the present application uses the Dirichlet Process Mixture Model to adaptively determine the number of categories according to the properties of the image itself, thereby achieving optimal clustering. Among them, the input information of the Dirichlet Process Mixture Model is the magnetic resonance signal intensity value of each voxel of the TOF-MRA magnetic resonance vascular imaging image. Based on this, the implementation process of step 104 provided above in the present application may include: 41. Randomly sample from the Dirichlet process to obtain the Dirichlet distribution and generate the probability density function of the Dirichlet distribution. For example, from the hyperparameter , the base distribution is The Dirichlet distribution is obtained by random sampling in the Dirichlet process. ,have: .

[0061] In the formula, Represents a random sampling operation.

[0062] Sampled Dirichlet distribution The probability density function of the categorical variable )for: .

[0063] In the formula, is the probability density function, For about Beta distribution, for The dimension of For the Dirichlet distribution The corresponding hyperparameters.

[0064] 42. Randomly sample the value of the categorical variable from the Dirichlet distribution (denoted as ), which is used as the parameter of the multivariate distribution of each voxel and is derived from the parameter Generates the distribution of voxel signals in a multivariate distribution of .

[0065] 43. The Gibbs sampling method (which has been implemented in the PyMC python package) is used for iterative calculation to obtain the vascular classification results based on the probability density function and voxel signal value.

[0066] 44. Construct a three-dimensional classification probability map based on the blood vessel classification results, and construct a blood vessel classification result perspective group. The blood vessel classification results obtained by the Dirichlet process mixture model are used as the fourth perspective of blood vessel segmentation.

[0067] In another exemplary embodiment of the present application, the present application constructs four cerebrovascular enhanced segmentation view groups (corresponding to the view groups constructed in step 101, step 102, step 103 and step 104) based on the different basic signal characteristics of cerebrovascular in magnetic resonance images. On this basis, the present application uses a multi-view fusion method based on atlas segmentation to fuse the view images of the four cerebrovascular enhanced segmentation view groups, and finally constructs a unified three-dimensional cerebrovascular segmentation image (that is, the cerebrovascular segmentation image finally obtained by the present application). 51. In the multi-view fusion method, in each view vascular segmentation image in the Frangi filter view group, the Sato filter view group, the multi-scale vascular diffusion enhancement view group and the vascular classification result view group, a sparse representation method is used to construct a similarity matrix between voxels in each view.

[0068] 52. Similarity matrix between voxels in each perspective Get the unified similarity matrix , .

[0069] In the formula, Representative View Voxel and voxels The similarity between Representative View Voxel and voxels The similarity between . , Representative View Voxel and voxels The vascular probability value, Representative In the similarity matrix of perspective The similarity value of the elements. is the total number of viewing angles, is the number of voxels.

[0070] 53. Determine the degree matrix based on the unified similarity matrix, and determine the graph Laplacian matrix based on the unified similarity matrix and the degree matrix. The graph Laplacian matrix is ; , where is the degree matrix element, To unify the similarity matrix elements, and represents a variable. In this formula, and Represents the row and column numbers of the matrix.

[0071] For example, considering that the information between different perspectives is relevant, the results of each perspective are combined, and the similarity matrices of each perspective are weighted and summed to obtain a unified similarity matrix ,for: .

[0072] In the formula, Representative The similarity matrix of perspectives, Representative voxel The unified similarity of Representative voxel and voxels The weight value of each perspective The similarity matrix from this perspective With the unified similarity matrix The norm between is determined by: .

[0073] In the formula, Represents the norm.

[0074] 54. Perform eigenvalue decomposition on the graph Laplace matrix, and arrange the decomposed eigenvalues ​​in descending order, and keep the top The eigenvalues ​​and the eigenvectors corresponding to the retained eigenvalues ​​are used to adjust the sensitivity of blood vessel recognition and obtain a cerebral blood vessel segmentation image. is the number of eigenvalues, is a parameter. The final cerebral vascular segmentation result is expressed as: .

[0075] In the formula, is the final cerebral vascular segmentation result. Indicates an intermediate parameter.

[0076] Based on the above description, the unsupervised multi-view automatic segmentation method of cerebrovascular provided by the present application can realize the accurate and stable identification and segmentation of blood vessels in multimodal magnetic resonance vascular imaging data without labeling. This advantage comes from the fact that the present application adopts the strategy of multi-view fusion unsupervised machine learning, combines prior knowledge with mathematical modeling, and comprehensively considers multiple essential common characteristics of cerebrovascular tissue in multimodal magnetic resonance images. Starting from the continuity of cerebrovascular signals at different scales, differences in signal distribution, and tubular geometric features, different feature capture methods are selected and constructed, and the cross-scale segmentation information of multiple feature perspectives is integrated by a multi-view fusion method, and merged into a unified segmentation result. The information between each perspective can realize interactive verification. Therefore, the segmentation result has the advantages of accuracy, strong generalization, and no reliance on large-scale labeled data.

[0077] Furthermore, in order to address the problems of artifacts in magnetic resonance vascular imaging signals and the inconsistency of vascular signals of different scales, the present application sets up a "two-stage" algorithm that separates multi-scale filtering signal extraction and information integration, increases the perspective image information that can be used for integration, and can obtain stable and accurate segmentation performance on blood vessels of different scales.

[0078] In addition, this application also summarizes and analyzes the essential characteristics of cerebrovascular signals widely found in multimodal magnetic resonance images - continuity, tubularity, and signal distribution contrast. These essential characteristics are used as prior knowledge for the segmentation algorithm, and different signal extraction and recognition methods in machine learning are constructed to generate segmentation view groups corresponding to different cerebrovascular features. Finally, by using the technology of multi-view fusion, the segmentation results of multiple feature view groups are integrated and unified, and accurate and stable cerebrovascular segmentation images can be obtained unsupervisedly for application in magnetic resonance angiography images with different modalities and different scanning parameters.

[0079] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 2As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store cerebrovascular unsupervised multi-view automatic segmentation data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for unsupervised multi-view automatic segmentation of cerebrovascular is implemented.

[0080] Those skilled in the art will understand that Figure 2 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0081] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0082] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0083] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0084] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0085] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0086] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0087] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for unsupervised multi-view automatic segmentation of cerebral blood vessels, characterized in that: include: Acquire multimodal 3D magnetic resonance images of the brain; Multimodal 3D MRI images include T1-weighted imaging and TOF-MRA MRI images; A Frangi filter viewing angle group is constructed based on the multi-modal magnetic resonance three-dimensional image using a multi-scale Frangi filter method; A multi-scale Sato filtering method is used to construct a Sato filtering viewing angle group based on the multi-modal magnetic resonance three-dimensional image; A multi-scale vascular diffusion enhancement method is used to construct a multi-scale vascular diffusion enhancement viewing angle group based on the multi-modal magnetic resonance three-dimensional image; Using a Dirichlet process mixture model to construct a blood vessel classification result perspective group based on the multimodal magnetic resonance three-dimensional image; A multi-view fusion method based on atlas segmentation is used to fuse the Frangi filter view group, the Sato filter view group, the multi-scale vascular diffusion enhancement view group and the vascular classification result view group, and a cerebral vascular segmentation image is obtained through post-processing optimization.

2. The unsupervised multi-view automatic segmentation method of cerebral blood vessels according to claim 1, characterized in that: A multi-scale Frangi filtering method is used to construct a Frangi filtering viewing angle group based on the multi-modal magnetic resonance three-dimensional image, including: In the multimodal magnetic resonance three-dimensional image space, voxel points are obtained, and a voxel-level Hessian matrix is ​​constructed based on the voxel points; Performing eigenvalue decomposition on the Hessian matrix and determining the absolute value of the eigenvalue obtained by the decomposition; Arrange the eigenvalues ​​in descending order according to the absolute values ​​to obtain an absolute value sequence of the eigenvalues; constructing a vascular enhancement response value based on the absolute values ​​of all eigenvalues ​​in the absolute value sequence of the eigenvalues, and determining probability information of the existence of a tubular continuum structure at different scales at each voxel point in the multimodal magnetic resonance three-dimensional image space based on the vascular enhancement response value; Based on the probability information, a three-dimensional blood vessel probability distribution matrix having the same dimension as the multimodal magnetic resonance three-dimensional image is generated at each scale, and the three-dimensional blood vessel probability distribution matrix is ​​used as a Frangi filter viewing angle at the corresponding scale; The Frangi filter view group is formed based on the Frangi filter view at all scales.

3. The unsupervised multi-view automatic segmentation method of cerebral blood vessels according to claim 1, characterized in that: A multi-scale Sato filtering method is used to construct a Sato filtering perspective group based on the multi-modal magnetic resonance three-dimensional image, including: In the multimodal magnetic resonance three-dimensional image space, voxel points are obtained, and a voxel-level Hessian matrix is ​​constructed based on the voxel points; Performing eigenvalue decomposition on the Hessian matrix, and sorting the eigenvalues ​​to form an eigenvalue sequence; Constructing a Sato vascular enhancement response value based on all eigenvalues ​​in the eigenvalue sequence, and determining a Sato probability value of the existence of a tubular continuum structure at different scales at each voxel point in the multimodal magnetic resonance three-dimensional image space based on the Sato vascular enhancement response value; Based on the Sato probability value, a three-dimensional blood vessel probability distribution matrix having the same dimension as the multimodal magnetic resonance three-dimensional image is generated at each scale, and the three-dimensional blood vessel probability distribution matrix is ​​used as the Sato filter perspective at the corresponding scale; The Sato filter perspective group is formed based on the Sato filter perspectives at all scales.

4. The unsupervised multi-view automatic segmentation method of cerebral blood vessels according to claim 1, characterized in that: A multi-scale vascular diffusion enhancement method is used to construct a multi-scale vascular diffusion enhancement viewing angle group based on the multi-modal magnetic resonance three-dimensional image, including: In the multimodal magnetic resonance three-dimensional image space, voxel points are obtained, and a voxel-level Hessian matrix is ​​constructed based on the voxel points; According to the Hessian matrix, different scales are selected, and smoothing and enhancement processing is performed based on the Frangi filter results to establish the vascular response function at different scales; constructing a three-dimensional diffusion tensor using the vascular response function and the Hessian matrix; The three-dimensional diffusion tensor is used to perform specific diffusion enhancement along the axial direction of the blood vessel, and after a preset diffusion iteration time, a three-dimensional blood vessel probability distribution matrix with the same dimension as the multi-modal magnetic resonance three-dimensional image is generated at each scale, and the three-dimensional blood vessel probability distribution matrix is ​​used as a multi-scale blood vessel diffusion enhancement perspective at a corresponding scale; The multi-scale vascular diffusion enhancement viewing angle group is formed based on the multi-scale vascular diffusion enhancement viewing angles at all scales.

5. The unsupervised multi-view automatic segmentation method of cerebral blood vessels according to claim 4, characterized in that: The vascular response function at different scales is expressed as: ； In the formula, exp( ) represents the exponential function, , , and are all hyperparameters, , and are the eigenvalues ​​of the Hessian matrix, e is the natural logarithm, Used to distinguish disc-like and linear structures, For identification of globular structures, = Indicates structural strength.

6. The unsupervised multi-view automatic segmentation method of cerebral blood vessels according to claim 1, characterized in that: A Dirichlet process mixture model is used to construct a blood vessel classification result perspective group based on the multimodal magnetic resonance three-dimensional image, including: Random sampling from the Dirichlet process yields a Dirichlet distribution and generates a probability density function for the mixed distribution. Sampling a categorical variable from the mixed distribution, and sampling a voxel signal value from a corresponding multivariate Gaussian distribution based on the categorical variable; using the categorical variable as a cluster label; Using Gibbs sampling method, a blood vessel classification result is obtained based on the probability density function and the voxel signal value; The blood vessel classification result viewing angle group is constructed based on the blood vessel classification result.

7. The unsupervised multi-view automatic segmentation method of cerebral blood vessels according to claim 1, characterized in that: The Frangi filter perspective group, the Sato filter perspective group, the multi-scale vascular diffusion enhancement perspective group and the vascular classification result perspective group are subjected to perspective image fusion by using a multi-perspective fusion method based on atlas segmentation to obtain a cerebral vascular segmentation image, including: For each perspective blood vessel segmentation image in the Frangi filter perspective group, the Sato filter perspective group, the multi-scale blood vessel diffusion enhancement perspective group and the blood vessel classification result perspective group, a similarity matrix between voxels is constructed based on a Gaussian kernel function using a sparse representation method; Through the weighted average method, a unified similarity matrix is ​​obtained based on the similarity matrix ; According to the unified similarity matrix Degree of Certainty Matrix , and based on the unified similarity matrix Sum degree matrix Determine the graph Laplacian matrix; the graph Laplacian matrix is ; , where is the degree matrix element, To unify the similarity matrix elements, and represents a variable. In this formula, and Represents the row and column numbers of the matrix; Perform eigenvalue decomposition on the graph Laplace matrix, and arrange the decomposed eigenvalues ​​in descending order, retaining the top eigenvalues ​​and the previous The eigenvector corresponding to the eigenvalue is used to adjust the sensitivity of blood vessel recognition to obtain the cerebral blood vessel segmentation image; is the number of eigenvalues, as a parameter.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the unsupervised multi-view automatic segmentation method for cerebral blood vessels according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for unsupervised multi-view automatic segmentation of cerebral blood vessels described in any one of claims 1 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the unsupervised multi-view automatic segmentation method for cerebral blood vessels according to any one of claims 1 to 7 is implemented.

Images