CN112598619A

CN112598619A - Method for establishing intracranial vascular simulation three-dimensional narrowing model based on transfer learning

Info

Publication number: CN112598619A
Application number: CN202011324122.8A
Authority: CN
Inventors: 贾艳楠
Original assignee: Xian Cresun Innovation Technology Co Ltd
Current assignee: Xian Cresun Innovation Technology Co Ltd
Priority date: 2020-11-23
Filing date: 2020-11-23
Publication date: 2021-04-02

Abstract

The invention discloses a method for establishing an intracranial vascular simulation three-dimensional narrowing model based on transfer learning, which comprises the following steps: acquiring a bright blood image group and an enhanced black blood image group of intracranial blood vessels; preprocessing each bright blood image and the corresponding enhanced black blood image to obtain a first bright blood image and a first black blood image; carrying out image registration by using mutual information based on Gaussian distribution sampling and a registration method of an image pyramid; obtaining MIP images in all directions from the registered bright blood image group; obtaining a two-dimensional blood vessel segmentation map based on the MIP map and the fundus blood vessel map; the two-dimensional vessel segmentation graph is subjected to back projection synthesis to obtain first three-dimensional vessel volume data, and an intracranial vessel simulation three-dimensional model is obtained by utilizing the second three-dimensional vessel volume data; and aiming at each section of blood vessel in the model, obtaining a numerical value of a target parameter representing the stenosis degree of the section of blood vessel, and marking the intracranial blood vessel simulation three-dimensional model by using the numerical value of the target parameter to obtain a simulated three-dimensional intracranial blood vessel stenosis analysis model.

Description

Method for establishing intracranial vascular simulation three-dimensional narrowing model based on transfer learning

Technical Field

The invention belongs to the field of image processing, and particularly relates to a method for establishing an intracranial vascular simulation three-dimensional narrowing model based on transfer learning.

Background

Medical data showed that the first cause of death among residents in 34 provinces (including hong Kong and Australia) in China was stroke from 1990 to 2017. Cerebral apoplexy is a series of symptoms caused by brain tissue necrosis caused by rupture, stenosis or blockage of intracranial blood vessels, including cerebral hemorrhage, cerebral infarction and the like.

Currently, for clinically assessing the degree of intracranial vascular lesion and vascular stenosis, a lumen-based imaging method, such as Digital Subtraction Angiography (DSA), CT Angiography (CTA), Magnetic Resonance Angiography (MRA), and High-Resolution Magnetic Resonance Angiography (HRMRA), is commonly used.

Because the images corresponding to the bright blood sequence and the black blood sequence obtained by the magnetic resonance blood vessel imaging technology are two-dimensional images, the method has limitations. Is not favorable for simply, quickly and visually obtaining the integral state of the intracranial blood vessel clinically.

Disclosure of Invention

In order to simply, quickly and intuitively obtain the whole state of the intracranial blood vessel in clinic. The embodiment of the invention provides a method for establishing an intracranial vascular simulation three-dimensional stenosis model based on transfer learning. The method comprises the following steps:

acquiring a bright blood image group and an enhanced black blood image group of intracranial blood vessels; the bright blood image group and the enhanced black blood image group respectively comprise K bright blood images and K enhanced black blood images; the images in the bright blood image group and the enhanced black blood image group correspond to each other one by one; k is a natural number greater than 2; taking each bright blood image and the corresponding enhanced black blood image as an image pair, and preprocessing each image pair to obtain a first bright blood image and a first black blood image of the image pair; aiming at each first bright blood image, taking the corresponding first black blood image as a reference, and performing image registration by using a mutual information based on Gaussian distribution sampling and image pyramid registration method to obtain a registered bright blood image group comprising K registered bright blood images; and projecting the registered bright blood image group in three preset directions by using a maximum intensity projection method to obtain MIP (maximum intensity projection) images in all directions. Taking the MIP images in all directions as target domains and the fundus blood vessel images as source domains, and obtaining two-dimensional blood vessel segmentation images corresponding to the MIP images in all directions by using a migration learning method; synthesizing the two-dimensional vessel segmentation maps in the three directions by using a back projection method to obtain first three-dimensional vessel volume data; wherein the voxel value of the blood vessel part in the first three-dimensional blood vessel volume data is 0, and the voxel value of the non-blood vessel part is minus infinity; obtaining an intracranial blood vessel simulation three-dimensional model based on the first three-dimensional blood vessel volume data and second three-dimensional blood vessel volume data corresponding to the registered bright blood image group; aiming at each segment of blood vessel in the intracranial blood vessel simulation three-dimensional model, obtaining a numerical value of a target parameter representing the stenosis degree of the segment of blood vessel; and marking the intracranial blood vessel simulation three-dimensional model by using the numerical value of the target parameter of each section of blood vessel to obtain the simulated three-dimensional intracranial vascular stenosis analysis model.

In the scheme provided by the embodiment of the invention, the bright blood image and the enhanced black blood image obtained by the magnetic resonance blood vessel imaging technology are subjected to image registration by adopting a mutual information and image pyramid registration method based on Gaussian distribution sampling, so that the registration efficiency and the registration precision are improved. And training a network model for segmenting the fundus blood vessels by using the registered bright blood images and the similarity characteristics of the intracranial blood vessel bright blood sequence MIP image and the fundus blood vessel image, performing characteristic transformation on the intracranial blood vessel bright blood sequence MIP image to obtain a characteristic MIP image with the same sample distribution as the fundus blood vessel image, and obtaining two-dimensional blood vessel segmentation images in all directions corresponding to the intracranial blood vessel bright blood sequence MIP image. The embodiment of the invention applies the transfer learning to the field of the segmentation of intracranial blood vessels, and obtains a more accurate blood vessel segmentation effect. And obtaining first three-dimensional blood vessel volume data by using a back projection method, and realizing an intracranial blood vessel simulation three-dimensional model. Aiming at each segment of blood vessel in the intracranial blood vessel simulation three-dimensional model, obtaining a numerical value of a target parameter representing the stenosis degree of the segment of blood vessel; and marking the intracranial blood vessel simulation three-dimensional model by using the numerical value of the target parameter of each section of blood vessel to obtain the simulated three-dimensional intracranial vascular stenosis analysis model. The simulated three-dimensional intracranial vascular stenosis analysis model can provide visual intracranial vascular three-dimensional spatial information without the need of reducing vascular tissue structures, disease characteristics and the like through imagination of doctors, is convenient for visual observation and is convenient for positioning and displaying narrow focus areas. The analytical data about the intracranial vascular stenosis degree can be obtained clinically intuitively and quickly.

Of course, not all of the advantages described above need to be achieved at the same time in the practice of any one product or method of the invention.

The present invention will be described in further detail with reference to the accompanying drawings and examples.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

Fig. 1 is a schematic flow chart of a method for establishing a three-dimensional stenosis simulation model of an intracranial blood vessel based on transfer learning according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of coordinate transformation of an intracranial vascular magnetic resonance image according to an embodiment of the invention;

FIG. 3 shows the results of registration comparison of two search strategies according to an embodiment of the present invention;

FIG. 4 is a pre-registered result of an intracranial vascular magnetic resonance image in accordance with an embodiment of the present invention;

FIG. 5 is a schematic diagram of a region to be registered of an intracranial vascular magnetic resonance image in accordance with an embodiment of the invention;

fig. 6(a) is a bright blood gaussian pyramid and a black blood gaussian pyramid of an intracranial vascular magnetic resonance image according to an embodiment of the invention; fig. 6(b) is a bright blood laplacian pyramid and a black blood laplacian pyramid of an intracranial vascular magnetic resonance image according to an embodiment of the present invention;

FIG. 7 is a result of registration of Laplacian pyramid images of intracranial vascular magnetic resonance images according to an embodiment of the invention;

FIG. 8 is a schematic diagram of a Gaussian pyramid image registration step based on mutual information for an intracranial vascular magnetic resonance image in an embodiment of the invention;

FIG. 9 is a normalized mutual information for different iterations according to an embodiment of the present invention;

FIG. 10 is a registration result of an intracranial vascular magnetic resonance image including a plurality of registration methods of a mutual information pyramid method;

fig. 11 is a result of the registration of the mutual information and image pyramid based on gaussian distribution sampling and the intracranial vascular magnetic resonance image of the mutual information pyramid method in the embodiment of the present invention;

fig. 12 is an exemplary MIP diagram of an embodiment of the present invention;

FIG. 13 is an inverse graph, a characteristic MIP graph corresponding to the MIP graph of the present invention;

FIG. 14 is an effect diagram of a three-dimensional model of an intracranial vascular simulation in accordance with an embodiment of the invention;

FIG. 15 is a graph showing the effect of a simulated three-dimensional intracranial vascular stenosis analysis model according to an embodiment of the invention;

fig. 16 is a simulated three-dimensional intracranial vascular stenosis analysis model and a sectional view display effect diagram according to an embodiment of the invention.

Detailed Description

In order to simply, quickly and intuitively obtain the whole state of the intracranial blood vessel in clinic. The embodiment of the invention provides a method for establishing an intracranial vascular simulation three-dimensional stenosis model based on transfer learning.

As shown in fig. 1, a method for establishing a three-dimensional stenosis simulation model of an intracranial blood vessel based on migration learning according to an embodiment of the present invention may include the following steps:

s1, acquiring a bright blood image group and an enhanced black blood image group of the intracranial blood vessel;

the bright blood image group and the enhanced black blood image group respectively comprise K bright blood images and K enhanced black blood images; the images in the bright blood image group and the enhanced black blood image group correspond to each other one by one; k is a natural number greater than 2;

the bright blood image group is an image group obtained by performing bright blood sequence scanning on intracranial blood vessels by using a magnetic resonance blood vessel imaging technology. In particular, the set of bright blood images may be a TOF-MRA sequence. Among them, TOF is one of bright blood sequence scanning methods, and is called Time of flight (TOF). The enhanced black blood image group is an image group obtained by injecting paramagnetic contrast agent into a patient and then performing black blood sequence scanning on intracranial blood vessels by using a magnetic resonance blood vessel imaging technology. In an embodiment of the invention, the magnetic resonance angiography technique is preferably HRMRA.

The bright blood image group and the enhanced black blood image group correspond in such a manner that the order of images formed in accordance with the scanning time is the same.

S2, taking each bright blood image and the corresponding enhanced black blood image as an image pair, and preprocessing each image pair to obtain a first bright blood image and a first black blood image of the image pair;

this step can be understood as a pre-processing procedure of the image. In an alternative embodiment, the preprocessing of each image pair to obtain the first bright blood image and the first black blood image of the image pair may include S21 and S22:

s21, aiming at each image pair, taking the enhanced black blood image as a reference, carrying out coordinate transformation and image interpolation on the bright blood image, using the similarity measurement based on mutual information, and adopting a preset search strategy to obtain a first bright blood image after pre-registration;

the enhanced black blood image is imaged by coronal plane scanning, while the bright blood image is imaged by axial plane scanning, and the difference of the sequence scanning direction causes the difference of the two final magnetic resonance imaging layers, so that the magnetic resonance images of different imaging layers need to be observed under a standard reference coordinate system through coordinate transformation.

Specifically, the enhanced black blood image and the bright blood image are to-be-registered images, and the enhanced black blood image is used as a reference image, the bright blood image is used as a floating image, and the bright blood image is subjected to coordinate transformation according to the orientation tag information in the DICOM file of the bright blood image, so that the purpose of rotating the bright blood image to the same coordinate system as the enhanced black blood image is achieved, and the scanning direction of the rotated bright blood image is also changed into a coronal plane.

To facilitate an understanding of the method of the embodiments of the present invention, a brief description is provided below in connection with an image registration process, which can be understood by referring to the related art.

For the registration of the two images a and B, each coordinate position in the image a is actually mapped to the image B through a mapping relationship. Specific coordinate transformation methods may include rigid body transformation, affine transformation, projective transformation, nonlinear transformation, and the like. Because the intracranial blood vessel can be regarded as a rigid body, the embodiment of the invention selects rigid body transformation as a coordinate transformation method.

However, in the coordinate transformation process, the coordinate system of the floating image may stretch or deform, the image pixel coordinate after the coordinate transformation does not completely coincide with the sampling grid of the original image, that is, the original integer pixel coordinate point may not be an integer after the coordinate transformation, which causes some areas of the image to lose part of pixels, therefore, in the image coordinate transformation process, the image needs to be resampled and interpolated at the same time to determine the gray value of the image pixel coordinate point after the coordinate transformation, which is convenient for the subsequent processing. The image interpolation method comprises nearest neighbor interpolation, bilinear interpolation, bicubic interpolation and the like. The embodiment of the invention performs experiments on three interpolation methods, and sets 5 evaluation indexes which are respectively root mean square error RMSE, peak signal-to-noise ratio PSNR, normalized cross correlation coefficient NCC, normalized mutual information NMI and Time consumption Time, wherein the smaller the RMSE is, the more accurate the registration is, and the higher the PSNR, NCC and NMI values are, the more accurate the registration is. From the whole experimental data, the accuracy of bicubic interpolation is obviously better than that of nearest neighbor interpolation and bilinear interpolation, so that bicubic interpolation is selected.

After image restoration is carried out on missing pixel points by using an image interpolation method, certain similarity measurement is needed to be used for calculating the similarity between a reference image and a changed floating image, then the optimal similarity measurement is found by using a search strategy, iteration optimization is carried out repeatedly until the similarity measurement of two images reaches the optimal value, iteration is stopped, and finally coordinate conversion is carried out on the floating image according to a determined space transformation matrix (rotation matrix) so as to realize complete image registration. After the images to be registered are optimized by an iterative algorithm, the spatial position registration relationship and the registered images of the two images can be calculated, so that the similarity between the registered floating images and the reference images is the highest.

The scale for measuring the feature similarity between the two images is the similarity measurement, and the selection of the proper similarity measurement can improve the registration accuracy, effectively inhibit noise and the like, and has very important function in the registration of the images. Common similarity measures are mainly classified into three major categories, namely distance measures, correlation measures and information entropy. In the embodiment of the invention, the intracranial blood vessel can be regarded as a rigid body, hardly deforms, and organs such as heart or lung change along with movement such as respiration of a person, so that mutual information or normalized mutual information can be selected as similarity measurement for the intracranial blood vessel, and the registration effect is more accurate.

Mutual information and normalized mutual information are one of information entropies. Mutual Information (MI), which measures the correlation between two images, or the amount of Information contained in each other, is used to explain whether the two images have reached optimal registration, and the larger the value of Mutual Information, the more similar the two images. Normalized Mutual Information (NMI), which is an improvement in Mutual Information measurement, is used as a similarity measurement when the pixel gray scale levels of two images to be registered are similar, so that the obtained registered image has higher accuracy and is more reliable. The value range of NMI is [0,1], and the closer the value is to 1, the more similar the two images are. The concept of normalization mutual information solves the problems that when the overlapping part of two images is small or most of the overlapping area is background information, the image registration based on the mutual information is not high in precision and poor in registration effect, and reduces the sensitivity of the mutual information to the image overlapping area.

Image registration is essentially a multi-parameter optimization problem, namely, spatial coordinate change is performed on images by using a certain search strategy, and finally, the similarity measurement of the two images is optimized, wherein the search strategy and the spatial coordinate change are performed in a mutual intersection manner in the actual calculation process. The algorithm idea is to calculate the similarity measurement between two images in each iteration, adjust the floating image through the operations of coordinate transformation such as translation or rotation and the like, and interpolate the images at the same time until the similarity measurement of the two images is the maximum. Currently, commonly used search strategies include a gradient descent optimizer, (1+1) -ES based on an Evolution Strategy (ES), and the like, and the predetermined search Strategy in the embodiment of the present invention may be selected as needed.

Specific experimental results are shown as follows, referring to fig. 2, fig. 2 is a schematic diagram of coordinate transformation of an intracranial vascular magnetic resonance image according to an embodiment of the present invention, where a first row is an enhanced black blood image and a bright blood image, respectively, and a second row is an enhanced black blood image and a bright blood image after coordinate transformation, respectively.

Respectively registering 160 bright blood images and 160 enhanced black blood images of corresponding scanning layers by using a gradient descent optimizer and two search strategies (1+1) -ES, wherein the enhanced black blood images are reference images, the bright blood images are floating images, the registration result is shown in fig. 3, and fig. 3 is a registration comparison result of the two search strategies according to the embodiment of the invention; the left image in fig. 3 is the result of the pair-wise display of two images without using optimizer registration, the middle image is the result of the pair-wise display of images using gradient descent optimizer registration, and the right image is the result of the pair-wise display of images using (1+1) -ES optimizer registration. The right image display adopts a montage effect, and a black blood image and a bright blood image are enhanced by using pseudo-color transparency processing, purple is the enhanced black blood image, and green is the bright blood image (the image in the figure is the image after gray processing of the original image, and the color is not shown). As can be seen from the figure, in the images which are not registered by using the optimizer, the enhanced black blood image and the bright blood image are not overlapped and have more shadows; when the gradient descent optimizer is used for registering images, although the registration effect is better than that of a left image, the obvious misalignment phenomenon still occurs at the gray brain matter; in the image using the (1+1) -ES optimizer, the registration result is accurate, and the misaligned shadow part in the image completely disappears. The data shown in table 1 are 3 evaluation indexes of the registration result, namely normalized mutual information NMI, normalized cross correlation coefficient NCC and algorithm Time. From the experimental result graph, the registration image effect of (1+1) -ES is displayed more clearly and is better than that of a gradient descent optimizer; from the experimental data, the three evaluation indexes all represent the good performance of the (1+1) -ES optimizer, and therefore, in the embodiment of the present invention, the predetermined search strategy is preferably (1+1) -ES.

TABLE 1 analysis of results under different search strategies

^aThe value in (1) is based on the mean value of the evaluation indexes of the registration of 160 bright blood images and 160 enhanced black blood images +/-mean square error

Referring to fig. 4, fig. 4 is a diagram illustrating pre-registered results of an intracranial vascular magnetic resonance image according to an embodiment of the present invention. The left image is a pre-registered first bright blood image, wherein the interpolation method adopts bicubic interpolation; the middle image is an enhanced black blood image, both images are coronal planes, the right image is an effect image obtained by directly superimposing the two images, and the right image shows that although the bright blood image and the enhanced black blood image under the current imaging layer can be observed under the same coronal plane after pre-registration, the bright blood image and the enhanced black blood image are still misaligned, so that subsequent image fine registration is required.

Through the pre-registration of the step, the magnetic resonance images of the same scanning layer can be compared under the same coordinate system preliminarily, but because the scanning time of the bright blood sequence and the scanning time of the black blood sequence are different, and the patient possibly moves slightly before and after the scanning, the operation is only a rough coordinate transformation, the complete registration of the multi-mode magnetic resonance images can not be realized only through the pre-registration, but the step can omit unnecessary processing procedures for the subsequent accurate registration link, and the processing speed is improved.

S22, the same area content as the scanning range of the first bright blood image is extracted from the enhanced black blood image, and a first black blood image is formed.

Because the scanning ranges of the blood vessel imaging in different magnetic resonance sequences are different, after the bright blood image is subjected to image coordinate transformation, the information of the coronal plane of the bright blood image is not rich in the information of the enhanced black blood image, so that the same scanning area can be extracted from the enhanced black blood image according to the scanning area of the first bright blood image, and the registration range of the subsequent image is reduced.

Optionally, S22 may include the following steps:

1. obtaining edge contour information of a blood vessel in the first bright blood image;

specifically, the edge contour information may be obtained by using a Sobel edge detection method or the like. The edge profile information contains coordinate values of the respective edge points.

2. Extracting the minimum value and the maximum value of the abscissa and the ordinate from the edge profile information, and determining an initial extraction frame based on the obtained four coordinate values;

in other words, in the edge profile information, extracting a minimum abscissa value, a maximum abscissa value, a minimum ordinate value and a maximum ordinate value, and determining four vertexes of the square frame by using the four coordinate values, thereby obtaining an initial extracted frame;

3. in the size range of the first bright blood image, the size of the initial extraction frame is respectively enlarged by a preset number of pixels along four directions to obtain a final extraction frame;

wherein, the four directions are respectively the positive and negative directions of the horizontal and vertical coordinates; the preset number is reasonably selected according to the type of the blood vessel image, so as to ensure that the expanded final extraction frame does not exceed the size range of the first bright blood image, for example, the preset number may be 20.

4. And extracting the corresponding area content in the final extracted frame from the enhanced black blood image to form a first black blood image.

And extracting the content of the corresponding area in the enhanced black blood image according to the coordinate range defined by the final extraction frame, and forming the extracted content into a first black blood image. The step obtains the common scanning range of the magnetic resonance images under the two modes by extracting the region to be registered, thereby being beneficial to subsequent rapid registration.

Referring to fig. 5, fig. 5 is a schematic diagram of a region to be registered of an intracranial vascular magnetic resonance image according to an embodiment of the present invention, where the left image is a first bright blood image after pre-registration, the right image is an enhanced black blood image, and the box is a region to be extracted in the enhanced black blood image. The region contains the common scanning range of a bright blood sequence and a black blood sequence in an intracranial vascular magnetic resonance image, and useful information can be focused more quickly by determining the region to be extracted.

The two-step preprocessing process of the embodiment of the invention plays a very important role, the preprocessed image can pay more attention to useful information and exclude irrelevant information, and in actual use, the image preprocessing can be used for improving the reliability of intracranial blood vessel image registration and identification.

S3, aiming at each first bright blood image, taking the corresponding first black blood image as a reference, and carrying out image registration by using a mutual information based on Gaussian distribution sampling and an image pyramid registration method to obtain a registered bright blood image group comprising K registered bright blood images;

in an alternative embodiment, S3 may include S31-S34:

s31, sampling by adopting Gaussian distribution to select a part of preprocessed image pair as a test image pair;

the test image pair of the embodiment of the invention is the image pair to be registered, and the random selection of the image pair to be registered of the embodiment of the invention adopts Gaussian distribution sampling. The scanning directions of the bright blood image and the enhanced black blood image are different, and in the image preprocessing process, in order to observe the bright blood image and the enhanced black blood image on the same imaging layer, the bright blood image is subjected to coordinate transformation and interpolation, so that each bright blood image corresponds to the enhanced black blood image on the current layer; meanwhile, due to the fact that the scanning ranges of the bright blood image and the enhanced black blood image are different, data of the edge layer of the bright blood image may not be complete. In summary, the bright blood data and the enhanced black blood data of the scanned middle layer are most abundant, so that the gaussian mean μ is selected to be half of the total number of the images to be registered, and the probability of registration of the images in the middle layer selected by gaussian random is the largest.

S32, performing image registration on the first bright blood image and the first black blood image in each test image pair by adopting a registration method based on mutual information and an image pyramid to obtain a rotation matrix corresponding to the first bright blood image in the test image pair after registration; in an optional embodiment, S32 may specifically include steps S321 to S324:

s321, aiming at each test image pair, obtaining a bright blood Gaussian pyramid from the first bright blood image based on downsampling processing, and obtaining a black blood Gaussian pyramid from the first black blood image; the bright blood Gaussian pyramid and the black blood Gaussian pyramid comprise m images with resolution ratios which are sequentially reduced from bottom to top; m is a natural number greater than 3;

in order to improve the accuracy of image registration and avoid the convergence of an image to a local maximum value in the registration process, a multi-resolution strategy can be used for solving the problem of a local extreme value, and meanwhile, the multi-resolution strategy can improve the execution speed of the algorithm and increase the robustness under the condition of meeting the image registration precision. The method is an effective way to improve the registration accuracy and speed by increasing the complexity of the model, namely, in the registration process, the registration is performed in the order from coarse registration to fine registration, firstly, the registration is performed on the low-resolution image, and then, on the basis of the completion of the registration of the low-resolution image, the registration is performed on the high-resolution image.

In an alternative embodiment, the step S321 may include:

obtainingFiltering the input image of the ith layer by using a Gaussian kernel, deleting even rows and even columns of the filtered image to obtain an image G of the ith layer of the Gaussian pyramid_iAnd the ith layer image G_iObtaining an i +1 layer image G of a Gaussian pyramid as an i +1 layer input image_i+1；

Wherein i is 1, 2, …, m-1; when the gaussian pyramid is a bright blood gaussian pyramid, the input image of the 1 st layer is a first bright blood image, and when the gaussian pyramid is a black blood gaussian pyramid, the input image of the 1 st layer is a first black blood image.

Specifically, the multiple images in the gaussian pyramid are corresponding to the same original image with different resolutions. The Gaussian pyramid acquires an image through Gaussian filtering and downsampling, and each layer of construction steps can be divided into two steps: firstly, smoothing filtering is carried out on an image by using Gaussian filtering, namely filtering is carried out by using a Gaussian kernel; and then deleting even rows and even columns of the filtered image, namely reducing the width and height of the lower layer image by half to obtain the current layer image, so that the current layer image is one fourth of the size of the lower layer image, and finally obtaining the Gaussian pyramid by continuously iterating the steps.

The embodiment of the invention optimizes the two-dimensional Gaussian template. The two-dimensional Gaussian filter can be split into two independent one-dimensional Gaussian filters, and image filtering is performed in the horizontal direction and the vertical direction respectively. The Gaussian function is separated, so that the edge generated by the two-dimensional Gaussian template can be eliminated, and the running speed of the program can be greatly accelerated. Compared with other blurring filters, the Gaussian filtering can not only realize the blurring effect of the image, but also better keep the marginal effect.

In this step, the first bright blood image and the first black blood image after the preprocessing are subjected to the processing, so that a bright blood gaussian pyramid and a black blood gaussian pyramid can be obtained. As shown in fig. 6(a), fig. 6(a) is a bright blood gaussian pyramid and a black blood gaussian pyramid of an intracranial vascular magnetic resonance image according to an embodiment of the present invention.

These resolutions are gradually reduced, and the images from the same image combined at different resolutions are arranged to resemble a pyramid, and are therefore referred to as an image pyramid, where the highest resolution image is located at the bottom of the pyramid and the lowest resolution image is located at the top of the pyramid. In the aspect of image information processing, the multi-resolution images can more easily acquire the essential characteristics of the images compared with the traditional single-resolution images.

S322, based on the upsampling processing, obtaining a bright blood Laplacian pyramid by using a bright blood Gaussian pyramid, and obtaining a black blood Laplacian pyramid by using a black blood Gaussian pyramid; wherein the bright blood Laplacian pyramid and the black blood Laplacian pyramid comprise m-1 images with resolution which is sequentially reduced from bottom to top;

since the gaussian pyramid is downsampled, i.e., the image is reduced, a portion of the data of the image is lost. Therefore, in order to avoid data loss of the image in the zooming process and recover detailed data, the embodiment of the invention uses the Laplacian pyramid to realize image reconstruction together with the Gaussian pyramid, and details are highlighted on the basis of the Gaussian pyramid image.

In an alternative embodiment, the step S322 may include:

for the i +1 layer image G of the Gaussian pyramid_i+1Performing upsampling, and filling the newly added rows and columns with data 0 to obtain a filled image;

performing convolution on the filling image by utilizing a Gaussian kernel to obtain an approximate value of the filling pixel to obtain an amplified image;

the ith layer image G of the Gaussian pyramid_iSubtracting the amplified image to obtain the ith layer image L of the Laplacian pyramid_i；

When the gaussian pyramid is the bright blood gaussian pyramid, the laplacian pyramid is the bright blood laplacian pyramid, and when the gaussian pyramid is the black blood laplacian pyramid, the laplacian pyramid is the black blood laplacian pyramid.

Since the laplacian pyramid is a residual between the image and the original image after downsampling, the laplacian pyramid is compared from bottom to top, and the laplacian pyramid has one layer of higher-level image less than the laplacian pyramid structure.

Specifically, the mathematical formula for generating the Laplacian pyramid structure is shown as (1), wherein L_iIndicating the Laplacian pyramid (bright blood Laplacian pyramid or black blood Laplacian pyramid) of the i-th layer G_iRepresenting the i-th level gaussian pyramid (bright blood gaussian pyramid or black blood gaussian pyramid), and the UP operation is an UP-sampled magnified image, symbol

Is a sign of the convolution of the symbols,

is the gaussian kernel used in constructing the gaussian pyramid. The formula shows that the laplacian pyramid is essentially formed by subtracting residual data of an image which is reduced and then enlarged from an original image, and is a residual prediction pyramid. Since a part of information lost in the previous downsampling operation cannot be completely restored by upsampling, that is, downsampling is irreversible, the display effect of the image after downsampling and upsampling is blurred compared with the original image. By storing the residual between the image and the original image after the down-sampling operation, the detail can be added to the images of different frequency layers on the basis of the Gaussian pyramid image, and the detail and the like can be highlighted.

Corresponding to the gaussian pyramid with 4 layers, the step can obtain a bright blood laplacian pyramid and a black blood laplacian pyramid with 3 image layers. As shown in fig. 6(b), fig. 6(b) is a bright blood laplacian pyramid and a black blood laplacian pyramid of an intracranial vascular magnetic resonance image according to an embodiment of the present invention. The image display uses gamma correction to achieve a clearer effect, and the gamma value is 0.5.

S323, registering images of corresponding layers in the bright blood Laplacian pyramid and the black blood Laplacian pyramid to obtain a registered bright blood Laplacian pyramid;

in an alternative embodiment, the step S323 may include:

aiming at each layer of the bright blood Laplacian pyramid and the black blood Laplacian pyramid, taking the corresponding black blood Laplacian image of the layer as a reference image, taking the corresponding bright blood Laplacian image of the layer as a floating image, using a similarity measure based on mutual information and adopting a preset search strategy to realize image registration, and obtaining the registered bright blood Laplacian image of the layer;

forming a registered Laplacian pyramid of the bright blood from bottom to top according to the sequence of the sequential reduction of the resolution by the registered multilayer Laplacian images of the bright blood;

the black blood laplacian image is an image in the black blood laplacian pyramid, and the bright blood laplacian image is an image in the bright blood laplacian pyramid.

The registration process in this step is similar to the pre-registration process, and the image registration is realized by performing coordinate transformation and image interpolation on the bright blood laplacian image and using the similarity measure and the predetermined search strategy, so that the registered bright blood laplacian image can be obtained. The coordinate transformation, the image interpolation, the similarity measurement and the predetermined search strategy are not repeated.

As shown in fig. 7, fig. 7 is a registration result of laplacian pyramid images of an intracranial vascular magnetic resonance image according to an embodiment of the present invention, where the left image is a reference image in a black blood laplacian pyramid, the middle image is a registered image in a bright blood laplacian pyramid, the right image is an effect image obtained by directly superimposing the left and middle images, and the superimposed image displays a montage effect, and the black blood image and the bright blood image are enhanced by using pseudo-color transparency processing, where purple is the enhanced black blood laplacian pyramid image, and green is the bright blood laplacian pyramid image (the image is an image of an original image subjected to gray processing, and the color is not shown).

And S324, registering the images of all layers in the bright blood Gaussian pyramid and the black blood Gaussian pyramid from top to bottom by using the registered bright blood Laplacian pyramid as superposition information to obtain a registered bright blood Gaussian pyramid, and obtaining a rotation matrix corresponding to the first bright blood image in the test image pair after registration.

In the registration of the step, images with different resolutions in the Gaussian pyramid need to be registered, and the registration of the low-resolution images can easily hold essential characteristics of the images, so that the embodiment of the invention registers the high-resolution images on the basis of the registration of the low-resolution images, namely, the images of the Gaussian pyramid are registered from top to bottom, and the registration result of the image of the upper layer is used as the registration input of the image of the lower layer.

In an optional embodiment, in S324, the registering of the blood-brightening gaussian pyramid and the black blood gaussian pyramid from top to bottom is performed by using the registered blood-brightening laplacian pyramid as the overlay information, so as to obtain the registered blood-brightening gaussian pyramid, which may include:

for the j-th layer from top to bottom in the bright blood Gaussian pyramid and the black blood Gaussian pyramid, taking the black blood Gaussian image corresponding to the layer as a reference image, taking the bright blood Gaussian image corresponding to the layer as a floating image, using similarity measurement based on mutual information, and adopting a preset search strategy to realize image registration to obtain a registered j-th layer bright blood Gaussian image;

performing up-sampling operation on the registered jth layer of bright blood Gaussian image, adding the up-sampling operation to the registered corresponding layer of bright blood Laplacian image, and replacing the jth +1 layer of bright blood Gaussian image in the bright blood Gaussian pyramid by using the added image;

taking the black blood Gaussian image of the j +1 th layer as a reference image, taking the replaced bright blood Gaussian image of the j +1 th layer as a floating image, and using a preset similarity measure and a preset search strategy to realize image registration to obtain a registered bright blood Gaussian image of the j +1 th layer;

where j is 1, 2, …, m-1, the black blood gaussian image is an image in the black blood gaussian pyramid, and the bright blood gaussian image is an image in the bright blood gaussian pyramid.

And repeating the operations until the high-resolution registration of the bottom layer Gaussian pyramid image is completed to obtain the registered bright blood Gaussian pyramid. In the registration process, the pre-registration process is similar to the one described above. The coordinate transformation, the image interpolation, the similarity measurement and the predetermined search strategy are not repeated.

The specific steps of mutual information-based gaussian pyramid image registration are shown in fig. 8, and fig. 8 is a schematic diagram of mutual information-based gaussian pyramid image registration steps of an intracranial vascular magnetic resonance image in an embodiment of the present invention. Firstly, registering the low-resolution black blood Gaussian image of the top layer and the low-resolution bright blood Gaussian image of the top layer based on mutual information; then, performing up-sampling operation on the registered bright blood Gaussian image, and adding the up-sampled bright blood Gaussian image and the bright blood Laplacian image of the corresponding layer which retains high-frequency information and is registered according to the operation to be used as a next layer of bright blood Gaussian image; and then, taking the bright blood Gaussian image obtained by the operation as an input image, registering the input image with the black blood Gaussian image of the corresponding layer, and repeating the operation until the high-resolution registration of the bottom layer Gaussian pyramid image is completed.

In the registration of Gaussian pyramid images based on mutual information, the registration of each layer of bright blood Gaussian image and black blood Gaussian image is carried out by taking normalized mutual information as similarity measurement, and the NMI of the two images is calculated through loop iteration until the NMI reaches the maximum. When the iteration times are too small, accurate registration of the images cannot be completed, but when the iteration times are too large, the calculated amount is increased rapidly, fig. 9 is normalized mutual information under different iteration times of the embodiment of the invention, and when the registration of the first-layer image, namely the bottom-layer image with the highest resolution in the gaussian pyramid reaches the maximum NMI value and the data is stable, the iteration is stopped.

Thus, a registered bright blood Gaussian pyramid is obtained, the coordinate system of the bright blood image is consistent with the coordinate system of the enhanced black blood image, and the images have high similarity, so that the blood vessel image registration process of the embodiment of the invention can be completed. And after registration, a rotation matrix corresponding to the first bright blood image in the test image pair after registration can be obtained.

In order to verify the effectiveness and the practicability of the mutual information and image pyramid-based registration method (referred to as the mutual information pyramid method for short) in the embodiment of the invention, a comparison experiment is also performed, and intracranial vascular magnetic resonance images of five patients are used together, wherein the enhanced black blood image and the bright blood image of the patient A, B, C, D are 160 respectively, and the enhanced black blood image and the bright blood image of the patient E are 150 respectively; meanwhile, an algorithm which only uses DICOM image orientation label information for registration and a registration algorithm based on mutual information measurement are selected and compared with the mutual information pyramid method in the embodiment of the invention, wherein the algorithm based on mutual information measurement is to search the optimal transformation between a reference image and a floating image by a multi-parameter optimization method, so that the mutual information value of the two images is maximum, and the image pyramid algorithm is not used.

The experimental platform was Matlab R2016 b. And combining qualitative analysis and quantitative analysis according to the image registration result of the experiment. In the aspect of qualitative analysis, because large gray scale difference exists between the multi-modal medical images, a difference image obtained by subtracting the registration image from the reference image cannot effectively reflect the registration result of the multi-modal medical images, in the embodiment of the present invention, the registration image is overlapped with the reference image to obtain a color overlapped image capable of reflecting the alignment degree of the registration image and the reference image, the registration effect of the multi-modal registration algorithm is qualitatively analyzed through the color overlapped image, fig. 10 shows the registration result of the multi-modal intracranial vascular magnetic resonance image, and fig. 10 shows the registration result of the intracranial vascular magnetic resonance image of the multiple registration methods including the mutual information pyramid method. Wherein, (a) is a reference image; (b) is a floating image; (c) is an overlay image based on image orientation label information; (d) is an overlay image based on a mutual information metric; (e) the invention discloses a superposed image of a mutual information pyramid method. The figures are gray scale images of the original image, not shown in color. In the aspect of quantitative analysis, since the root mean square error RMSE and the peak signal-to-noise ratio PSNR of the evaluation indexes are not suitable for evaluating images with large gray scale changes, in order to better evaluate the registration result of the multi-modal medical image, the normalized cross-correlation coefficient NCC is adopted, the normalized mutual information NMI is used as the evaluation index, when the values of the normalized cross-correlation coefficient NCC and the normalized mutual information NMI are larger, the higher the image registration accuracy is, and table 2 shows the evaluation index result analysis of different registration algorithms.

Table 2 analysis of the results of different registration methods

^aThe value in (1) is the mean value of the evaluation index +/-mean square error based on the registration of a plurality of images of a patient

And (3) qualitative analysis: as is apparent from the overlaid images of fig. 10, the method based on mutual information metric has a large registration shift, and the analysis reason may be that it is easy to fall into a local optimum value rather than a global optimum value only using the method based on mutual information metric; the registration effect based on the image orientation label information is not good enough, and the images are partially not overlapped; the mutual information pyramid method has good image registration effect, the images are displayed more clearly, and the images are almost completely overlapped.

Quantitative analysis: as can be seen from table 2, from the two evaluation indexes NCC and NMI, compared with the registration algorithm using only the orientation tag information of the DICOM image and the registration algorithm based on the mutual information measurement, the mutual information pyramid method according to the embodiment of the present invention is improved in registration accuracy, which shows that the registration method based on the mutual information and the image pyramid according to the embodiment of the present invention can well process the registration of the multi-modal intracranial vascular magnetic resonance image.

S33, obtaining the mean value of the rotation matrix of all the test image pairs;

in the last step, the rotation matrix corresponding to the first bright blood image after registration in each test image pair can be obtained, and then the mean value of the rotation matrices of all the test image pairs can be calculated and obtained.

And S34, performing coordinate transformation on the first bright blood image in the other preprocessed image pairs except the test image pair by using the mean value of the rotation matrix, completing image registration, and obtaining a plurality of registered image pairs.

Considering that when a patient uses a magnetic resonance bright blood sequence scan, if slight movement occurs, the coordinate position of an intracranial blood vessel image obtained by the bright blood sequence scan slightly changes, and then each bright blood image needs to be subjected to a spatial coordinate transformation operation so as to keep the same coordinate position as that of an enhanced black blood image. Mutual information of every two images to be registered needs to be continuously and iteratively calculated in the registration process of the registration method based on the mutual information and the image pyramid, when the size and the number of the images are large, time consumption is overlarge, and if the registration method based on the mutual information and the image pyramid is used for all the images to be registered, the calculation speed is slow. The inventor considers that the intracranial blood vessel can be regarded as a rigid body, which is different from organs such as heart or lung and the like which change along with the movement of human breath and the like, so the space coordinate transformation operation of each bright blood image is almost consistent, namely almost the same rotation matrix is used. Referring to table 3, table 3 shows the mean and mean square error of the rotation matrix calculated by using the mutual information pyramid method, the mean square error of the rotation matrix is very small as can be seen from the data, and the rotation matrix is obtained by registering 160 400 × 400 enhanced black blood images and 160 400 × 400 bright blood images derived from the patient a by using the mutual information pyramid method. Then, the calculated rotation matrix can be registered by a few bright blood images of the patient to perform the same spatial coordinate transformation on all the bright blood images without the need of solving the rotation matrix for each bright blood image, thereby accelerating the image registration process.

TABLE 3 mean and mean square error of rotation matrix

In the step, the mean value of the rotation matrix, namely a new rotation matrix, is used for carrying out coordinate transformation on the first bright blood image in the rest of the preprocessed image pairs, so that the registration of all images can be completed quickly, and the registration speed is improved greatly. For the processes of implementing coordinate transformation of the bright blood image by using the rotation matrix, image interpolation, using the similarity measurement based on mutual information, and implementing image registration by using a predetermined search strategy, reference may be made to the foregoing, and it may also be understood by combining with the related art, and details are not described herein again.

After the step, a plurality of registered bright blood images can be obtained, and the bright blood images and the corresponding enhanced black blood images are in the same coordinate system and have higher similarity.

In order to verify the feasibility of the registration method based on the mutual information of gaussian distribution sampling and the image pyramid in the embodiment of the present invention, intracranial vascular magnetic resonance images of five patients were used together for registration, wherein the number of the enhanced black blood images and the number of the bright blood images of patient A, B, C, D were 160, and the number of the enhanced black blood images and the number of the bright blood images of patient E were 150, respectively. For the image registration result of an experiment, because the multi-mode magnetic resonance images are different in acquisition principle and different in presented information, a unified gold standard is not available at present to evaluate which registration algorithm is the best, and the quality of the registration result is evaluated according to a specific registration object and an application purpose, so that qualitative analysis and quantitative analysis are combined. In terms of qualitative analysis, the registration algorithm results are qualitatively analyzed by color overlay images that reflect the degree of alignment of the registered image and the reference image, and fig. 11 shows a comparison of the registration results of the multi-modality intracranial vascular magnetic resonance images. Fig. 11 is a result of the registration of the mutual information based on gaussian distribution sampling and the image pyramid method and the intracranial vascular magnetic resonance image by the mutual information pyramid method according to the embodiment of the present invention. Wherein (a) is a reference image; (b) is a floating image; (c) an overlay image for mutual information pyramid method; (d) for the overlapped images of the method of the invention, the standard deviation sigma is taken as 1; (e) for the overlapped images of the method of the invention, the standard deviation sigma is taken as 2; (f) for the overlapped images of the method of the invention, the standard deviation sigma is taken as 3; (g) for the overlapped images of the method of the invention, the standard deviation sigma is taken as 4; (h) for the overlapped images of the method of the invention, the standard deviation sigma is taken as 5; (i) for the overlaid images of the method of the invention, the standard deviation σ is taken to be 6. Each image is a processed gray scale image, not shown in color. In the aspect of quantitative analysis, a normalized cross-correlation coefficient NCC is adopted, normalized mutual information NMI is used as an evaluation index, and the higher the value of NCC and NMI is, the higher the image registration precision is. Table 4 shows the registration results of all image pairs of each patient using the mutual information and image pyramid-based registration method (referred to as the mutual information pyramid method for short) and the mutual information and image pyramid-based registration method based on gaussian distribution sampling (referred to as the present method for short) proposed in the embodiment of the present invention (the rest of the patient data is not shown by space limitation), and the experimental platform is Matlab R2016 b. As the method of the invention does not need to register all images, only needs to randomly select a few images for registration, the experiment sets the Gaussian distribution mean value mu to be half of the total number of the images to be registered, and randomly extracts 20 enhanced black blood images to register with the corresponding bright blood images when the standard deviation sigma is 1, 2, 3, 4, 5 and 6 respectively.

TABLE 4 analysis of results of different registration algorithms

The value in a is based on the mean value of evaluation indexes of the registration of 160 bright blood images and 160 enhanced black blood images +/-mean square error

^aThe value in (1) is based on the matching of 160 bright blood images and 160 enhanced black blood imagesMean value of standard evaluation index ± mean square error

^aThe value in (1) is based on the mean value of evaluation indexes of registration of 150 bright blood images and 150 enhanced black blood images +/-mean square error

As is apparent from the overlapped images of fig. 11, the registered images of the mutual information pyramid method and the registered images of the method of the present invention have good performance, and the images are almost completely overlapped together; as can be seen from table 4, from the two evaluation indexes, NCC and NMI, although the accuracy of the method of the present invention is lower than that of the mutual information pyramid method, the difference between them is not too large under different gaussian distribution function settings. When the size of the images participating in registration is large and the number of the images is large, the mutual information pyramid method has a large calculation amount. The method improves the algorithm aiming at the rigid space transformation characteristics of the intracranial blood vessels, accelerates the image registration by utilizing the similarity of the transformation matrix, and the experimental result proves that the time consumed by the improved algorithm registration is only one fifth of the registration time of the mutual information pyramid method, so that the registration speed can be greatly improved, and the registration of the multi-mode intracranial blood vessel magnetic resonance image can be well realized.

In the registration scheme provided by the embodiment of the invention, the intracranial blood vessel can be regarded as a rigid body, and the space coordinate transformation operation of each bright blood image is almost consistent, so that the same rotation matrix can be used. Therefore, on one hand, a small number of image pairs are selected for carrying out image registration of the bright blood images, the average value of the rotation matrix of the registered small number of bright blood images is utilized, the same space coordinate transformation is carried out on the bright blood images of the rest image pairs, and the rotation matrix of each of the rest bright blood images is not required to be solved, so that the image registration process can be accelerated. On the other hand, in the image registration process, mutual information is used as similarity measurement, and an image pyramid algorithm is adopted to increase the complexity of the model, so that the registration accuracy and speed can be improved. Compared with the prior art, the method needs doctors to observe the bright blood image and the black blood image of the intracranial blood vessel according to spatial imagination and subjective experience. The embodiment of the invention adopts the image registration method to unify the bright blood image and the enhanced black blood image in the same coordinate system, so that doctors can conveniently understand the intracranial blood vessel images corresponding to the black blood sequence and the bright blood sequence, the comprehensive information required by diagnosis can be simply, conveniently and quickly obtained, and accurate and reliable reference information can be provided for subsequent medical diagnosis, operation plan formulation, radiotherapy plan and the like. The registration scheme provided by the embodiment of the invention can provide a better reference mode for the registration of other medical images, and has great clinical application value.

And S4, projecting the registered bright blood image group in three preset directions by using a maximum intensity projection method to obtain MIP (maximum intensity projection) maps in all directions, taking the MIP maps in all directions as target domains, taking the fundus blood vessel map as a source domain, and obtaining two-dimensional blood vessel segmentation maps corresponding to the MIP maps in all directions by using a migration learning method.

The Maximum Intensity Projection (MIP) is one of the CT three-dimensional image reconstruction techniques, and is referred to as MIP. Which traverses a volume data series along a preselected viewing angle using a set of projection lines, the highest CT value on each projection line being encoded to form a two-dimensional projection image. Is a method of generating a two-dimensional image by calculating the maximum density of pixels encountered along each ray of the scanned object. Specifically, when the fiber bundle passes through an original image of a section of tissue, the pixels with the highest density in the image are retained and projected onto a two-dimensional plane, thereby forming an MIP reconstruction image (referred to as an MIP map in the embodiment of the present invention). The MIP can reflect the X-ray attenuation value of the corresponding pixel, small density change can be displayed on the MIP image, and stenosis, expansion and filling defects of the blood vessel can be well displayed, and calcification on the blood vessel wall and contrast agents in the blood vessel cavity can be well distinguished.

It will be understood by those skilled in the art that the group of registered bright blood images is actually a three-dimensional volume data, and the three-dimensional volume data can be projected in three predetermined directions by using the above MIP method to obtain a two-dimensional MIP map in each direction, where the three predetermined directions include: axial, coronal, and sagittal.

For the MIP method, please refer to the related description of the prior art, which is not repeated herein, refer to fig. 12, and fig. 12 is an exemplary MIP diagram according to an embodiment of the present invention.

The inventor finds out through research that the MIP map of the intracranial vascular bright blood sequence has a distribution of a vascular tree similar to that of the fundus blood vessel. In the intracranial vascular bright blood sequence MIP diagram, a common carotid artery and a vertebral artery are main branches of a blood vessel tree, an intracranial blood vessel is a branch of the blood vessel tree, and a small blood vessel is a small branch of the blood vessel tree. Because the MIP image of the intracranial blood vessel bright blood sequence has certain similarity with the fundus blood vessel image, namely has the same characteristics, the inventor considers that the model pre-trained in the fundus blood vessel (source domain) segmentation task is migrated into the intracranial blood vessel segmentation task by means of a migration learning method and particularly by adopting a characteristic migration mode. Feature based TL (Feature based TL) is to transform the features of the source domain and the target domain into the same space by Feature transformation, assuming that the source domain and the target domain contain some common cross features, so that the source domain data and the target domain data in the space have the same distributed data distribution, and then perform conventional machine learning.

For S4, an optional embodiment may include S41 to S43:

s41, obtaining a target neural network pre-trained for the fundus angiogram segmentation task;

the target neural network is obtained by pre-training according to the fundus blood vessel map data set and the improved U-net network model.

As described above, the embodiment of the present invention intends to migrate a pre-trained model of a fundus blood vessel segmentation task to an intracranial blood vessel segmentation task by means of a feature migration learning manner. Therefore, it is necessary to obtain a mature network model for the vessel segmentation of the fundus blood vessel map. Specifically, obtaining the target neural network may be performed by the following steps:

step 1, obtaining an original network model;

in the embodiment of the invention, the structure of the existing U-net network model can be improved, and each sub-module of the U-net network model is respectively replaced by a residual module with a residual connection form, so that the improved U-net network model is obtained. According to the embodiment of the invention, the residual error module is introduced into the U-net network model, so that the problem that the training error does not decrease or inversely increase due to the disappearance of the gradient caused by the deepening of the layer number of the neural network can be effectively solved.

Step 2, obtaining sample data of the fundus blood vessel map;

embodiments of the present invention acquire a fundus angiogram dataset, the DRIVE dataset, which is a dataset that has been labeled.

And 3, training the original network model by using the sample data of the fundus blood vessel map to obtain the trained target neural network.

The following summary describes some parameter characteristics of the target neural network of embodiments of the present invention:

the improved U-net network model in the embodiment of the invention has 5 levels, and a 2.5M parameter ladder network is formed. Each residual module uses 0.25 droout rate (droout means that the neural network unit is temporarily discarded from the network according to a certain probability in the training process of the deep learning network, generally, the droout rate can be set to be 0.3-0.5); and Batch Normalization (BN) is used, the variance size and the mean position are changed by optimization, so that the new distribution is more suitable for the real distribution of data, and the nonlinear expression capability of the model is ensured. The activating function adopts LeakyRelu; the last layer of the network model is activated using Softmax.

Moreover, because of the problem of uneven foreground and background distribution of the medical image sample, the loss function uses a common Dice coefficient (Dice coefficient) loss function for medical image segmentation, and specifically uses an improved Dice loss function, so as to solve the unstable condition of Dice loss function training.

In the aspect of neural network optimization, an Adam optimization algorithm and default parameters are adopted, and the batch size is 256. 250 epochs are trained using the "reduced learning rate" strategy, setting the learning rates at

epochs

0, 20, and 150 to 0.01, 0.001, and 0.0001, respectively, and the total learning rate to 250. And the data enhancement is carried out by using a random clipping mode, and the training sample in the DRIVE data set is enlarged by 20000 times.

The process of obtaining the target neural network is briefly introduced, and the trained target neural network can realize the blood vessel segmentation of the fundus blood vessel map to obtain a corresponding two-dimensional blood vessel segmentation map.

S42, respectively carrying out gray inversion processing and contrast enhancement processing on the MIP images in all directions to obtain corresponding characteristic MIP images;

the realization of the feature transfer learning requires that a source domain (fundus blood vessel image) and a target domain (intracranial blood vessel bright blood sequence MIP image) have high similarity and realize the same data distribution.

Therefore, in step S42, the MIP map is subjected to the gradation inversion processing and the contrast enhancement processing, and a characteristic MIP map is obtained so that the characteristic MIP map is closer to the fundus blood vessel image.

In an alternative embodiment, S42 may include S421 and S422:

s41, carrying out pixel transformation on the MIP graph by utilizing a gray inversion formula to obtain an inversion graph; wherein, the grayscale inversion formula is t (x) 255-x, x is the pixel value in the MIP map, and t (x) is the pixel value in the inversion map;

the step can be understood in a popular way as grayscale inversion processing, since the pixel range of the MIP map is between 0 and 255, the original brighter region can be darkened and the original darker region can be lightened through the step, specifically, the pixel transformation can be performed through the grayscale inversion formula, the obtained inversion map please refer to the left map in fig. 13, and the left map in fig. 13 is the inversion map corresponding to the MIP map in the embodiment of the present invention.

And S422, enhancing the contrast of the inversion graph by using a contrast-limited self-adaptive histogram equalization method to obtain a characteristic MIP (maximum intensity distribution) graph.

The main purpose of this step is to enhance the contrast of the inversion map to show a clearer vascularity. As for the way of enhancing the Contrast, any one of the prior arts can be used, and in an alternative embodiment, this step may employ a Contrast Limited Adaptive Histogram Equalization (CLAHE) to enhance the Contrast. For the CLAHE method, reference is made to the prior art for understanding, and no further description is given here. The obtained characteristic MIP map is shown in the right diagram of fig. 13, and the right diagram of fig. 13 is the characteristic MIP map corresponding to the MIP map of the embodiment of the present invention. It can be seen that the contrast of the characteristic MIP map is significantly enhanced and the blood vessels are clearer compared to the inversion map.

After S422, corresponding characteristic MIP maps can be obtained for the MIP maps in each direction.

In the embodiment of the invention, the cross characteristics of the intracranial blood vessel bright blood sequence MIP and the fundus blood vessel image are considered, so that the MIP image characteristics are mapped to the fundus blood vessel image by adopting a characteristic migration learning method, and the intracranial blood vessel input sample and the fundus blood vessel input sample corresponding to the target neural network have the same sample distribution. Wherein, the S51 and the S52 can be in no sequence.

S43, respectively inputting the feature MIP graphs of each direction into a target neural network to obtain corresponding two-dimensional blood vessel segmentation graphs;

and respectively inputting the characteristic MIP images of all directions into a target neural network to obtain a two-dimensional blood vessel segmentation image corresponding to each direction, wherein the obtained two-dimensional blood vessel segmentation image is a binary image, namely pixels are only 0 and 255, white represents a blood vessel, and black represents a background.

S5, synthesizing the two-dimensional vessel segmentation maps in the three directions by using a back projection method to obtain first three-dimensional vessel volume data;

the principle of the back projection method is to evenly distribute measured projection values to each passing point according to the original projection path, back-project the projection values in all directions, and accumulate the back-projected images at all angles to estimate the original image. By synthesizing the two-dimensional vessel segmentation maps in the three directions by using a back projection method, three-dimensional volume data can be obtained, which is referred to as first three-dimensional vessel volume data in the embodiment of the invention. The back projection method in the embodiment of the present invention may be a direct back projection method, a filtered back projection method, a convolution back projection method, and the like, which is not limited herein.

In the embodiment of the present invention, the voxel value of the blood vessel portion in the obtained first three-dimensional blood vessel volume data is 0, and the voxel value of the non-blood vessel portion is minus infinity through the pixel control of the back projection method.

And S6, obtaining an intracranial blood vessel simulation three-dimensional model based on the first three-dimensional blood vessel volume data and the second three-dimensional blood vessel volume data corresponding to the registered bright blood image group.

In an alternative embodiment, S6 may include S61 and S62:

s61, adding the first three-dimensional blood vessel volume data and the second three-dimensional blood vessel volume data to obtain third three-dimensional blood vessel volume data;

the method can be used for directly correspondingly adding each voxel value in the first three-dimensional blood vessel volume data and the second three-dimensional blood vessel volume data to obtain third three-dimensional blood vessel volume data, and cerebrospinal fluid and fat signals with the same intracranial and blood vessel signal intensity can be eliminated through the step.

And S62, processing the third three-dimensional blood vessel volume data by using a threshold segmentation method to obtain an intracranial blood vessel simulation three-dimensional model.

The threshold segmentation method is an image segmentation technology based on regions, and the principle is to divide image pixel points into a plurality of classes. The purpose of image thresholding is to divide the set of pixels by gray level, each resulting subset forming a region corresponding to the real scene, each region having consistent properties within it, while adjacent regions do not have such consistent properties. Threshold segmentation is a method for processing an image into a high-contrast, easily recognizable image with a proper pixel value as a boundary.

The threshold segmentation method adopted by the embodiment of the invention comprises a maximum inter-class variance method, a maximum entropy, an iteration method, a self-adaptive threshold, a manual method, an iteration method, a basic global threshold method and the like. In an alternative implementation manner, the embodiment of the present invention may adopt a maximum inter-class variance method.

The maximum inter-class variance method (or called Otsu method, OTSU for short) is a method for automatically solving a threshold value suitable for a bimodal situation, and uses a clustering idea to divide the gray scale number of an image into two parts according to gray scale, so that the gray scale difference between the two parts is maximum, the gray scale difference between each part is minimum, and a suitable gray scale is searched for through variance calculation to divide the gray scale. Therefore, the OTSU algorithm can be adopted to automatically select the threshold value for binarization during binarization. The method divides an image into a background part and an object part according to the gray characteristic of the image. The larger the inter-class variance between the background and the object, the larger the difference between the two parts constituting the image. The OTSU algorithm is considered as the optimal algorithm for selecting the threshold value in image segmentation, is simple to calculate and is not influenced by the brightness and the contrast of an image. Thus, a segmentation that maximizes the inter-class variance means that the probability of false positives is minimized. Performing S72 with an OTSU may include the steps of:

firstly, calculating a first threshold corresponding to centered fourth three-dimensional blood vessel volume data in third three-dimensional blood vessel volume data by using the OTSU;

in this step, one threshold corresponding to a plurality of images in one small cube (referred to as fourth three-dimensional blood vessel volume data) located near the middle of the large three-dimensional cube of the third three-dimensional blood vessel volume data is determined as a first threshold by using the OTSU method. Because the blood information is substantially concentrated in the middle of the image in the third three-dimensional blood vessel volume data, the small cube data (fourth three-dimensional blood vessel volume data) in the middle is selected to determine the first threshold value in the third three-dimensional blood vessel volume data, so that the calculation amount of the threshold value can be reduced, the calculation speed can be improved, and the first threshold value can be accurately applied to all the blood information in the third three-dimensional blood vessel volume data.

For the size of the fourth three-dimensional blood vessel volume data, the central point of the third three-dimensional blood vessel volume data can be determined firstly, and then the preset side length extends in six directions corresponding to the cube, so that the size of the fourth three-dimensional blood vessel volume data is determined; the preset side length may be determined according to an empirical value including a Willis ring, such as 1/4 that is the side length of the cube of the third three-dimensional blood vessel volume data. The Willis loop is the most important collateral circulation pathway in the cranium, linking the bilateral hemisphere with the anterior and posterior circulation.

And then, threshold segmentation of the third three-dimensional blood vessel volume data is realized by utilizing the first threshold, and an intracranial blood vessel simulation three-dimensional model is obtained.

It can be understood by those skilled in the art that, by threshold segmentation, the gray-scale value of the point on the image corresponding to the third three-dimensional blood vessel volume data can be set to 0 or 255, that is, the whole image exhibits a distinct black-and-white effect, the blood information is highlighted as white, and the irrelevant information is displayed as black. For the processing procedure of threshold segmentation, please refer to the prior art, and will not be described herein. Fig. 14 shows the effect of the three-dimensional intracranial vascular simulation model according to the embodiment of the invention, fig. 14 shows the effect of the three-dimensional intracranial vascular simulation model according to the embodiment of the invention. The map is grey-scale processed and the colours are not shown, in practice the vessel regions may be displayed in colour, such as red.

In an optional embodiment, after S6, the method may further include:

displaying the intracranial blood vessel simulation three-dimensional model, specifically, displaying the intracranial blood vessel simulation three-dimensional model on a display screen of a computer and other equipment so as to facilitate observation by a doctor; and, it is reasonable that it can be displayed simultaneously with the bright blood image and the enhanced black blood image.

S7, aiming at each segment of blood vessel in the intracranial blood vessel simulation three-dimensional model, obtaining the value of the target parameter representing the stenosis degree of the segment of blood vessel;

in an alternative embodiment, S7 includes:

s71, aiming at each section of blood vessel in the intracranial blood vessel simulation three-dimensional model, segmenting from three preset positions to obtain two-dimensional sectional diagrams of each position;

in the step, the blood vessels in the intracranial blood vessel simulation three-dimensional model can be divided, and for each section of blood vessel, the blood vessel simulation three-dimensional model is divided from three preset positions to obtain a two-dimensional sectional view of each position.

Wherein, three preset positions include: axial, coronal, and sagittal.

The segmentation of a certain orientation is performed on the blood vessel simulation three-dimensional model to obtain a two-dimensional sectional view of the orientation, which can be implemented by adopting the prior art and is not described herein again.

S72, carrying out corrosion operation on the blood vessel in the two-dimensional sectional diagram of each direction, and recording the target corrosion times when the blood vessel is corroded to a single pixel;

the corrosion operation is one of morphological operations, and the basic idea of the morphological operation is to extract image data interested by a user by using a structural element in an original image, remove irrelevant information, retain the essential characteristics of an interested region, generally apply to a binary image, generally apply to extracting a connected region or eliminating noise and the like, and have wide application in image processing.

The corrosion operation can eliminate the edge data of the object, the corroded object has a smaller area than the original area and even can completely disappear, and the corrosion can also break some small and long communication areas. The etching operation can be recorded as A theta B and defined as

Wherein B is a structural element and A is an original drawing.

When the blood vessel is thick, a plurality of corrosion operations can be carried out, and when the blood vessel is thin, only a few corrosion operations can be carried out. It will be appreciated by those skilled in the art that the blood vessel erodes to a single pixel, i.e., to the thinnest state, which may be a point or a line. The specific process of the etching operation can be referred to the related art, and is not described herein.

In step S72, performing erosion operation on the blood vessel in the axial two-dimensional sectional view, and recording the target erosion times n corresponding to the erosion of the blood vessel in the axial two-dimensional sectional view to a single pixel₁(ii) a Carrying out corrosion operation on the blood vessel in the two-dimensional sectional diagram of the coronal position, and recording the corresponding target corrosion times n when the blood vessel in the two-dimensional sectional diagram of the azimuth corrodes to a single pixel₂(ii) a Carrying out corrosion operation on the blood vessel in the two-dimensional sectional diagram of the sagittal position, and recording the corresponding target corrosion times n when the blood vessel in the two-dimensional sectional diagram of the azimuth corrodes to a single pixel₃；

Namely, the target corrosion times corresponding to the three orientations respectively comprise: target corrosion times n corresponding to axial position₁Target erosion number n corresponding to crown position₂Number of target erosion times n corresponding to sagittal position₃。

S73, obtaining the value of the target parameter representing the stenosis degree of the section of the blood vessel according to the target corrosion times of the section of the blood vessel corresponding to the three directions respectively;

in an alternative embodiment, the target parameter includes stenosis rate and/or flatness; those skilled in the art will appreciate that both of these parameters may be indicative of the degree of vascular stenosis.

When the target parameter includes a stenosis rate, S73 may include:

according to n₁、n₂、n₃Obtaining the value of the stenosis rate of the section of blood vessel by using a stenosis rate formula of the blood vessel; wherein, the stenosis rate formula is:

wherein, the resolution is the resolution of each azimuth two-dimensional sectional image (the resolution of the three azimuth two-dimensional sectional images is the same), and the smaller the numerical value of the stenosis rate is, the narrower the blood vessel is.

When the target parameter includes flatness, S53 may include:

according to n₁、n₂、n₃All right (1)Obtaining the value of the flatness of the section of the blood vessel by using a blood vessel flatness formula; wherein, the flatness formula is as follows:

a larger value of the degree of flattening indicates a narrower vessel.

S8, marking the intracranial blood vessel simulation three-dimensional model by using the numerical value of the target parameter of each segment of blood vessel to obtain the simulated three-dimensional intracranial blood vessel stenosis analysis model.

Through the steps, the numerical value of the target parameter of each segment of blood vessel can be obtained, and then the numerical values of each segment of blood vessel can be marked on the intracranial blood vessel simulation three-dimensional model to obtain the simulated three-dimensional intracranial blood vessel stenosis analysis model. The numerical value of the target parameter of each point is embedded into the simulated three-dimensional intracranial vascular stenosis analysis model, so that the numerical value of the target parameter of each point can be extracted and displayed when needed, a doctor can conveniently obtain the data of the vascular stenosis degree of each position in time when observing the overall three-dimensional vascular state, for example, when the simulated three-dimensional intracranial vascular stenosis analysis model is displayed on a display screen of a computer, the numerical value of the stenosis rate and/or the flatness of the mouse position point can be displayed in a blank area of the model.

In an alternative embodiment, S8 may include:

and marking the intracranial blood vessel simulation three-dimensional model by using the numerical values of the target parameters of each section of blood vessel and adopting the color corresponding to each numerical value to obtain the simulated three-dimensional intracranial vascular stenosis analysis model.

For convenience of visual display, different numerical values can be marked on the blood vessel simulation three-dimensional model by different colors to obtain a simulated three-dimensional blood vessel stenosis analysis model, for example, multiple colors from light to dark can be correspondingly marked for stenosis rate numerical values from small to large, and for flatness numerical values, because the numerical values are fewer and only 2 numerical values are possible, two colors which are distinguished from the stenosis rate can be correspondingly marked. The narrowing degree of the blood vessel can be more intuitively shown by adopting the color display of different tones, so that the attention of a doctor can be attracted.

In a preferred embodiment, the color corresponding to different values on one intracranial blood vessel simulation three-dimensional model can be used for marking the stenosis rate value, and the color corresponding to different values on the other intracranial blood vessel simulation three-dimensional model can be used for marking the flatness value, so that a doctor can observe the stenosis rate condition and the flatness condition respectively.

Referring to fig. 15, fig. 15 is a graph showing the effect of the simulated three-dimensional intracranial vascular stenosis analysis model according to the embodiment of the invention. Wherein the left graph is the stenosis rate marking effect and the right graph is the flatness marking effect. In practice, different colors are displayed on the model, so that the degree of narrowing can be distinguished, for example, a thinner part of a blood vessel is warm, the narrowest part is red, a thicker part of the blood vessel is cool, the thickest part is green, and the like, a white arrow indicates abrupt narrowing of the intracranial blood vessel, and color display with different colors can more intuitively show the narrowing of the blood vessel. In the figure are the effects of the grey scale processing, the colours not being shown.

The embodiment of the invention provides a simulated three-dimensional intracranial vascular stenosis analysis model and simultaneously provides two-dimensional sectional views of three directions, namely images of a coronal plane, a sagittal plane and an axial plane of a current point corresponding to each point in the simulated three-dimensional intracranial vascular stenosis analysis model are displayed. Referring to fig. 16, fig. 16 is a diagram showing the effect of a simulated three-dimensional intracranial vascular stenosis analyzing model and a sectional view according to an embodiment of the invention. In fig. 16, there may be a blood vessel narrowing at the warm tone of the blood vessel, there is no obvious blood vessel narrowing at the cold tone, and the three two-dimensional images on the right side of the image are respectively imaged from top to bottom on the axial plane, the sagittal plane and the coronal plane where the current point is located; when the simulated three-dimensional intracranial vascular stenosis analysis model is displayed, the functions of measuring the distance by two points and measuring the angle by three points can be realized by using the points with three colors, such as red, green and blue, and the three points are displayed on the left lower side of the display screen, and the volume size of the currently selected model is displayed on the right lower side of the display screen. So that the doctor can obtain more detailed data of the intracranial blood vessel.

In the scheme provided by the embodiment of the invention, firstly, the bright blood image and the enhanced black blood image obtained by the magnetic resonance blood vessel imaging technology are subjected to image registration by adopting a mutual information and image pyramid registration method based on Gaussian distribution sampling, so that the registration efficiency can be improved, and the registration accuracy of the images is improved layer by layer from low resolution to high resolution. The bright blood image and the enhanced black blood image can be unified under the same coordinate system through the image registration, so that subsequent unified observation is facilitated. In the field, the bright blood images after registration are two-dimensional images, and although the registration is performed, and the corresponding enhanced black blood images are in the same coordinate system, the difficulty of observation by a doctor is reduced, the two-dimensional images have limitations, the doctor needs to combine a plurality of two-dimensional images to imagine the specific form of the blood vessel, and the integral state of the intracranial blood vessel cannot be obtained simply, quickly and intuitively in clinic. According to the embodiment of the invention, by utilizing the maximum intensity projection method to obtain the MIP (maximum intensity projection) images in all directions of the registered bright blood images, and utilizing the characteristic that the intracranial blood vessel bright blood sequence MIP images have similarity with the fundus blood vessel images, on one hand, a network model for fundus blood vessel segmentation is trained by utilizing the labeled sample of the fundus blood vessel images, on the other hand, the characteristic transformation is carried out on the intracranial blood vessel bright blood sequence MIP images to obtain the characteristic MIP images with the same sample distribution as the fundus blood vessel images, and the network model pre-trained in the fundus blood vessel segmentation task is migrated into the intracranial blood vessel segmentation task by adopting a characteristic migration mode to obtain the two-dimensional blood vessel segmentation images in all directions corresponding to the intracranial blood vessel bright blood sequence MIP images. The embodiment of the invention applies the transfer learning to the field of the segmentation of intracranial blood vessels, and can obtain more accurate blood vessel segmentation effect. And then, obtaining first three-dimensional blood vessel volume data by using a back projection method, and realizing an intracranial blood vessel simulation three-dimensional model by using second three-dimensional blood vessel volume data corresponding to the bright blood image group after registration. Aiming at each segment of blood vessel in the intracranial blood vessel simulation three-dimensional model, obtaining a numerical value of a target parameter representing the stenosis degree of the segment of blood vessel; and finally, marking the intracranial blood vessel simulation three-dimensional model by using the numerical value of the target parameter of each section of blood vessel to obtain the simulated three-dimensional intracranial vascular stenosis analysis model. The simulated three-dimensional intracranial vascular stenosis analysis model can provide visual intracranial vascular three-dimensional space information without the need of reducing vascular tissue structure, disease characteristics and the like through imagination of doctors, and is convenient for visually observing, positioning and displaying a narrow focus region. The analytical data about the intracranial vascular stenosis degree can be obtained clinically intuitively and quickly.

Note: the patient experimental data in the embodiment of the invention are all from people hospitals in Shaanxi province, and the images can be used for general scientific research.

The above description is only for the preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for establishing a three-dimensional corrosion narrowing model of intracranial vascular simulation based on migration learning is characterized by comprising the following steps:

acquiring a bright blood image group and an enhanced black blood image group of intracranial blood vessels; wherein the bright blood image group and the enhanced black blood image group respectively include K bright blood images and K enhanced black blood images; the images in the bright blood image group and the enhanced black blood image group correspond to each other one by one; k is a natural number greater than 2;

taking each bright blood image and the corresponding enhanced black blood image as an image pair, and preprocessing each image pair to obtain a first bright blood image and a first black blood image of the image pair;

aiming at each first bright blood image, taking the corresponding first black blood image as a reference, and performing image registration by using a mutual information based on Gaussian distribution sampling and image pyramid registration method to obtain a registered bright blood image group comprising K registered bright blood images;

projecting the registered bright blood image group in three preset directions by using a maximum intensity projection method to obtain MIP (maximum intensity projection) images in all directions, taking the MIP images in all directions as a target domain, taking the fundus blood vessel image as a source domain, and obtaining two-dimensional blood vessel segmentation images corresponding to the MIP images in all directions by using a migration learning method;

synthesizing the two-dimensional vessel segmentation maps in the three directions by using a back projection method to obtain first three-dimensional vessel volume data; wherein the voxel value of the blood vessel part in the first three-dimensional blood vessel volume data is 0, and the voxel value of the non-blood vessel part is minus infinity;

obtaining an intracranial blood vessel simulation three-dimensional model based on the first three-dimensional blood vessel volume data and second three-dimensional blood vessel volume data corresponding to the registered bright blood image group;

aiming at each segment of blood vessel in the intracranial blood vessel simulation three-dimensional model, obtaining a numerical value of a target parameter representing the stenosis degree of the segment of blood vessel;

and marking the intracranial blood vessel simulation three-dimensional model by using the numerical value of the target parameter of each section of blood vessel to obtain a simulated three-dimensional intracranial vascular stenosis analysis model.

2. The method of claim 1, wherein preprocessing each image pair to obtain a first bright blood image and a first black blood image of the image pair comprises:

for each image pair, taking the enhanced black blood image as a reference, performing coordinate transformation and image interpolation on the bright blood image, using similarity measurement based on mutual information, and adopting a preset search strategy to obtain a pre-registered first bright blood image;

and extracting the same area content as the scanning range of the first bright blood image from the enhanced black blood image to form a first black blood image.

3. The method according to claim 1 or 2, wherein the performing, for each first bright blood image, image registration by using a mutual information based on gaussian distribution sampling and image pyramid registration method with reference to the corresponding first black blood image to obtain a registered bright blood image group including K registered bright blood images comprises:

adopting Gaussian distribution sampling to select a part of preprocessed image pair as a test image pair;

performing image registration on the first bright blood image and the first black blood image in each test image pair by adopting a registration method based on mutual information and an image pyramid to obtain a rotation matrix corresponding to the first bright blood image in the test image pair after registration;

obtaining the mean value of the rotation matrix of all the test image pairs;

and performing coordinate transformation on the first bright blood image in the other preprocessed image pairs except the test image pair by using the mean value of the rotation matrix to complete image registration to obtain a registered bright blood image group comprising K registered bright blood images.

4. The method according to claim 3, wherein the performing image registration on the first bright blood image and the first black blood image in each test image pair by using a registration method based on mutual information and an image pyramid to obtain a rotation matrix corresponding to the first bright blood image in the test image pair after registration comprises:

aiming at each test image pair, based on down-sampling processing, obtaining a bright blood Gaussian pyramid from the first bright blood image, and obtaining a black blood Gaussian pyramid from the first black blood image; the bright blood Gaussian pyramid and the black blood Gaussian pyramid comprise m images with resolution becoming smaller in sequence from bottom to top; m is a natural number greater than 3;

based on the upsampling processing, obtaining a bright blood Laplacian pyramid by using the bright blood Gaussian pyramid, and obtaining a black blood Laplacian pyramid by using the black blood Gaussian pyramid; the bright blood Laplacian pyramid and the black blood Laplacian pyramid comprise m-1 images with resolution which is sequentially reduced from bottom to top;

registering images of corresponding layers in the bright blood Laplacian pyramid and the black blood Laplacian pyramid to obtain a registered bright blood Laplacian pyramid;

and registering the images of all layers in the bright blood Gaussian pyramid and the black blood Gaussian pyramid from top to bottom by using the registered bright blood Laplacian pyramid as superposition information to obtain a registered bright blood Gaussian pyramid, and obtaining a rotation matrix corresponding to the first bright blood image in the registered test image pair.

5. The method according to claim 4, wherein the registering the images of the respective layers in the blood-brightening Gaussian pyramid and the black blood Gaussian pyramid from top to bottom by using the registered blood-brightening Gaussian pyramid as overlay information to obtain the registered blood-brightening Gaussian pyramid comprises:

performing upsampling operation on the registered jth layer of bright blood Gaussian image, adding the upsampled operation to the registered corresponding layer of bright blood Laplacian image, and replacing the jth +1 layer of bright blood Gaussian image in the bright blood Gaussian pyramid by using the added image;

wherein j is 1, 2, …, m-1, the black blood gaussian image is an image in the black blood gaussian pyramid, and the bright blood gaussian image is an image in the bright blood gaussian pyramid.

6. The method according to claim 1, wherein the projecting the registered bright blood image group in three preset directions by using a maximum intensity projection method to obtain MIP maps in all directions, and obtaining two-dimensional vessel segmentation maps corresponding to the MIP maps in all directions by using a migration learning method with the MIP maps in all directions as a target domain and a fundus blood vessel map as a source domain, comprises:

obtaining a pre-trained target neural network aiming at the eye fundus blood vessel map segmentation task; the target neural network is obtained by pre-training according to the fundus blood vessel map data set and the improved U-net network model;

respectively carrying out gray level inversion processing and contrast enhancement processing on the MIP images in all directions to obtain corresponding characteristic MIP images; wherein the characteristic MIP map has the same sample distribution as the fundus blood vessel map;

and respectively inputting the characteristic MIP maps of all directions into the target neural network to obtain corresponding two-dimensional vessel segmentation maps.

7. The method according to claim 6, wherein the performing gray-scale inversion processing and contrast enhancement processing on the MIP-maps in each direction respectively to obtain corresponding characteristic MIP-maps comprises:

carrying out pixel transformation on the MIP graph by utilizing a gray inversion formula to obtain an inversion graph; wherein, the grayscale inversion formula is t (x) 255-x, x is a pixel value in the MIP map, and t (x) is a pixel value in the inversion map;

and enhancing the contrast of the inversion graph by using a contrast-limited self-adaptive histogram equalization method to obtain a characteristic MIP (maximum intensity distribution) graph.

8. The method according to claim 1 or 7, wherein obtaining an intracranial vascular simulation three-dimensional model based on the first three-dimensional vascular volume data and the second three-dimensional vascular volume data corresponding to the post-registration bright blood image group comprises:

adding the first three-dimensional blood vessel volume data and the second three-dimensional blood vessel volume data to obtain third three-dimensional blood vessel volume data;

and processing the third three-dimensional blood vessel volume data by using a threshold segmentation method to obtain an intracranial blood vessel simulation three-dimensional model.

9. The method according to claim 1, wherein the simulating for each segment of the blood vessel in the three-dimensional model of the intracranial blood vessel, obtaining a value of a target parameter characterizing a stenosis degree of the segment of the blood vessel, comprises:

cutting each section of blood vessel in the intracranial blood vessel simulation three-dimensional model from three preset directions to obtain a two-dimensional sectional view of each direction;

carrying out corrosion operation on the blood vessel in the two-dimensional sectional diagram of each direction, and recording the target corrosion times when the blood vessel is corroded to a single pixel;

and obtaining a numerical value of a target parameter representing the stenosis degree of the section of the blood vessel according to the target corrosion times of the section of the blood vessel in the three directions respectively.

10. The method according to claim 1, wherein said labeling said three-dimensional model of intracranial vascular simulation with values of said target parameter for each segment of blood vessel, resulting in a simulated three-dimensional intracranial vascular stenosis analysis model, comprises:

and marking the intracranial blood vessel simulation three-dimensional model by using the numerical values of the target parameters of all the sections of blood vessels and adopting colors corresponding to all the numerical values to obtain a simulated three-dimensional intracranial vascular stenosis analysis model.