Magnetic resonance image cerebrovascular segmentation method, system and device and medium

Abstract

The invention discloses a magnetic resonance image cerebral vessel segmentation method, system and device and a medium, and relates to the field of image processing, and the method comprises the steps: obtaining a to-be-processed magnetic resonance image; carrying out noise reduction and blood vessel specific enhancement processing on the to-be-processed magnetic resonance image to obtain a preprocessed magnetic resonance image; based on a deep neural network model, determining a cerebrovascular tissue in the preprocessed magnetic resonance image, and segmenting the cerebrovascular tissue to obtain a cerebrovascular segmented image; wherein the deep neural network model is obtained by adopting a training sample set for training in advance, and the training sample set comprises a plurality of sample preprocessing magnetic resonance images and a real label of each voxel in each sample preprocessing magnetic resonance image; the real label is a cerebrovascular tissue or a non-cerebrovascular tissue; the deep neural network model is to introduce an attention gating mechanism into jump connection of 3D-UNet. The method improves the segmentation precision of the brain blood vessel in the magnetic resonance image.

Description

Technical Field

[0001] The present application relates to the field of image processing, and in particular to a method, system, device and medium for segmenting cerebral blood vessels in magnetic resonance images. 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 neurodegeneration. The development of magnetic resonance vascular imaging technology has enabled non-invasive and repeated observation of the structure of individual cerebral blood vessels in clinical research. However, due to the complexity of the cerebrovascular network structure, in actual work, we can only rely on the subjective qualitative judgment of magnetic resonance vascular images by radiologists based on their professional knowledge and experience. Therefore, at this stage, group-level research on cerebrovascular structure and clinical precision diagnosis are still limited.

[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 often based on the recognition of the single characteristics of cerebral vascular signals, and often cannot accurately and stably identify the cerebral vascular structure under multi-scale and multi-modal conditions; the other is based on deep neural network methods. The neural network method can automatically extract richer and more dimensional vascular features, but due to the complexity of the cerebral vascular network structure and the lack of cerebral vascular magnetic resonance data, there is still a lack of high-quality standardized cerebral vascular magnetic resonance annotation datasets. At the same time, due to problems with model architecture design and the principle of magnetic resonance vascular imaging, the current methods are generally unsatisfactory in the segmentation and recognition of small blood vessels, and thus their clinical research application value is severely limited. Summary of the invention

[0004] The purpose of the present application is to provide a method, system, device and medium for segmenting cerebral blood vessels in magnetic resonance images, which can improve the segmentation accuracy of cerebral blood vessels in magnetic resonance images.

[0005] To achieve the above objectives, this application provides the following solutions: In a first aspect, the present application provides a method for segmenting cerebral blood vessels in a magnetic resonance image, comprising: acquiring a magnetic resonance image to be processed; Performing noise reduction and blood vessel-specific enhancement processing on the magnetic resonance image to be processed to obtain a preprocessed magnetic resonance image; Based on the deep neural network model, determining the cerebral vascular tissue in the preprocessed magnetic resonance image, and segmenting the cerebral vascular tissue to obtain a cerebral vascular segmentation image; Among them, the deep neural network model is obtained by pre-training using a training sample set, and the training sample set includes multiple sample pre-processed magnetic resonance images and the true label of each voxel in each sample pre-processed magnetic resonance image; the true label is cerebrovascular tissue or non-cerebrovascular tissue; the deep neural network model introduces an attention gating mechanism in the jump connection of 3D-UNet.

[0006] In a second aspect, the present application provides a magnetic resonance image cerebral blood vessel segmentation system, comprising: An image acquisition module, used for acquiring a magnetic resonance image to be processed; A preprocessing module, used for performing noise reduction and blood vessel-specific enhancement processing on the magnetic resonance image to be processed to obtain a preprocessed magnetic resonance image; An image segmentation module, used for determining the cerebral vascular tissue in the preprocessed magnetic resonance image based on a deep neural network model, and segmenting the cerebral vascular tissue to obtain a cerebral vascular segmentation image; Among them, the deep neural network model is obtained by pre-training using a training sample set, and the training sample set includes multiple sample pre-processed magnetic resonance images and the true label of each voxel in each sample pre-processed magnetic resonance image; the true label is cerebrovascular tissue or non-cerebrovascular tissue; the deep neural network model introduces an attention gating mechanism in the jump connection of 3D-UNet.

[0007] In a third 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 above-mentioned magnetic resonance image cerebral blood vessel segmentation method.

[0008] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the above-mentioned magnetic resonance image cerebral blood vessel segmentation method when executed by a processor.

[0009] According to the specific embodiments provided in this application, this application has the following technical effects: The present application provides a method, system, device and medium for cerebral blood vessel segmentation in magnetic resonance images. By performing noise reduction and blood vessel-specific enhancement processing on the magnetic resonance images to be processed, the artifact noise and the risk of imbalance in the segmentation of large and small blood vessels in the magnetic resonance images are reduced. The deep neural network model integrates 3D-UNet with an attention gating mechanism, which ensures the stability of the deep neural network model in cerebral vascular tissue segmentation and the sensitivity of small blood vessel identification, thereby improving the segmentation accuracy of cerebral blood vessels in magnetic resonance images. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] 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.

[0011] Figure 1 This is a diagram of the application environment of a method for segmenting cerebral blood vessels in a magnetic resonance image in one embodiment of the present application.

[0012] Figure 2 A flowchart of a method for segmenting cerebral blood vessels in a magnetic resonance image is provided in one embodiment of the present application.

[0013] Figure 3 Schematic diagram of the attention gating mechanism fusion process in one embodiment of the present application.

[0014] Figure 4 A schematic diagram of the functional modules of a magnetic resonance image cerebral blood vessel segmentation system provided in one embodiment of the present application.

[0015] Figure 5 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0016] 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 creative work are within the scope of protection of this application.

[0017] 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.

[0018] The magnetic resonance image cerebral blood vessel segmentation method provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, or it can be integrated on the server 104, or it can be placed on the cloud or other servers. The terminal 102 can send the magnetic resonance image to be processed to the server 104. After the server 104 receives the magnetic resonance image to be processed, it performs noise reduction and vascular specific enhancement processing on the magnetic resonance image to be processed, and segments the cerebral vascular tissue based on the deep neural network model to obtain a cerebral vascular segmentation image. The server 104 can feedback the obtained cerebral vascular segmentation image to the terminal 102. In addition, in some embodiments, the magnetic resonance image cerebral vascular segmentation method can also be implemented by the server 104 or the terminal 102 alone.

[0019] The terminal 102 may be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart vehicle-mounted devices, etc. The portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.

[0020] In an exemplary embodiment, Figure 2 As shown, a method for segmenting cerebral blood vessels in magnetic resonance images is provided. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used for explanation, and the steps include the following steps 201 to 203.

[0021] Step 201, obtaining a magnetic resonance image to be processed. The magnetic resonance image to be processed includes a T1-weighted image and a time-of-flight magnetic resonance image queue. The time-of-flight magnetic resonance image queue is obtained using the time-of-flight magnetic resonance angiography (TOF-MRA) technology.

[0022] Step 202: performing noise reduction and blood vessel-specific enhancement processing on the magnetic resonance image to be processed to obtain a preprocessed magnetic resonance image.

[0023] In an exemplary embodiment, step 202 includes steps 21 to 26 below.

[0024] Step 21, performing brain tissue specific identification and segmentation on the time-of-flight magnetic resonance image queue to obtain a brain tissue structure magnetic resonance image.

[0025] Specifically, the Brain Extraction Tool (BET) in the FMRIB Software Library (FSL) toolkit was used to perform brain tissue-specific identification and segmentation on TOF-MRA magnetic resonance images. The brain tissue and non-brain tissue were distinguished by the grayscale information and its spatial distribution in the magnetic resonance images, and non-brain tissue structures such as the scalp, skull, facial features and neck were removed. The stripping threshold was set to 0.3 to eliminate the interference of non-brain tissue structure image signals on subsequent processing.

[0026] BET detects the edge between the skull and brain tissue by calculating the gradient of the MRI image. The gradient is a measure of the grayscale change in the MRI image and can be used to detect the edge in the MRI image. i The individual element is , the following formula is used to calculate the first i The gradient of a voxel: ; in, The magnetic resonance image i The gradient of a voxel, The magnetic resonance image i The 3D coordinates of the voxel.

[0027] After detecting the edge, BET estimates the boundary between brain tissue and non-brain tissue using a spherical fitting method. The spherical fitting method constructs an approximate skull contour based on a spherical model. Given a preliminary estimate of the spherical boundary , the sphere is matched with the gradient information in the magnetic resonance image, and the sphere is fitted using the following energy function: ; in, is the energy function value, is the threshold value to determine the boundary between brain tissue and non-brain tissue. After minimizing the above energy function, a best-fit spherical boundary is obtained. After obtaining the spherical boundary, BET further generates a mask of brain tissue, removes non-brain tissue, and leaves the brain tissue part.

[0028] Step 22, registering the T1-weighted image and the brain tissue structure magnetic resonance image to obtain a registered magnetic resonance image.

[0029] Specifically, the linear rigid body transformation with 6 degrees of freedom and the spline interpolation method are used to register the T1-weighted image and the brain tissue structure magnetic resonance image of the same voxel and resample them to the same image space. By converting each sequence image at different voxel coordinates to the same anatomical coordinates, it is convenient to integrate cross-sequence signal information.

[0030] First, set the initial rigid body transformation parameters and use mutual information as the similarity metric. Then use the rigid body transformation formula , maps the pixel coordinates of the floating image to the fixed image coordinate system. Among them, is the pixel coordinate in the fixed image coordinate system, R is the rotation matrix, including x axis, y axis, z 3 rotation parameters of the axis , , , t is the translation matrix, including x axis, y axis, z 3 translation parameters of the axis , , Then, the pixel values ​​of the transformed floating image at non-integer coordinates are calculated using spline interpolation, and the rigid body transformation parameters are adjusted by the gradient descent method to maximize the similarity measure between the fixed image and the floating image, thus obtaining the registered floating image.

[0031] Step 23, performing noise reduction and B0 field correction processing on the registered magnetic resonance image in sequence to obtain a noise-reduced magnetic resonance image.

[0032] Specifically, the adaptive non-local means algorithm is used to reduce the noise of the registered magnetic resonance images. The specific process is: extract image blocks, calculate the similarity between image blocks, perform weighted averaging on the neighborhood of each pixel, remove Gaussian noise that may exist in the magnetic resonance image, and generate a clean image. This avoids the amplification of high-frequency noise in artifact correction. Subsequently, the signal intensity remapping method after Gaussian deconvolution is used to restore the original signal in the frequency domain through deconvolution, remove the magnetic resonance image artifacts caused by B0 field inhomogeneity, and improve the signal-to-noise ratio of the image.

[0033] Step 24, performing contrast-limited adaptive histogram equalization processing on the de-noised magnetic resonance image to obtain an equalized magnetic resonance image.

[0034] Due to the principle limitations of vascular imaging magnetic resonance sequences such as TOF-MRA, the signal enhancement amplitude of small blood vessels will be significantly lower than that of large blood vessels, making it difficult to identify small blood vessels. This application adopts a method of limited contrast adaptive histogram equalization to adaptively enhance local cerebrovascular signals so that cerebrovascular signals of different scales can obtain similar enhancement amplitudes.

[0035] Step 25: Perform blood vessel edge signal enhancement processing on the equalized magnetic resonance image to obtain an enhanced magnetic resonance image.

[0036] Due to the partial volume effect caused by the principle of magnetic resonance imaging, the edge signal of the blood vessel is significantly lower than the center signal of the blood vessel, so it is not easy to identify. This application adopts the Sobel convolution method to identify the edge position where the signal intensity mutation occurs in the equalized magnetic resonance image, realize the specific enhancement of the edge signal of the blood vessel, improve the contrast between the blood vessel and the surrounding non-vascular tissue, so that the cerebral blood vessel signals at different distances from the center line can obtain similar enhancement.

[0037] Step 26, performing gamma correction processing on the enhanced magnetic resonance image to obtain a preprocessed magnetic resonance image.

[0038] Specifically, by adjusting the transformation coefficient Gamma, the vascular tissue signal (high signal) of the enhanced magnetic resonance image is nonlinearly enhanced and the non-vascular tissue signal (low signal) is suppressed. The transformation coefficient Gamma, as a processing hyperparameter, can control the relative amplitude of the enhanced and suppressed signals according to the nature of the processed image to obtain better image contrast.

[0039] In step 202, the present application specifically enhances the vascular signal according to the essential characteristics of the cerebral blood vessels, thereby reducing the artifact noise of TOF-MRI and the risk of imbalance in the segmentation of large and small blood vessels.

[0040] Step 203: Based on the deep neural network model, the cerebral vascular tissue in the preprocessed magnetic resonance image is determined, and the cerebral vascular tissue is segmented to obtain a cerebral vascular segmentation image.

[0041] The deep neural network model is pre-trained using a training sample set, wherein the training sample set includes a plurality of sample pre-processed magnetic resonance images and a true label of each voxel in each sample pre-processed magnetic resonance image. The true label is cerebrovascular tissue or non-cerebrovascular tissue.

[0042] The source of the training sample set is: T1-weighted images and TOF-MRA paired data of 100 healthy people collected by different equipment using routine clinical scanning sequences. After the acquired images were preprocessed according to the methods of steps 21 to 26 above, two neuroradiologists with rich image annotation experience independently annotated the cerebrovascular tissue in the magnetic resonance images. A third senior neuroradiologist conducted a comprehensive inspection and evaluation of the annotation results to ensure the consistency and accuracy of the labels, and finally constructed the training sample set. The training of the deep neural network model is based on the deep annotated training data set on a large-scale TOF-MRA magnetic resonance cohort of healthy people, which ensures its sensitivity and specificity in the task of cerebrovascular tissue identification and segmentation, reduces the risk of overfitting, and has good generalization and robustness in the deployment and application of the deep neural network model.

[0043] In a specific application example, a five-fold cross-validation method is used to train and test the deep neural network model. In order to cope with the differences in specificity and sensitivity of vascular segmentation results at different scales during vascular segmentation, the Tversky function is introduced as the loss function of the deep neural network model. On the basis of the Tversky function, a weighted Tversky loss function is further proposed, that is, the inverse of the Euclidean distance of the voxel from the center line is used as the mismatch loss weight, which strengthens the topological structure restrictions of the segmentation results, so that the vascular segmentation results have more consistent performance at all scales. This application selects the deep neural network model with the best results as the final deep neural network model by default, and the user can also choose to deploy an average model. The weighted Tversky loss function is: ; in, is the loss function value, For the The predicted value of the voxel, ranging from [0,1], For the The true label of the voxel, which takes a value of 0 or 1. For the The weighting coefficient of the voxel is calculated based on the Euclidean distance of the voxel from the center line, specifically The inverse of the Euclidean distance from the voxel to the target centerline, To control the weight of false positives, To control the weight of false negatives, it is similar to the standard Tversky loss function.

[0044] The deep neural network model introduces an attention gating mechanism in the skip connection of 3D-UNet.

[0045] In this application, 3D-UNet adopts an adaptive design to automatically adjust hyperparameters such as batch size and network depth and width according to the spatial resolution and size of the input image.

[0046] In the deep neural network model, Figure 3 As shown, the attention gating mechanism uses the following method to c The feature tensor on the encoder corresponding to the skip connection is fused with the feature tensor on the decoder.

[0047] (1) Add the feature tensor on the encoder and the feature tensor on the decoder to obtain the summed feature.

[0048] (2) The summed features are sequentially subjected to ReLU activation, Sigmoid activation, resampling, and ReLU activation to obtain the converted features. The processing formula can be expressed as: ; in, is the feature tensor on the encoder, is the feature tensor on the decoder, for The linear transformation coefficients of for The linear transformation coefficients of is the first bias coefficient, is the ReLU activation function, is the linear transformation coefficient, is the second bias coefficient, is the Sigmoid activation function, It is the feature after Sigmoid activation.

[0049] (3) After multiplying the feature tensor on the encoder with the converted feature, it is connected to the feature tensor on the decoder to obtain the first c The features after skip connection. c The features after the skip connection are used as the input features of the next decoder.

[0050] By multiplying the feature tensor on the encoder with the converted feature, feature information at different scales is fused, thereby suppressing background noise to a great extent.

[0051] The deep neural network model of this application integrates the adaptive 3D-UNet architecture of the attention gating mechanism and uses the weighted Tversky loss function for training, which ensures the stability of the topology of the segmentation structure and the sensitivity of small blood vessel identification.

[0052] In order to ensure the continuity and accuracy of the segmentation result, the magnetic resonance image cerebral blood vessel segmentation method further includes the following step 204 .

[0053] Step 204 , integrating the spatial continuity information of the cerebral blood vessel segmentation image to obtain an optimized cerebral blood vessel segmentation image.

[0054] In a specific application example, step 204 includes the following steps 41 and 42 .

[0055] Step 41 : segment the cerebral blood vessel segmentation image using a three-dimensional conditional random field model to obtain a spatially continuous segmentation image.

[0056] The three-dimensional conditional random field model takes into account the continuity of vascular tissue in space. By establishing potential functions between pairs and on individual voxels, adjacent voxels with similar signal intensities and different labels on the cerebral vascular segmentation image are penalized to obtain a more continuous and reasonable segmentation image in three-dimensional space. The energy function formula of the three-dimensional conditional random field is: ; Among them, the potential function formula of a single pixel is: ; The potential function formula between two voxels is: ; in, is the energy function value, and For the m Individual pixels and n The 3D coordinates of the voxels, and For the m Individual pixels and n Label information of whether the voxel is a blood vessel, For the m The value of the single voxel potential function of each voxel, is the conditional probability, For the m Individual pixels and n The potential function value between voxels, For definition in and The label compatibility function on 1 if yes, 0 otherwise. For the m Individual pixels and n Gaussian distance between voxels.

[0057] By minimizing the above energy function, a spatially continuous and topologically reasonable segmentation result can be obtained. The labels of adjacent voxels are encouraged to be consistent, thereby maintaining the spatial continuity of the segmentation result and reducing the break phenomenon.

[0058] Step 42, removing connected components whose voxels are smaller than a set threshold in the spatially continuous segmented image, to obtain an optimized cerebral vascular segmentation image.

[0059] The present application further searches for the three-dimensional maximum connected component on the spatially continuous segmented image, and removes the connected components whose voxel size is smaller than the threshold by manually setting a threshold, thereby reducing small block noise and obtaining the final cerebral vascular segmentation image.

[0060] The present application considers the spatial continuity characteristics of cerebrovascular signals through step 204, and uses a probabilistic graph model to constrain unreasonable segmentation, thereby ensuring the continuity and accuracy of the segmentation results.

[0061] This application can realize accurate and stable identification and segmentation of secondary blood vessels on routine clinical TOF-MRA magnetic resonance vascular imaging data, and can ensure the continuity of segmentation labels. It has an extremely high processing speed and can be deployed offline in the medical system to achieve a full range of automatic digital evaluation of clinical patients' cerebral vascular structural characteristics, assist in the precise individualized diagnosis and treatment of cerebrovascular diseases, and has a strong clinical application value.

[0062] Based on the same inventive concept, the embodiment of the present application also provides a magnetic resonance image cerebral blood vessel segmentation system for implementing the magnetic resonance image cerebral blood vessel segmentation method involved above. The implementation scheme for solving the problem provided by the system is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more magnetic resonance image cerebral blood vessel segmentation system embodiments provided below can refer to the limitations of the magnetic resonance image cerebral blood vessel segmentation method above, and will not be repeated here.

[0063] In an exemplary embodiment, Figure 4 As shown, a magnetic resonance image cerebral blood vessel segmentation system is provided, including: an image acquisition module 401, a preprocessing module 402 and an image segmentation module 403.

[0064] The image acquisition module 401 is used to acquire the magnetic resonance image to be processed.

[0065] The preprocessing module 402 is used to perform noise reduction and blood vessel-specific enhancement processing on the magnetic resonance image to be processed to obtain a preprocessed magnetic resonance image.

[0066] The image segmentation module 403 is used to determine the cerebral vascular tissue in the preprocessed magnetic resonance image based on a deep neural network model, and segment the cerebral vascular tissue to obtain a cerebral vascular segmentation image.

[0067] The deep neural network model is obtained by pre-training with a training sample set, wherein the training sample set includes a plurality of sample pre-processed magnetic resonance images and the true label of each voxel in each sample pre-processed magnetic resonance image. The true label is cerebrovascular tissue or non-cerebrovascular tissue. The deep neural network model introduces an attention gating mechanism in the skip connection of 3D-UNet.

[0068] As an optional implementation, the magnetic resonance image cerebral blood vessel segmentation system further includes a post-processing module 404. The post-processing module 404 integrates spatial continuity information of the cerebral blood vessel segmentation image to obtain an optimized cerebral blood vessel segmentation image.

[0069] 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 5 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, 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 magnetic resonance images to be processed. 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 cerebral vascular segmentation of magnetic resonance images is implemented.

[0070] Those skilled in the art will understand that Figure 5 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.

[0071] 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.

[0072] 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.

[0073] 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.

[0074] In this application, all actions to obtain signals, information or data are carried out in compliance with the relevant data protection laws and policies of the country where they are located and with the authorization given by the owner of the corresponding device.

[0075] 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).

[0076] 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.

[0077] 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.

[0078] 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 segmenting cerebral blood vessels in magnetic resonance images, characterized in that: The magnetic resonance image cerebral blood vessel segmentation method comprises: acquiring a magnetic resonance image to be processed; Performing noise reduction and blood vessel-specific enhancement processing on the magnetic resonance image to be processed to obtain a preprocessed magnetic resonance image; Based on the deep neural network model, determining the cerebral vascular tissue in the preprocessed magnetic resonance image, and segmenting the cerebral vascular tissue to obtain a cerebral vascular segmentation image; Among them, the deep neural network model is obtained by pre-training using a training sample set, and the training sample set includes multiple sample pre-processed magnetic resonance images and the true label of each voxel in each sample pre-processed magnetic resonance image; the true label is cerebrovascular tissue or non-cerebrovascular tissue; the deep neural network model introduces an attention gating mechanism in the jump connection of 3D-UNet.

2. The method for segmenting cerebral blood vessels in magnetic resonance images according to claim 1, characterized in that: The magnetic resonance images to be processed include T1-weighted images and a time-of-flight magnetic resonance image queue.

3. The method for segmenting cerebral blood vessels in magnetic resonance images according to claim 2, characterized in that: The denoising and vascular-specific enhancement processing is performed on the magnetic resonance image to be processed to obtain a preprocessed magnetic resonance image, specifically comprising: Performing brain tissue-specific identification and segmentation on the time-of-flight magnetic resonance image queue to obtain a brain tissue structure magnetic resonance image; Registering the T1-weighted image and the brain tissue structure magnetic resonance image to obtain a registered magnetic resonance image; performing noise reduction and B0 field correction processing on the registered magnetic resonance image in sequence to obtain a noise-reduced magnetic resonance image; Performing contrast-limited adaptive histogram equalization processing on the de-noised magnetic resonance image to obtain an equalized magnetic resonance image; Performing blood vessel edge signal enhancement processing on the equalized magnetic resonance image to obtain an enhanced magnetic resonance image; Gamma correction processing is performed on the enhanced magnetic resonance image to obtain a preprocessed magnetic resonance image.

4. The method for segmenting cerebral blood vessels in magnetic resonance images according to claim 1, characterized in that: In the deep neural network model, the attention gating mechanism uses the following method to c The feature tensor on the encoder corresponding to the skip connection is fused with the feature tensor on the decoder: Add the feature tensor on the encoder and the feature tensor on the decoder to obtain the sum feature; The summed features are sequentially subjected to ReLU activation, Sigmoid activation, resampling, and ReLU activation to obtain converted features; After multiplying the feature tensor on the encoder with the converted feature, it is connected to the feature tensor on the decoder to obtain the first c Features after skip connection.

5. The method for segmenting cerebral blood vessels in magnetic resonance images according to claim 1, characterized in that: The loss function during the training of the deep neural network model is the weighted Tversky loss function: ； in, is the loss function value, For the The predicted value of a voxel, For the The true label of the voxel, For the The weighting coefficient of the voxel, To control the weight of false positives, To control the weight of false negative examples.

6. The method for segmenting cerebral blood vessels in magnetic resonance images according to claim 1, characterized in that: The magnetic resonance image cerebral blood vessel segmentation method further comprises: The spatial continuity information of the cerebral blood vessel segmentation image is integrated to obtain an optimized cerebral blood vessel segmentation image.

7. The method for segmenting cerebral blood vessels in magnetic resonance images according to claim 6, characterized in that: Integrating the spatial continuity information of the cerebral blood vessel segmentation image to obtain an optimized cerebral blood vessel segmentation image specifically includes: Using a three-dimensional conditional random field model to segment the cerebral blood vessel segmentation image to obtain a spatially continuous segmentation image; The connected components whose voxels are smaller than a set threshold value in the spatially continuous segmented image are removed to obtain an optimized cerebral blood vessel segmentation image.

8. A magnetic resonance image cerebral blood vessel segmentation system, applied to the magnetic resonance image cerebral blood vessel segmentation method according to any one of claims 1 to 7, characterized in that: The magnetic resonance image cerebral blood vessel segmentation system comprises: An image acquisition module, used for acquiring a magnetic resonance image to be processed; A preprocessing module, used for performing noise reduction and blood vessel-specific enhancement processing on the magnetic resonance image to be processed to obtain a preprocessed magnetic resonance image; An image segmentation module, used for determining the cerebral vascular tissue in the preprocessed magnetic resonance image based on a deep neural network model, and segmenting the cerebral vascular tissue to obtain a cerebral vascular segmentation image; Among them, the deep neural network model is obtained by pre-training using a training sample set, and the training sample set includes multiple sample pre-processed magnetic resonance images and the true label of each voxel in each sample pre-processed magnetic resonance image; the true label is cerebrovascular tissue or non-cerebrovascular tissue; the deep neural network model introduces an attention gating mechanism in the jump connection of 3D-UNet.

9. 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 magnetic resonance image cerebral blood vessel segmentation method according to any one of claims 1 to 7.

10. 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 segmenting cerebral blood vessels in magnetic resonance images according to any one of claims 1 to 7 is implemented.

Images