WO2019232688A1 - Method, apparatus, and computing device for segmenting veins in magnetic susceptibility weighted image - Google Patents

Abstract

The present invention pertains to the field of image processing, and provides a method, an apparatus, and a computing device for segmenting veins in a magnetic susceptibility weighted image, so as to accurately segment veins in a magnetic susceptibility weighted image. The method comprises: extracting and normalizing a brain region in an original brain SWI to obtain a normalized brain SWI; extracting a set of first image blocks and second image blocks having coincidental centers by using any one of sampling points defined in the normalized brain SWI at an interval of m voxels; inputting n sets of first image blocks and second image blocks into a trained convolutional neural network, such that the trained convolutional neural network marks veins in the n sets of first image blocks and second image blocks to obtain n sets of third image block pairs having marked veins; and mapping the veins in the n sets of third image block pairs back to the original brain SWI to obtain a vein segmentation result. The present invention enables accurate identification of veins in a brain SWI and an improvement in the precision of vein segmentation in the brain SWI.

Description

Method, device and computing device for segmenting venous blood vessels in magnetically sensitive weighted image Technical field

The invention belongs to the field of image processing, and particularly relates to a method, a device, and a computing device for segmenting a venous blood vessel in a magnetically sensitive weighted image.

Background technique

Acute ischemic stroke has high morbidity, high mortality, and high recurrence rate. In recent years, studies have shown that magnetically sensitive weighted images (SWI) are more sensitive to acute cerebral ischemia than magnetic resonance diffusion weighted images (DWI), and they can show lower veins on the affected side of cerebral ischemia Signals are getting more and more attention. The asymmetric features of the venous and normal sides of the brain in patients with cerebral ischemia can be used for diagnosis, treatment planning and prognosis prediction of acute cerebral ischemia.

Most of the current work is based on qualitative analysis, but lack of quantitative analysis methods. The key to quantitative SWI analysis is accurate segmentation of low-vein signals. Experienced experts are very focused and carefully carry out layer-by-layer manual labeling of venous low signals on SWI to obtain better segmentation results. However, manual labeling depends on expert experience and effort, which is a very time-consuming and labor-intensive task and is not repeatable. Therefore, the automatic segmentation of SWI veins with low signal becomes an urgent need. This is also a technical problem to be solved by the present invention.

The low signal segmentation of SWI veins has the following challenges: 1) Compared with cerebral arteries, SWI veins are smaller. There is only 1 to 2 voxels in the radius of the narrow area; 2) there is a very large difference in the apparent appearance of the SWI vein signal, which makes it difficult for experts to mark; And gray level changes are various.

In response to these challenges, the prior art generally uses shallow features to classify voxels in a brain computed tomography image (CTA) or magnetic resonance imaging (MRA) image to achieve cerebral arterial segmentation. For example, the region-based active contour method can utilize both grayscale and shape information, and iteratively optimizes through level set to achieve blood vessel segmentation. However, because this method uses shallow features such as grayscale and shape, it has limited recognition ability and low segmentation accuracy.

Summary of the Invention

An object of the present invention is to provide a method, a device, and a computing device for segmenting a venous blood vessel in a magnetically sensitive weighted image to accurately segment a venous blood vessel in a magnetically sensitive weighted image.

A first aspect of the present invention provides a method for segmenting a venous blood vessel in a magnetically sensitive weighted image. The method includes:

Extract and normalize the brain regions in the original brain magnetically sensitive weighted image SWI to obtain a standardized brain SWI;

A set of first image blocks and second image blocks with coincident centers are extracted centering on any one sampling point defined by m voxels per interval m in the standardized brain SWI. The area of the first image block includes M ₁ * A voxel arranged in a M ₁ order matrix, the region of the second image block includes a voxel arranged in a M ₂ * M ₂ order matrix, the n, m, M ₁ and M ₂ are natural numbers, and M ₂ > M ₁ >m;

Input the n groups of first image blocks and second image blocks to a trained convolutional neural network, and the trained convolutional neural network performs venous blood vessels in the n groups of first image blocks and second image blocks After labeling, two sets of three third image blocks labeled with venous blood vessels are obtained. The trained convolutional neural network is trained on the convolutional neural network through a supervised learning method, and the area of the third image block includes m * m-order matrix arranged voxels;

Map the venous blood vessel labels of the two third image blocks of the n groups to the original brain magnetically sensitive weighted image SWI to obtain the venous blood vessel segmentation result.

According to a second aspect of the present invention, a device for segmenting a venous blood vessel in a magnetically sensitive weighted image is provided. The device includes:

A normalization module for extracting and normalizing brain regions in the original brain magnetically sensitive weighted image SWI to obtain a standardized brain SWI;

An image block extraction module, for extracting a set of first image blocks and second image blocks with coincident centers centered on any one sampling point defined by every voxel in the standardized brain SWI The region of contains the voxels arranged in a M ₁ * M ₁ order matrix, and the region of the second image block contains the voxels arranged in a M ₂ * M ₂ order matrix, where m, M ₁ and M ₂ are Natural number, and M ₂ > M ₁ >m;

A labeling module, configured to input n groups of first image blocks and second image blocks to a trained convolutional neural network, and the trained convolutional neural network The venous vessels are labeled to obtain two sets of n third image blocks labeled with venous vessels. The trained convolutional neural network is trained by a convolutional neural network in a supervised learning manner, and the area of the third image block includes Voxels arranged in an m * m order matrix, where n is the number of the sampling points;

A mapping module, configured to map the venous blood vessel marks of the n sets of two third image blocks back to the original brain magnetically sensitive weighted image SWI to obtain a venous blood vessel segmentation result.

A third aspect of the present invention provides a computing device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following method is implemented: step:

Extract and normalize the brain regions in the original brain magnetically sensitive weighted image SWI to obtain a standardized brain SWI;

A set of first image blocks and second image blocks with coincident centers are extracted centering on any one sampling point defined by m voxels per interval m in the standardized brain SWI. The area of the first image block includes M ₁ * A voxel arranged in a matrix of order M ₁ , the region of the second image block includes a voxel arranged in a matrix of order M ₂ * M ₂ , the m, M ₁ and M ₂ are natural numbers, and M ₂ > M ₁ >m;

Input n sets of first image blocks and second image blocks to a trained convolutional neural network, and the trained convolutional neural network marks the venous blood vessels in the n sets of first image blocks and second image blocks The n groups of two third image blocks labeled with venous blood vessels are obtained, and the trained convolutional neural network is trained on the convolutional neural network through a supervised learning method. The region of the third image block includes m * m orders. Voxels arranged in a matrix, where n is the number of sampling points;

Map the venous blood vessel labels of the two third image blocks of the n groups to the original brain magnetically sensitive weighted image SWI to obtain the venous blood vessel segmentation result.

A fourth aspect of the present invention provides a computer-readable storage medium, where the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the following method are implemented:

Extract and normalize the brain regions in the original brain magnetically sensitive weighted image SWI to obtain a standardized brain SWI;

A set of first image blocks and second image blocks with coincident centers are extracted centering on any one sampling point defined by m voxels per interval m in the standardized brain SWI. The area of the first image block includes M ₁ * A voxel arranged in a matrix of order M ₁ , the region of the second image block includes a voxel arranged in a matrix of order M ₂ * M ₂ , the m, M ₁ and M ₂ are natural numbers, and M ₂ > M ₁ >m;

Input n sets of first image blocks and second image blocks to a trained convolutional neural network, and the trained convolutional neural network marks the venous blood vessels in the n sets of first image blocks and second image blocks The n groups of two third image blocks labeled with venous blood vessels are obtained, and the trained convolutional neural network is trained on the convolutional neural network through a supervised learning method. The region of the third image block includes m * m orders. Voxels arranged in a matrix, where n is the number of sampling points;

Map the venous blood vessel labels of the two third image blocks of the n groups to the original brain magnetically sensitive weighted image SWI to obtain the venous blood vessel segmentation result.

It can be known from the technical solution of the present invention that, since the trained convolutional neural network is trained on the convolutional neural network through a supervised learning method, the supervised learning automatically learns prior knowledge from existing data and extracts deep features, thereby being able to Accurately identifying the venous blood vessels in the brain SWI, compared with the prior art, the accuracy of venous blood vessel segmentation in the brain SWI is improved.

BRIEF DESCRIPTION OF THE DRAWINGS

1 is a schematic flowchart of an implementation method of a method for segmenting a venous blood vessel in a magnetically sensitive weighted image according to an embodiment of the present invention;

2 is a schematic diagram of a voxel and its sampling points in a standardized brain SWI according to an embodiment of the present invention;

3 is a schematic structural diagram of a convolutional neural network according to an embodiment of the present invention;

FIG. 4 is a schematic structural diagram of a device for segmenting a venous blood vessel in a magnetically sensitive weighted image according to an embodiment of the present invention; FIG.

5 is a schematic structural diagram of a device for segmenting a vein and a blood vessel in a magnetically sensitive weighted image according to another embodiment of the present invention;

FIG. 6 is a schematic structural diagram of a computing device according to an embodiment of the present invention.

Detailed ways

In order to make the objectives, technical solutions, and beneficial effects of the present invention clearer, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

In the following description, for the purpose of explanation rather than limitation, specific details such as a specific system structure and technology are provided in order to thoroughly understand the embodiments of the present invention. However, it should be clear to a person skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present invention with unnecessary details.

FIG. 1 is a schematic flowchart of an implementation method of a method for segmenting a venous blood vessel in a magnetically sensitive weighted image according to an embodiment of the present invention, which mainly includes the following steps S101 to S104, which are described in detail below:

S101: Extract and normalize a brain region in the original brain magnetically sensitive weighted image SWI to obtain a standardized brain SWI.

In the embodiment of the present invention, the brain region in the original brain magnetically sensitive weighted image SWI is extracted in order to exclude the interference of the skull and other non-brain tissues; the extracted brain region in the SWI, that is, the brain mask image belongs to the pre- Process. Specifically, extracting and normalizing the brain region in the original brain magnetically sensitive weighted image SWI, and obtaining a standardized brain SWI includes the following implementation process: using a threshold method to extract all background voxels whose gray value in the original brain SWI is less than 50, And extract the largest connected region; after inverting the largest connected region, use a 3 * 3 * 3 size structural element to perform morphological closing operation to restore the threshold segmentation of the lost vein voxels; calculate the brain region in the original brain SWI Voxel mean and standard deviation, the mean value of each voxel of the brain region in the original brain SWI is subtracted from the mean and divided by the standard deviation to obtain a standardized brain SWI.

S102. A set of first and second image blocks with coincident centers is extracted centered on any one sampling point defined by every voxel in the normalized brain SWI. The area of the first image block includes M ₁ * The voxels are arranged in M ₁ order matrix. The area of the second image block includes voxels arranged in M ₂ * M ₂ order matrix, m, M ₁ and M ₂ are natural numbers, and M ₂ > M ₁ > m.

The experiments prove, m = 9, M ₁ = 25 and M ₂ = 57 and the phase value, for purposes of the present invention, the effect is optimal, following, therefore, m = 9, M ₁ = 25 and M ₂ = 57 to An example is used to explain the technical solution of the present invention. An example of the standardized brain SWI obtained in step S101 is shown in FIG. 2. The small white circles and small black circles represent the voxels of the standardized brain SWI. Among them, the small white circles labeled 1, 2, 3, and 4 Represents a sampling point. Obviously, these sampling points are also voxels, and are spaced apart by 9 voxels. The technical solution of the present invention is to use these sampling points as the center to extract a set of first image blocks and second image blocks with coincident centers. Taking M ₁ = 25, M ₂ = 57, and the sampling point labeled 1 as an example, extracting a set of first and second image blocks with coincident centers is an extraction region containing voxels arranged in a 25 * 25 order matrix. The image blocks and regions contain image blocks of voxels arranged in a 57 * 57 order matrix. These two image blocks are a group, and the centers of the image blocks are centered on the sampling point labeled 1. Groundly, taking M ₁ = 25, M ₂ = 57, and the sampling point labeled 2 as an example, extracting a set of first and second image blocks with coincident centers is an extraction region containing a 25 * 25 order matrix. The image block and region of the voxel contain image blocks of voxels arranged in a 57 * 57 order matrix. These two image blocks are a group, and the centers of the image blocks are centered on the sampling point labeled 2. .

S103. Input n sets of first image blocks and second image blocks to the trained convolutional neural network, and after training the convolutional neural network to mark the venous blood vessels in the n sets of first image blocks and second image blocks The n groups of two third image blocks labeled with venous blood vessels are obtained. Among them, the trained convolutional neural network is trained on the convolutional neural network through supervised learning. The region of the third image block includes a matrix arranged in order of m * m order. The voxel of the cloth, n is the number of sampling points.

Obviously, according to the example of FIG. 2, how many sets of first image blocks and second image blocks can be obtained with how many sampling points, so when there are n sampling points, n can be obtained according to the method illustrated in FIG. 2. Group a first image block and a second image block. Each time a set of first image blocks and second image blocks are input to the trained convolutional neural network, and the trained convolutional neural network marks the venous blood vessels in the set of first image blocks and the second image blocks Each third image block can be labeled with a third image block of the venous blood vessel, each second image block can be labeled with a third image block of the venous blood vessel, and n sets of two third images are labeled with the venous blood vessel. Piece. Therefore, when n sets of first image blocks and second image blocks are input to the trained convolutional neural network, the trained convolutional neural network performs venous blood vessels in the n sets of first image blocks and second image blocks. After labeling, two sets of three third image blocks labeled with venous blood vessels can be obtained, where the region of the third image block contains voxels arranged in a matrix of order m * m. When m = 9, the third image The area of the block contains voxels arranged in a 9 * 9 order matrix.

In the embodiment of the present invention, the trained convolutional neural network is trained on the convolutional neural network through a supervised learning method, which can be extracted and standardized from the brain region in the original brain SWI to obtain a standardized brain SWI. That is, before the brain region in the original brain SWI is extracted and standardized to obtain a standardized brain SWI, steps S1031 to S1033 of the following method are further included:

S1031: Extract a training image block pair and a vein segmentation gold standard image block corresponding to the training image block pair from the brain SWI for training, where the training image block pair is composed of a first training image block and a second training image block.

In the embodiment of the present invention, the extracted training image block pair and the vein segmentation gold standard image block corresponding to the training image block pair are used to train a convolutional neural network, wherein the vein segmentation gold standard image block is based on artificial (generally (Experts) Experience The image blocks obtained by labeling the venous low signal in the training image are all randomly collected the same number of sample voxels from the venous region and the background region of the brain SWI used for training. Extracted as the center. It should be noted that the sizes of the first training image block and the second training image block are respectively a set of center coincident with a set of centers extracted from any one of the sampling points defined in the normalized brain SWI in each interval m voxels. One image block and the second image block are the same size, that is, the area of the first training image block contains voxels arranged in a M ₁ * M ₁ order matrix, and the area of the second training image block contains M ₂ * M ₂ Order voxels, and the size of the gold standard image block for vein segmentation is the same as the size of the third image block in the previous embodiment, that is, the area of the gold standard image block for vein segmentation includes the m * m order matrix. Voxels. When the area of the first image block in the foregoing embodiment contains voxels arranged in a 25 * 25 order matrix, and the area of the second image block contains voxels arranged in a 57 * 57 order matrix, accordingly The area of the first training image block contains voxels arranged in a 25 * 25 order matrix, and the area of the second training image block contains voxels arranged in a 57 * 57 order matrix; when the third image block in the foregoing embodiment The region of contains the voxels arranged in a 9 * 9 order matrix, and the area of the gold standard image block for vein segmentation contains the voxels arranged in a 9 * 9 order matrix.

Considering that data augmentation helps to improve the performance of the convolutional neural network, in order to increase the number of training images, a training image block pair is extracted from the brain SWI used for training and the vein segmentation gold corresponding to the training image block pair Before the standard image block, the SWI of the brain used for training is also symmetrically reversed along the X-axis and Y-axis of the two-dimensional coordinate system, and in order to enhance the robustness of the convolutional neural network to grayscale changes, After extracting a training image block pair from a brain SWI used for training and dividing a vein standard gold standard image block corresponding to the training image block pair, the method further includes a voxel grayscale for the first training image block and the second training image block. Transform according to the formula I ' _S = I _S + r * σ _c , where r is a random number sampled from a normal distribution N (0,1), and σ _c is the first training image block or the second training image block The gray standard deviation, I ′ _S is a gray value after transforming the gray I _S in the first training image block or the second training image block.

S1032. Construct a convolutional neural network.

In the embodiment of the present invention, constructing a convolutional neural network includes constructing a first convolution path, a second convolution path, a third convolution path, and a classifier, and the first convolution path, the second convolution path, A third convolution path is connected to the classifier, where the first convolution path is used to process the first training image block, the second convolution path is used to process the second training image block, and the first convolution path Or the second convolution path includes 8 convolution modules and 3 series layers, the second convolution path also includes a downsampling unit and an upsampling unit, and the third convolution path includes 3 convolution modules and 2 series Layer, the convolution modules 1 to 8 of the eight convolution modules are connected in series, and the outputs of convolution module 2 and convolution module 4 are connected to the input of layer 1 in series, and the output of layer 1 and convolution are connected in series. The output of module 6 is connected to the input of serial layer 2 respectively, the output of serial layer 2 and the output of convolution module 8 are connected to the input of serial layer 3 respectively; The input is connected to the output of the downsampling unit, and the second convolution The output of serial layer 3 in the path is connected to the input of the upsampling unit. The output of the upsampling unit and the output of serial layer 3 in the first convolution path are connected to the input of serial layer 1 in the third convolution path. Connection; convolution module 1 in the third convolution path is connected in series with convolution module 2, the input end of convolution module 1 in the third convolution path is connected to the output end of series layer 1 in the third convolution path, and the third volume The output of convolution module 2 in the convolution path and the output of series layer 1 in the third convolution path are connected to the input of series layer 2 in the third convolution path, and the output of series layer 2 in the third convolution path. And the input of the convolution module 3 in the third convolution path, and the output of the convolution module 3 in the third convolution path is connected to the classifier. The structure of the entire convolutional neural network is shown in FIG. 3.

It should be noted that the first convolution path is used to process the first training image block, and the second convolution path is used to process the second training image block. It can be seen that the convolution nerve of the present invention illustrated in FIG. 3 The network is a multi-scale convolutional neural network. After the multi-scale convolutional neural network is trained, a multi-scale trained convolutional neural network is obtained.

In the convolutional neural network illustrated in FIG. 3, the convolution modules of the first convolution path, the second convolution path, and the third convolution path all have the same structure, and each includes a batch standardized BN layer and a non-linear mapping PReLU layer. And conv layer Conv, where the normalized BN layer solves the covariance shift problem in the neural network by normalizing the feature map of the incoming convolution layer, and the non-linear mapping PReLU layer implements the non-linear mapping of convolution features and avoids Conventional excitation functions, such as the disappearance of the gradient caused by saturation of the Sigmoid function, accelerate the convergence. The conv layer Conv defines the convolution kernel (the size of the convolution kernel in the first and second convolution paths is 3x3, and the third convolution The size of the convolution kernel in the path is 1x1), and the convolution operation is implemented (the convolution step is 1):

among them,

Represents the convolution kernel of the mth neuron of the l-th convolutional layer, and the l-th convolutional layer receives as input the feature map output by the previous convolutional layer.

Represents the n-th feature map output by the l-1 convolution layer, n _l-1 represents the number of feature maps output by the l-1 convolution layer,

Represents the bias of the mth neuron of the l-th convolution layer, * is the convolution operation,

Represents the m-th feature map output by the l-th convolution layer.

In the convolutional neural network exemplified in FIG. 3, dense connections are achieved through series layers to protect the neural network signals and improve the trainability of the deep convolutional network. In the present invention, the feature maps of the two convolutional layers are implemented in series:

Among them, [·] represents a tandem operation, and x _l represents a feature map of the output of the l-th convolution layer,

The center part of the feature map x _l-2 representing the output of the l-2 convolution layer is the same as x _l .

An upsampling unit is added at the end of the second convolution path to achieve the output feature map size matching of the two convolution paths. The feature maps sampled by the upsampling unit and the feature maps output from the third series layer of the first convolution path are connected in series and processed by the third convolution path. The third convolution path also consists of a convolution module and a series layer. The difference is that the convolution kernels of the convolution layer in the convolution module of the third convolution path have a size of 1x1 and a step size of 1, so the third volume The product path does not change the size of the output feature map. Finally, the feature map processed by the output of the third convolution module in the third convolution path is processed by a classifier, such as a Softmax classifier, to obtain the input image block center m * m. The individual voxels belong to the vein or background. Probability and labeling.

S1033. The first training image block and the second training image block are input to a convolutional neural network, and the convolutional neural network is trained according to the vein segmentation gold standard image block to obtain a trained convolutional neural network.

Specifically, the first training image block and the second training image block are input to the convolutional neural network, and the convolutional neural network is trained according to the vein segmentation gold standard image block. The trained convolutional neural network can be achieved through the following steps S1 and S2 :

S1. Use a predicted image block and the vein segmentation gold standard image block to define a loss function

The above loss function is a loss function defined by the Dice coefficient, where the predicted image block is the predicted output result after the first training image block and the second training image block are input to the convolutional neural network, and B is the process of training the convolutional neural network. The number of training image block pairs processed at a time, N is the number of voxels in the predicted image block, p _ij is the probability that the jth voxel in the corresponding prediction image block of the i-th training image block belongs to the vein, and y _ij The i-th training image block is the true label of the j-th voxel in the corresponding vein segmentation gold standard image block, and the training image block pair is a combination of the first training image block and the second training image block mentioned in the foregoing embodiment. Pair of training image blocks.

S2. Using the first training image block and the second training image block as input features, train a convolutional neural network based on a batch gradient descent algorithm, and obtain parameters of the convolutional neural network while minimizing a loss function.

Minimizing the loss function

The parameters of the convolutional neural network obtained at that time are the parameters of the trained convolutional neural network, which means that the convolutional neural network has been trained and can be used for data prediction, that is, to segment the vein blood vessels in the magnetically sensitive weighted image. Specifically, it will be extracted. The first image block of each of the n groups of the first image block and the second image block is input to the first convolution path of the trained neural network, and the second image block of each group is input to the second volume of the trained neural network. The product path, after a series of processing by the trained neural network, each group of the first image block and the second image block respectively obtains a group of two third image blocks labeled with a venous blood vessel.

As mentioned before, since the trained convolutional neural network of the present invention is a multi-scale trained convolutional neural network, n sets of first image blocks and second image blocks are input to the trained convolutional neural network. When multi-scale trained convolutional neural network marks the venous blood vessels in the n sets of first image blocks and second image blocks, different sizes of first image blocks and second image blocks are used to obtain different ranges of context information Among them, small image blocks mainly capture local information, while large image blocks focus more on global information. The two kinds of information complement each other, thereby improving the performance of the convolutional neural network and the accuracy of vein segmentation.

S104. Map the venous blood vessel labels of the two third image blocks in the n groups to the original brain magnetically sensitive weighted image SWI to obtain the venous blood vessel segmentation result.

Taking the third image block containing voxels arranged in a 9 * 9 order matrix as an example, after step S103, n groups of two third image blocks labeled with venous vessels are obtained, and n groups of two third images are obtained. The venous blood vessel label of each third image block of the block is mapped back to the original brain magnetically sensitive weighted image SWI to obtain the venous blood vessel segmentation result. The specific mapping method is based on the space of the center of each third image block in the original image. Position, map a 9x9 size prediction marker around the center of the third image block to the original image. For example, the center coordinate of a third image block is (100, 100). After segmentation of venous blood vessels by a trained convolutional neural network, a 9x9 labeled area is obtained, and the label is mapped to [96: 104, 96: 104] to obtain the segmentation results of venous blood vessels.

The following Table 1 compares the effects of the method provided by the present invention with several methods of the prior art.

Table 1

method	Dice coefficient	Sensitivity	Specificity
Method provided by the present invention	0.736 ± 0.046	0.821 ± 0.026	0.223 ± 0.003
Single prediction method convolutional neural network	0.705 ± 0.069	0.875 ± 0.076	0.990 ± 0.004
Multi-scale vascular enhancement	0.615 ± 0.06	0.714 ± 0.045	0.989 ± 0.006
Multi-directional straight gray distribution	0.388 ± 0.037	0.44 ± 0.096	0.985 ± 0.003

It can be concluded from the above Table 1 that the convolutional neural network provided by the present invention can extract deep features with stronger recognition capabilities. Compared with shallow features such as grayscale and shape, depth features have a better ability to identify problems such as uneven grayscale of veins and veins and background grayscale overlap. In addition, the convolutional network provided by the present invention adds residual connections, and uses the loss function defined by the Dice coefficient to guide the training of the convolutional neural network, and the performance of the convolutional neural network is further improved. As shown in Table 1, the method provided by the present invention has the highest Dice coefficient in the segmentation result. The single prediction type convolutional network adopts the same structure as the convolutional neural network provided by the present invention, but its input image block size is 17x17. At this time, Only the label of the center voxel of the input image block can be predicted. Because the same number of venous blood samples and background samples participate in the training, the actual distribution of the samples is different (the background samples are much larger than the venous samples), so the segmentation results appear over segmentation, and more background samples are classified as venous blood vessels. Therefore, while the sensitivity is increased, the specificity is reduced, and the Dice coefficient is correspondingly decreased. The other two methods use shallow features such as grayscale and shape, and have limited recognition capabilities, so their segmentation accuracy is low.

It should be noted that ischemic stroke is a recognized worldwide problem in the medical community, and its diagnosis process is quite complicated. Although the technical solution of the present invention combines a convolutional neural network, it accurately segmentes the magnetically sensitive weighted images through image processing. Venous blood vessels, but this process is only used as a process to obtain an intermediate result, and the result is only used as an intermediate result. It cannot be directly used as a diagnosis of ischemic stroke. Obtain the health status of patients with ischemic stroke directly.

From the method for segmenting venous blood vessels in the magnetically sensitive weighted image shown in the example in FIG. 1 above, since the trained convolutional neural network is trained on the convolutional neural network through a supervised learning method, the supervised learning is automatically used from the existing data. Learning prior knowledge and extracting deep features can accurately identify venous blood vessels in the brain SWI. Compared with the prior art, the accuracy of venous blood vessel segmentation in the brain SWI is improved.

FIG. 4 is a schematic diagram of a device for segmenting a venous blood vessel in a magnetically sensitive weighted image according to an embodiment of the present invention, which mainly includes a normalization module 401, an image block extraction module 402, a labeling module 403, and a mapping module 404. The detailed description is as follows:

A normalization module 401, configured to extract and normalize brain regions in the original brain magnetically sensitive weighted image SWI to obtain a standardized brain SWI;

An image block extraction module 402 is used to extract a set of first image blocks and second image blocks with coincident centers around any one sampling point defined by m voxels per interval m in standardized brain SWI. The region contains voxels arranged in a matrix of order M ₁ * M ₁ , the region of the second image block contains voxels arranged in matrix of order M ₂ * M ₂ , m, M ₁ and M ₂ are natural numbers, and M ₂ > M ₁ >m;

A labeling module 403, configured to input n sets of first image blocks and second image blocks to a trained convolutional neural network, and the trained convolutional neural network performs venous blood vessels in the n sets of first image blocks and second image blocks After labeling, two sets of third image blocks labeled with venous blood vessels are obtained. Among them, the trained convolutional neural network is trained on the convolutional neural network through a supervised learning method. The area of the third image block includes m * m orders. Voxels arranged in a matrix, n is the number of sampling points;

The mapping module 404 is configured to map the venous blood vessel marks of the n sets of two third image blocks back to the original brain magnetically sensitive weighted image SWI to obtain the venous blood vessel segmentation result.

It should be noted that, because the device provided by the embodiment of the present invention is based on the same concept as the method embodiment of the present invention, the technical effects brought by it are the same as those of the method embodiment of the present invention. For specific content, refer to the description in the method embodiment of the present invention , Will not repeat them here.

The normalization module 401 illustrated in FIG. 4 may include an extraction unit 501, a negation unit 502, and a calculation unit 503, such as the device for segmenting a vein and blood vessel in a magnetically sensitive weighted image illustrated in FIG. 5, where:

An extraction unit 501, configured to extract all background voxels with a gray value less than 50 in the original brain SWI by using a threshold method, and extract a maximum connected region;

The inversion unit 502 is configured to perform a morphological closing operation using a 3 * 3 * 3 size structural element after inverting the largest connected region to restore the threshold segmentation of the lost vein voxels;

The calculation unit 503 is configured to calculate the mean and standard deviation of the voxels of the brain region in the original brain SWI, subtract the mean value of each voxel of the brain region in the original brain SWI, and divide by the standard deviation to obtain a standardized brain SWI.

The apparatus illustrated in FIG. 4 may further include a training image block extraction module, a construction module, and a training module, where:

A training image block extraction module is configured to extract a training image block pair from a brain SWI used for training and a vein standard gold standard image block corresponding to the training image block pair. The training image block pair is composed of a first training image block. And a second training image block, the region of the first training image block includes voxels arranged in a M ₁ * M ₁ order matrix, and the region of the second training image block includes a M ₂ * M ₂ order matrix Arranged voxels, the region of the vein-segmented gold standard image block including voxels arranged in an m * m order matrix;

A construction module for constructing a convolutional neural network;

A training module, configured to input the first training image block and the second training image block into the convolutional neural network, and train the convolutional neural network according to the vein segmentation gold standard image block to obtain the trained Convolutional neural network.

The apparatus of the above embodiment further includes an inversion module and a transformation module, where:

Inversion module, used for training image block extraction module to extract the training image block pair from the brain SWI used for training and before dividing the vein standard gold standard image block corresponding to the training image block pair to the training brain block SWI performs symmetrical inversions along the X and Y axes of the two-dimensional coordinate system, respectively;

A transformation module for training image block extraction module to extract a training image block pair from a brain SWI used for training and a vein segmentation gold standard image block corresponding to the training image block pair, and then to the first training image block and The voxel gray level of the second training image block is transformed according to the formula I ′ _S = I _S + r * σ _c , where r is a random number sampled from a normal distribution N (0,1), and σ _c is the first standard deviation of gray training image blocks or the second training image blocks, I _'S is the value of gradation after gradation conversion I _S training image block within the first or the second training image blocks.

The construction module of the above embodiment is specifically configured to construct a first convolution path, a second convolution path, a third convolution path, and a classifier, and perform the first convolution path, the second convolution path, and the third convolution. The path and the classifier are connected, the first convolution path is used to process the first training image block, the second convolution path is used to process the second training image block, and the first A convolution path or a second convolution path includes 8 convolution modules and 3 series layers, the second convolution path further includes a downsampling unit and an upsampling unit, and the third convolution path includes 3 Convolution modules and two series layers; after the first to eighth convolution modules of the eight convolution modules are connected in series, the output ends of the second and fourth convolution modules are respectively connected to the first The inputs of the two series layers are connected, the output of the first series layer and the output of the sixth convolution module are connected to the inputs of the second series layer, the output of the second series layer and the eighth The output end of the convolution module is respectively connected to the input end of the third series layer; the second volume The input of the first convolution module in the path is connected to the output of the downsampling unit, and the output of the third series layer in the second convolution path is connected to the input of the upsampling unit. The output end of the upsampling unit and the output end of the third series layer in the first convolution path are respectively connected to the input end of the first series layer in the third convolution path; the third convolution The first convolution module in the path is connected in series with the second convolution module, and the input of the first convolution module in the third convolution path is connected to the output of the first series layer in the third convolution path. And the output end of the second convolution module in the third convolution path and the output end of the first series layer in the third convolution path are respectively connected to the second series layer in the third convolution path. The input end of the third convolution path is connected to the output end of the second series layer in the third convolution path and the input end of the third convolution module in the third convolution path. An output end of each convolution module is connected to the classifier.

The training module of the above embodiment further includes a function definition unit and a network training unit, where:

A function defining unit, configured to define a loss function using a predicted image block and the vein segmentation gold standard image block

Wherein, the predicted image block is a predicted output result after the first training image block and the second training image block are input to a convolutional neural network, and B is the number of training image block pairs processed at one time during the training process of the convolutional neural network, N is the number of voxels in the predicted image block, p _ij is the probability that the j-th voxel in the corresponding predicted image block belongs to the vein vein, and y _ij is the i-th training image block paired with the corresponding vein Segment the true label of the jth voxel in the gold standard image block;

A network training unit is configured to use a first training image block and a second training image block as input features to train a convolutional neural network based on a batch gradient descent algorithm, and obtain parameters of the convolutional neural network while minimizing a loss function.

FIG. 6 is a schematic structural diagram of a computing device according to an embodiment of the present invention. As shown in FIG. 6, the computing device 6 of this embodiment includes a processor 60, a memory 61, and a computer program 62 stored in the memory 61 and executable on the processor 60, such as segmentation of a venous vessel in a magnetically sensitive weighted image Method of Procedure. When the processor 60 executes the computer program 62, the steps in the embodiment of the method for segmenting venous blood vessels in the magnetically sensitive weighted image described above are implemented, for example, steps S101 to S104 shown in FIG. Alternatively, when the processor 60 executes the computer program 62, the functions of the modules / units in the foregoing device embodiments are realized, such as the functions of the standardization module 401, image block extraction module 402, marking module 403, and mapping module 404 shown in FIG.

Exemplarily, the computer program 62 of the method for segmenting venous blood vessels in a magnetically sensitive weighted image mainly includes: extracting and normalizing a brain region in the original brain magnetically sensitive weighted image SWI to obtain a standardized brain SWI; An arbitrary sampling point defined by an interval m voxels is taken as the center, and a set of first and second image blocks with coincident centers is extracted. The area of the first image block includes voxels arranged in a M ₁ * M ₁ order matrix. The region of the second image block contains voxels arranged in a M ₂ * M ₂ order matrix, m, M ₁ and M ₂ are natural numbers, and M ₂ > M ₁ >m; input n groups of the first image block and the second The image block to the trained convolutional neural network, and the trained convolutional neural network marks the venous vessels in the n sets of first and second image blocks to obtain two sets of n labeled third venous vessels Image block, which has been trained by the trained convolutional neural network through supervised learning. The area of the third image block contains voxels arranged in a matrix of order m * m, and n is the number of sampling points. Venous vessel marking of two third image blocks in n groups Exit back to the original brain susceptibility-weighted images to obtain the segmentation result SWI's vein. The computer program 62 may be divided into one or more modules / units, and the one or more modules / units are stored in the memory 61 and executed by the processor 60 to complete the present invention. One or more modules / units may be a series of computer program instruction segments capable of performing a specific function, and the instruction segments are used to describe the execution process of the computer program 62 in the computing device 6. For example, the computer program 62 may be divided into functions (modules in a virtual device) of the normalization module 401, the image block extraction module 402, the marking module 403, and the mapping module 404. The specific functions of each module are as follows: The brain region in the brain magnetically sensitive weighted image SWI is extracted and standardized to obtain a standardized brain SWI. The image block extraction module 402 is used to center any one of the sampling points defined by m voxels in the standardized brain SWI as The first image block and the second image block coincide with the center of the group. The area of the first image block contains voxels arranged in a M ₁ * M ₁ order matrix, and the area of the second image block contains a M ₂ * M ₂ order matrix. The arranged voxels, m, M ₁ and M ₂ are natural numbers, and M ₂ > M ₁ >m; the labeling module 403 is used to input n sets of first image blocks and second image blocks to the trained convolutional neural network , After training the convolutional neural network to label the venous blood vessels in the n groups of the first image block and the second image block, two sets of n third image blocks labeled with the venous blood vessel are obtained. The trained convolutional neural network passes the supervision Learning style Product neural network training, the area of the third image block contains voxels arranged in a matrix of order m * m, n is the number of sampling points; the mapping module 404 is used to convert n groups of two third image blocks into Vein vein markers are mapped back to the original brain magnetically sensitive weighted image SWI to obtain the segmentation results of vein veins.

The computing device 6 may include, but is not limited to, a processor 60 and a memory 61. Those skilled in the art can understand that FIG. 6 is only an example of the computing device 6 and does not constitute a limitation on the computing device 6. It may include more or fewer components than shown in the figure, or combine some components or different components. For example, computing devices may also include input and output devices, network access devices, and buses.

The so-called processor 60 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), Ready-made programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

The memory 61 may be an internal storage unit of the computing device 6, such as a hard disk or a memory of the computing device 6. The memory 61 may also be an external storage device of the computing device 6, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, and a flash memory card (Flash) provided on the computing device 6. Card) and so on. Further, the memory 61 may also include both an internal storage unit of the computing device 6 and an external storage device. The memory 61 is used to store computer programs and other programs and data required by the computing device. The memory 61 may also be used to temporarily store data that has been output or is to be output.

Those skilled in the art can clearly understand that, for the convenience and brevity of the description, only the above-mentioned division of functional units and modules is used as an example. In practical applications, the above functions can be assigned by different functional units, Module completion, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment may be integrated into one processing unit, or each unit may exist separately physically, or two or more units may be integrated into one unit, and the integrated unit may use hardware. It can be implemented in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the protection scope of the present application. For specific working processes of the units and modules in the above system, reference may be made to corresponding processes in the foregoing method embodiments, and details are not described herein again.

In the foregoing embodiments, the description of each embodiment has its own emphasis. For a part that is not detailed or recorded in an embodiment, reference may be made to related descriptions of other embodiments.

Those of ordinary skill in the art may realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

In the embodiments provided by the present invention, it should be understood that the disclosed apparatus / computing device and method may be implemented in other ways. For example, the device / computing device embodiments described above are only schematic. For example, the division of modules or units is only a logical function division. In actual implementation, there may be another division manner, such as multiple units or components. Can be combined or integrated into another system, or some features can be ignored or not implemented. In addition, the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or units, which may be electrical, mechanical or other forms.

The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, may be located in one place, or may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the objective of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist separately physically, or two or more units may be integrated into one unit. The above integrated unit may be implemented in the form of hardware or in the form of software functional unit.

When integrated modules / units are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the method of the above embodiment, and can also be performed by a computer program instructing related hardware. The computer program of the method for segmenting a vein in a magnetically sensitive weighted image can be stored in a computer In a readable storage medium, when the computer program is executed by a processor, it can implement the steps of the above method embodiments, that is, extract and normalize the brain region in the original brain magnetically sensitive weighted image SWI to obtain a standardized brain SWI ; Extracting a set of first and second image blocks that coincide with each other at the center of any sampling point defined by every voxel in the normalized brain SWI. The area of the first image block includes M ₁ * M ₁ order Voxels arranged in a matrix, the region of the second image block includes voxels arranged in a matrix of order M ₂ * M ₂ , m, M ₁ and M ₂ are natural numbers, and M ₂ > M ₁ >m; input n groups The first image block and the second image block to the trained convolutional neural network, and the trained convolutional neural network marks the venous vessels in the n groups of the first image block and the second image block to obtain a label The n groups of two third image blocks of the venous blood vessels are trained by the trained convolutional neural network to train the convolutional neural network through a supervised learning method. The region of the third image block contains voxels arranged in a m * m order matrix. , N is the number of sampling points; map the venous blood vessel marks of the two third image blocks of the n groups back to the original brain magnetic sensitive weighted image SWI to obtain the venous blood vessel segmentation result. The computer program includes computer program code, and the computer program code may be in a source code form, an object code form, an executable file, or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disc, a computer memory, a read-only memory (ROM, Read-Only Memory), random access Memory (RAM, Random Access Memory), electric carrier signals, telecommunication signals, and software distribution media. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of the legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to the legislation and patent practice, the computer-readable medium does not include Electric carrier signals and telecommunication signals. The above embodiments are only used to illustrate the technical solutions of the present invention, but not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still apply the foregoing embodiments. The recorded technical solutions are modified, or some of the technical features are equivalently replaced; and these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the present invention Within the scope of protection.

Claims (10)

1. A method for segmenting venous blood vessels in a magnetically sensitive weighted image, wherein the method includes:

   Extract and normalize the brain regions in the original brain magnetically sensitive weighted image SWI to obtain a standardized brain SWI;

   A set of first image blocks and second image blocks with coincident centers are extracted centering on any one sampling point defined by m voxels per interval m in the standardized brain SWI. The area of the first image block includes M ₁ * A voxel arranged in a matrix of order M ₁ , the region of the second image block includes a voxel arranged in a matrix of order M ₂ * M ₂ , the m, M ₁ and M ₂ are natural numbers, and M ₂ > M ₁ >m;

   Input n sets of first image blocks and second image blocks to a trained convolutional neural network, and the trained convolutional neural network marks the venous blood vessels in the n sets of first image blocks and second image blocks The n groups of two third image blocks labeled with venous blood vessels are obtained, and the trained convolutional neural network is trained on the convolutional neural network through a supervised learning method. The region of the third image block includes m * m orders. Voxels arranged in a matrix, where n is the number of sampling points;

   Map the venous blood vessel labels of the two third image blocks of the n groups to the original brain magnetically sensitive weighted image SWI to obtain the venous blood vessel segmentation result.
2. The method for segmenting venous vessels in a magnetically sensitive weighted image according to claim 1, wherein the extracting and normalizing the brain region in the original brain magnetically sensitive weighted image SWI to obtain a standardized brain SWI includes:

   Using a threshold method to extract all background voxels with a gray value less than 50 in the original brain SWI, and extract the largest connected region;

   After inverting the maximum connected region, using a 3x3x3 size structural element to perform a morphological closing operation to restore the threshold segmentation of the lost vein voxels;

   Calculate the mean and standard deviation of the voxels of the brain region in the original brain SWI, subtract the mean of each voxel of the brain region in the original brain SWI, and divide by the standard deviation to obtain the standardized brain SWI.
3. The method for segmenting venous blood vessels in a magnetically sensitive weighted image according to claim 1, wherein the brain region in the original brain magnetically sensitive weighted image SWI is extracted and standardized, and before the standardized brain SWI is obtained, the The method also includes:

   Extracting a training image block pair from the brain SWI used for training and a vein standard gold standard image block corresponding to the training image block pair, the training image block pair consisting of a first training image block and a second training image block, A region of the first training image block includes voxels arranged in a M ₁ * M ₁ order matrix, and a region of the second training image block includes voxels arranged in a M ₂ * M ₂ order matrix, the The region of the gold standard image block for vein segmentation includes voxels arranged in a matrix of order m * m;

   Construct a convolutional neural network;

   The first training image block and the second training image block are input to the convolutional neural network, and the convolutional neural network is trained according to the vein segmentation gold standard image block to obtain the trained convolutional neural network.
4. The method for segmenting venous blood vessels in a magnetically sensitive weighted image according to claim 3, wherein the training image block pair extracted from the brain SWI for training and the vein segmentation gold corresponding to the training image block pair Before the standard image block, the method further includes: performing symmetrical inversion of the brain SWI for training along the X-axis and Y-axis of the two-dimensional coordinate system, respectively;

   After the extracting a training image block pair from a brain SWI for training and a vein segmentation gold standard image block corresponding to the training image block pair, the method further includes: dividing the first training image block and the second training image block The voxel gray level of the training image block is transformed according to the formula I ′ _S = I _S + r * σ _c , where r is a random number sampled from a normal distribution N (0,1), and σ _c is the a first standard deviation of gray training image block or the second training image blocks, the I _'S is the value of gradation after gradation conversion I _S training image block within the first or the second training image blocks.
5. The method for segmenting venous blood vessels in a magnetically sensitive weighted image according to claim 3, wherein said constructing a convolutional neural network comprises:

   Constructing a first convolution path, a second convolution path, a third convolution path, and a classifier, and connecting the first convolution path, the second convolution path, the third convolution path, and the classifier, where A first convolution path is used to process the first training image block, and a second convolution path is used to process the second training image block. The first convolution path or the second convolution path The path includes 8 convolution modules and 3 series layers. The second convolution path also includes a downsampling unit and an upsampling unit. The third convolution path includes 3 convolution modules and 2 series layers. ;

   After the convolution modules 1 to 8 of the eight convolution modules are connected in series, the outputs of the convolution module 2 and the convolution module 4 are connected to the input of the series layer 1, respectively, and the output of the series layer 1 and the convolution module are connected. The output of 6 is connected to the input of serial layer 2 respectively, and the output of serial layer 2 and the output of convolution module 8 are connected to the input of serial layer 3 respectively;

   The input end of the convolution module 1 in the second convolution path is connected to the output end of the downsampling unit, and the output end of the series layer 3 in the second convolution path is connected to the input end of the upsampling unit. An output end of the upsampling unit and an output end of series layer 3 in the first convolution path are respectively connected to an input end of series layer 1 in the third convolution path;

   The convolution module 1 in the third convolution path is connected in series with the convolution module 2. The input terminal of the convolution module 1 in the third convolution path is connected to the output terminal of the series layer 1 in the third convolution path. The output of the convolution module 2 in the third convolution path and the output of the series layer 1 in the third convolution path are connected to the input of the series layer 2 in the third convolution path, respectively. The output end of the serial layer 2 in the three convolution paths is connected to the input end of the convolution module 3 in the third convolution path, and the output end of the convolution module 3 in the third convolution path is connected to the classifier.
6. The method for segmenting venous blood vessels in a magnetically sensitive weighted image according to claim 3, wherein the first training image block and the second training image block are input to the convolutional neural network, and according to the convolutional neural network, Training the convolutional neural network with a gold standard image block for vein segmentation to obtain the trained convolutional neural network includes:

   Defining a loss function using a predicted image block and the vein segmentation gold standard image block

   

   The predicted image block is a predicted output result after the first training image block and the second training image block are input to the convolutional neural network, and the B is a training for one processing in the convolutional neural network training process. The number of image block pairs, where N is the number of voxels in the predicted image block, and p _ij is the probability that the j-th voxel in the corresponding predicted image block of the i-th training image block belongs to a vein. Let y _ij be the true label of the jth voxel in the i-th training image block to the corresponding vein segmentation gold standard image block;

   With the first training image block and the second training image block as input features, the convolutional neural network is trained based on a batch gradient descent algorithm, and parameters of the convolutional neural network are obtained when the loss function is minimized.
7. A device for segmenting a venous blood vessel in a magnetically sensitive weighted image, wherein the device includes:

   A normalization module for extracting and normalizing brain regions in the original brain magnetically sensitive weighted image SWI to obtain a standardized brain SWI;

   An image block extraction module, for extracting a set of first image blocks and second image blocks with coincident centers centered on any one sampling point defined by every voxel in the standardized brain SWI The region of contains the voxels arranged in a M ₁ * M ₁ order matrix, and the region of the second image block contains the voxels arranged in a M ₂ * M ₂ order matrix, where m, M ₁ and M ₂ are Natural number, and M ₂ > M ₁ >m;

   A labeling module, configured to input n groups of first image blocks and second image blocks to a trained convolutional neural network, and the trained convolutional neural network The venous vessels are labeled to obtain two sets of n third image blocks labeled with venous vessels. The trained convolutional neural network is trained by a convolutional neural network in a supervised learning manner, and the area of the third image block includes Voxels arranged in an m * m order matrix, where n is the number of the sampling points;

   A mapping module, configured to map the venous blood vessel marks of the n sets of two third image blocks back to the original brain magnetically sensitive weighted image SWI to obtain a venous blood vessel segmentation result.
8. The device for segmenting venous blood vessels in a magnetically sensitive weighted image according to claim 7, wherein the normalization module comprises:

   An extraction unit, configured to use a threshold method to extract all background voxels with a gray value less than 50 in the original brain SWI, and extract a maximum connected region;

   A negation unit, configured to perform a morphological closing operation using a 3 * 3 * 3 size structural element after inverting the maximum connected region to restore the threshold voxel segmentation;

   A calculation unit for calculating a mean value and a standard deviation of a voxel of a brain region in the original brain SWI, subtracting a mean value of each voxel of the brain region in the original brain SWI and dividing by a standard deviation to obtain the Standardized brain SWI.
9. A computing device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the processor implements claims 1 to Steps of the method of any one of 6.
10. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images