WO2022105647A1 - 一种颅内血管造影增强三维模型的建立方法 - Google Patents

一种颅内血管造影增强三维模型的建立方法 Download PDF

Info

Publication number
WO2022105647A1
WO2022105647A1 PCT/CN2021/129608 CN2021129608W WO2022105647A1 WO 2022105647 A1 WO2022105647 A1 WO 2022105647A1 CN 2021129608 W CN2021129608 W CN 2021129608W WO 2022105647 A1 WO2022105647 A1 WO 2022105647A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
blood
bright
enhanced
black
Prior art date
Application number
PCT/CN2021/129608
Other languages
English (en)
French (fr)
Inventor
贾艳楠
王文杰
Original Assignee
西安科锐盛创新科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 西安科锐盛创新科技有限公司 filed Critical 西安科锐盛创新科技有限公司
Priority to US17/552,282 priority Critical patent/US12112489B2/en
Publication of WO2022105647A1 publication Critical patent/WO2022105647A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G06T11/003—Reconstruction from projections, e.g. tomography
    • G06T11/008—Specific post-processing after tomographic reconstruction, e.g. voxelisation, metal artifact correction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/20—Image enhancement or restoration using local operators
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
    • G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
    • G06T7/344—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods involving models
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20024—Filtering details
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30101—Blood vessel; Artery; Vein; Vascular
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2211/00—Image generation
    • G06T2211/40—Computed tomography
    • G06T2211/424—Iterative

Definitions

  • the invention belongs to the field of image processing, and in particular relates to a method for establishing an enhanced three-dimensional model of intracranial angiography.
  • the clinical evaluation of the degree of intracranial vascular lesions and the degree of vascular stenosis usually uses methods based on lumen imaging, such as Digital Subtraction Angiography (DSA), Computed Tomography Angiography (CTA) ), Magnetic Resonance Angiography (MRA), and High-Resolution Magnetic Resonance Angiography (HRMRA).
  • DSA Digital Subtraction Angiography
  • CTA Computed Tomography Angiography
  • MRA Magnetic Resonance Angiography
  • HRMRA High-Resolution Magnetic Resonance Angiography
  • the intracranial arteries are connected to the carotid and vertebral arteries, and form a ring-shaped structure at the base of the brain. Through magnetic resonance angiography, the path of intracranial arteries can be clearly delineated.
  • MRA magnetic resonance angiography
  • HRMRA magnetic resonance angiography
  • the images corresponding to the bright blood sequence and the black blood sequence obtained by magnetic resonance angiography are two-dimensional images
  • doctors need to combine the information of the two images to obtain the comprehensive situation of the intracranial blood vessels, so as to carry out Analysis of intracranial vascular lesions.
  • two-dimensional images have limitations, which are not conducive to obtaining real information of intracranial blood vessels easily and quickly.
  • the embodiment of the present invention provides a method for establishing an enhanced three-dimensional model of intracranial angiography.
  • the bright blood image group, the black blood image group, and the enhanced black blood image group respectively include K bright blood images Blood image, black blood image and enhanced black blood image; the images in the bright blood image group, the black blood image group, and the enhanced black blood image group are in one-to-one correspondence; K is a natural number greater than 2;
  • each image pair is preprocessed to obtain the first bright blood image and the first black blood image of the image pair;
  • each first bright blood image take the corresponding first black blood image as the benchmark, use the mutual information and image pyramid registration method based on Gaussian distribution sampling to perform image registration, and obtain K registered bright blood images.
  • the registered bright blood image group ;
  • the enhanced black blood image in the enhanced black blood image group is subjected to a flow void artifact removal operation to obtain an artifact-removed enhanced black blood image including K target enhanced black blood images Group;
  • an intracranial angiography-enhanced three-dimensional model is obtained.
  • the bright blood image and the enhanced black blood image scanned by the magnetic resonance angiography technique are firstly registered by the mutual information and image pyramid registration method based on Gaussian distribution sampling, which can improve the The registration efficiency enables the image to improve the registration accuracy layer by layer from low resolution to high resolution.
  • the bright blood image and the enhanced black blood image can be unified in the same coordinate system.
  • using the registered bright blood image to eliminate the flow-space artifacts on the enhanced black blood image it can display more accurate and comprehensive blood vessel information.
  • the solution provided by the embodiment of the present invention is to eliminate the flow void artifact from the perspective of image post-processing, without using new imaging technology, imaging mode or pulse sequence, so the flow void artifact can be eliminated simply, accurately and quickly, and the It can achieve better promotion in clinical application.
  • a three-dimensional model of blood and a three-dimensional model of blood vessels with expanded blood boundary are established by using the registered bright blood image, and a contrast-enhanced three-dimensional model with contrast-enhanced effect is obtained by subtracting the enhanced black blood image and the black blood image by eliminating the artifacts.
  • the blood three-dimensional model, the blood vessel three-dimensional model, and the contrast-enhanced three-dimensional model are used to obtain the intracranial angiography-enhanced three-dimensional model corresponding to the blood vessel wall with contrast-enhancing effect.
  • the intracranial angiography-enhanced three-dimensional model realizes the three-dimensional visualization of intracranial blood vessels, without the need for doctors to restore the structure and disease characteristics of intracranial blood vessels through imagination, and it is convenient for doctors to observe and analyze the morphological characteristics of blood vessels from any angle and level of interest. , which can provide realistic three-dimensional space information of blood vessels, which is convenient for intuitively displaying the blood vessel wall with obvious contrast enhancement, and is convenient for locating and displaying the lesion area. In clinical application, the real information of intracranial blood vessels can be easily and quickly obtained to analyze vascular lesions.
  • FIG. 1 is a schematic flowchart of a method for establishing an enhanced three-dimensional model of intracranial angiography according to an embodiment of the present invention
  • FIG. 2 is a schematic diagram of coordinate transformation of an intracranial blood vessel magnetic resonance image according to an embodiment of the present invention
  • FIG. 4 is a result diagram of an intracranial blood vessel magnetic resonance image after pre-registration according to an embodiment of the present invention
  • FIG. 5 is a schematic diagram of a region to be registered in an intracranial blood vessel magnetic resonance image according to an embodiment of the present invention
  • FIG. 6( a ) is a bright blood Gaussian pyramid and a black blood Gaussian pyramid of an intracranial blood vessel magnetic resonance image according to an embodiment of the present invention
  • FIG. 6( b ) is a bright blood Rapp of the intracranial blood vessel magnetic resonance image according to an embodiment of the present invention Pyramid of Las Vegas and Pyramid of Black Blood Laplace;
  • FIG. 8 is a schematic diagram of steps of Gaussian pyramid image registration based on mutual information of intracranial blood vessel magnetic resonance images according to an embodiment of the present invention
  • Fig. 10 is the registration result of the intracranial blood vessel magnetic resonance image of multiple registration methods including the mutual information pyramid method;
  • FIG. 11 is a registration result of an intracranial blood vessel magnetic resonance image based on a Gaussian distribution sampling based mutual information and image pyramid registration method and a mutual information pyramid method according to an embodiment of the present invention
  • FIG. 13 is a result diagram of a grayscale linear transformation according to an embodiment of the present invention.
  • FIG. 14 is an image binarization result diagram according to an embodiment of the present invention.
  • FIG. 15 is a result of eliminating flow void artifacts obtained by different methods for intracranial blood vessels in an embodiment of the present invention.
  • FIG. 16 are respectively an effect diagram of a blood 3D model, an effect diagram of a blood vessel 3D model, and a contrast-enhanced 3D model effect diagram of an intracranial blood vessel according to an embodiment of the present invention
  • FIG. 17 is an effect diagram of an enhanced three-dimensional model of intracranial angiography according to an embodiment of the present invention.
  • the embodiment of the present invention provides a method for establishing an enhanced three-dimensional model of intracranial angiography.
  • the execution subject of the method for establishing an enhanced three-dimensional model of intracranial angiography may be a device for establishing a three-dimensional model of enhanced intracranial angiography, and the device may run in an electronic device .
  • the electronic device may be a blood vessel imaging device, or an image processing device, of course, it is not limited to this.
  • a method for establishing an enhanced three-dimensional model for intracranial angiography may include the following steps:
  • the bright blood image group, black blood image group, and enhanced black blood image group respectively include K bright blood images, black blood images, and enhanced black blood images; One-to-one correspondence of the images; K is a natural number greater than 2;
  • the bright blood image group is an image group obtained by scanning the intracranial blood vessels with the bright blood sequence using magnetic resonance angiography.
  • the black blood image group is an image group obtained by scanning the intracranial blood vessels using magnetic resonance angiography.
  • the enhanced black blood image group is an image group obtained by first injecting paramagnetic contrast agent into the patient, and then using magnetic resonance angiography to scan the intracranial blood vessel.
  • the magnetic resonance angiography technology is preferably HRMRA.
  • the K images in the bright blood image group, the black blood image group, and the enhanced black blood image group are in one-to-one correspondence, and the corresponding manner is that the order of the images formed according to the scanning time is the same.
  • each image pair is preprocessed to obtain the first bright blood image and the first black blood image of the image pair, which may include S21 and S22:
  • the enhanced black blood image is scanned according to the coronal plane, while the bright blood image is scanned according to the axial plane.
  • MRI images of different imaging slices are observed in a standard reference frame.
  • the coordinate transformation of the image can be realized by using the orientation information in the medical image DICOM (Digital Imaging and Communications in Medicine) file.
  • DICOM Digital Imaging and Communications in Medicine
  • a DICOM file is an image storage format for medical equipment such as CT or MRI.
  • the DICOM3.0 format image file contains azimuth label information related to the imaging direction. This information briefly introduces the azimuth relationship between the patient and the imaging instrument.
  • the exact position of each pixel in the image can be known.
  • the enhanced black blood image and the bright blood image are images to be registered, and the enhanced black blood image can be used as a reference based on the coordinate system of the enhanced black blood image according to the orientation label information in the DICOM file of the bright blood image.
  • Image take the bright blood image as a floating image, and perform coordinate transformation on the bright blood image to achieve the purpose of rotating the bright blood image to the same coordinate system as the enhanced black blood image. After the rotation, the scanning direction of the bright blood image also becomes the coronal plane.
  • each coordinate position in the image A corresponds to the image B through a mapping relationship.
  • Specific coordinate transformation methods may include rigid body transformation, affine transformation, projection transformation, and nonlinear transformation. Since the intracranial blood vessel can be regarded as a rigid body, the embodiment of the present invention selects rigid body transformation as the coordinate transformation method.
  • the coordinate system of the floating image will be stretched or deformed, and the pixel coordinates of the image after coordinate transformation will not completely coincide with the sampling grid of the original image, that is, the pixel coordinates that were originally integers pass through the coordinates. After transformation, it may no longer be an integer, causing some areas of the image to lose some pixels. Therefore, in the process of image coordinate transformation, it is necessary to resample and interpolate the image at the same time to determine the gray value of the pixel coordinate point of the image after coordinate transformation. , which is convenient for subsequent processing.
  • the coordinates of the bright blood image after coordinate transformation may be mapped to the non-integer coordinates of the original image, so it is necessary to perform image interpolation on the bright blood image at the same time.
  • Image interpolation methods include nearest neighbor interpolation, bilinear interpolation, and bicubic interpolation.
  • experiments are carried out on three interpolation methods, and a total of 5 evaluation indicators are set, which are the root mean square error RMSE, the peak signal-to-noise ratio PSNR, the normalized cross-correlation coefficient NCC, and the normalized mutual information NMI.
  • the scale to measure the feature similarity between two images is the similarity measure. Selecting the appropriate similarity measure can improve the registration accuracy, effectively suppress noise, etc. It plays a very important role in image registration.
  • Commonly used similarity measures are mainly divided into three categories, namely distance measure, correlation measure and information entropy.
  • distance measure a measure of the distance measure
  • correlation measure a measure of the correlation measure
  • information entropy a measure of the correlation measure
  • intracranial blood vessels it can be regarded as a rigid body and hardly deforms. Unlike organs such as the heart or the lungs, which change with movements such as human respiration, the intracranial blood vessels are targeted for Mutual information or normalized mutual information can be selected as the similarity measure to make the registration effect more accurate.
  • MI Mutual Information
  • NMI Normalization Mutual Information
  • the concept of normalized mutual information solves the problem that when the overlapping part of the two images is small or most of the overlapping area is background information, the image registration based on mutual information has low accuracy and poor registration effect.
  • the sensitivity of mutual information to the overlapping area of the image is calculated.
  • Image registration is essentially a multi-parameter optimization problem, that is, by using a certain search strategy to change the spatial coordinates of the image, the similarity measure of the two images is finally optimized.
  • the search strategy and the spatial coordinate change are calculated in actual.
  • the processes are intertwined with each other.
  • the idea of the algorithm is to calculate the similarity measure between the two images in each iteration, adjust the floating image through coordinate transformation operations such as translation or rotation, and at the same time interpolate the image until the similarity measure of the two images is the largest.
  • search strategies include gradient descent optimizer, (1+1)-ES based on evolution strategy (Evolution Strategy, ES), etc.
  • the predetermined search strategy in the embodiment of the present invention can be selected as required.
  • FIG. 2 is a schematic diagram of coordinate transformation of an intracranial blood vessel magnetic resonance image according to an embodiment of the present invention, wherein the first row is the enhanced black blood image and the bright blood image, and the second row is the From the enhanced black blood image and the bright blood image after coordinate transformation, it can be seen that after the coordinate transformation, the scanning direction of the bright blood image and the enhanced black blood image are consistent, and both are coronal planes.
  • the gradient descent optimizer and (1+1)-ES two search strategies are used to register 160 bright blood images and 160 enhanced black blood images at the corresponding scanning level, where the enhanced black blood images are the reference images,
  • the bright blood image is a floating image
  • the registration result is displayed as shown in Figure 3, which is the registration comparison result of the two search strategies according to the embodiment of the present invention;
  • the left image in Figure 3 is two images registered without an optimizer
  • the results are displayed in pairs.
  • the middle picture shows the paired display results of images registered with the gradient descent optimizer, and the right picture shows the paired display results of images registered with the (1+1)-ES optimizer.
  • the right image shows a montage effect, using pseudo-color transparency processing to enhance the black blood image and the bright blood image, the purple is the enhanced black blood image, and the green is the bright blood image (the attached image is the grayscale processed image of the original image, the color is not shown out).
  • the enhanced black blood image and the bright blood image do not overlap, and there are many shadows; when the gradient descent optimizer is used to register the image, although it is more The registration effect is good, but there is still obvious misalignment in the gray matter of the brain; while 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 the three evaluation indicators of the registration results, which are the normalized mutual information NMI, the normalized cross-correlation coefficient NCC and the algorithm time-consuming Time. From the experimental results, the registration image effect of (1+1)-ES is clearer and better than that of the gradient descent optimizer; from the experimental data, the three evaluation indicators all show (1+1)-ES The optimizer has good performance. Therefore, in this embodiment of the present invention, preferably, the predetermined search strategy is (1+1)-ES.
  • the value in a is the mean ⁇ mean square error of the evaluation index based on the registration of 160 bright blood images and 160 enhanced black blood images
  • FIG. 4 is a result diagram of an intracranial blood vessel magnetic resonance image after pre-registration according to an embodiment of the present invention.
  • the left picture is the first bright blood image after pre-registration, in which the interpolation method adopts bicubic interpolation;
  • the middle picture is the enhanced black blood image, it can be seen that both are coronal planes, and
  • the right picture is the effect picture after the two are directly superimposed, It can be seen from the image on the right that although pre-registration has been carried out, the bright blood image and the enhanced black blood image under the current imaging layer can be observed under the same coronal plane, but the two still do not overlap, so follow-up image refinement is required. registration.
  • this step it is possible to initially compare the magnetic resonance images of the same scan level in the same coordinate system, but due to the different scanning times of the bright blood sequence and the black blood sequence, and the patient may have a slight movement before and after the scan , so the above operation is just a rough coordinate transformation, and the complete registration of multi-modal magnetic resonance images cannot be achieved only through pre-registration, but this step can omit unnecessary processing for the subsequent precise registration process, improve processing speed.
  • S22 may include the following steps:
  • edges contour information includes the coordinate values of each edge point.
  • the edge contour information extract the minimum abscissa value, the maximum abscissa value, the minimum ordinate value, and the maximum ordinate value, and use these four coordinate values to determine the four vertices of the square frame, thereby obtaining the initial extraction frame;
  • the four directions are the positive and negative directions of the horizontal and vertical coordinates respectively;
  • the preset number is reasonably selected according to the type of blood vessel image, and the purpose is to ensure that the enlarged final extraction frame does not exceed the size range of the first bright blood image, such as the preset number Can be 20 and so on.
  • the content of the corresponding area in the enhanced black blood image is extracted, and the extracted content is formed into a first black blood image.
  • the common scanning range of the magnetic resonance images in the two modalities is obtained by extracting the region to be registered, which is beneficial to the subsequent rapid registration.
  • FIG. 5 is a schematic diagram of an area to be registered in an intracranial blood vessel magnetic resonance image according to an embodiment of the present invention, wherein the left image is the first bright blood image after pre-registration, the right image is the enhanced black blood image, and the box To enhance the area to be extracted in the black blood image.
  • This area includes the common scanning range of the bright blood sequence and the black blood sequence in the intracranial blood vessel magnetic resonance image. By determining the area to be extracted, useful information can be focused more quickly.
  • the above-mentioned two-step preprocessing process in the embodiment of the present invention plays a very important role.
  • the preprocessed image can pay more attention to useful information and exclude irrelevant information.
  • using this image preprocessing can improve the intracranial Reliability of blood vessel image registration and recognition.
  • S3 may include S31 to S34:
  • the test image pair in the embodiment of the present invention is the image pair to be registered, and the random selection of the image pair to be registered in the embodiment of the present invention adopts Gaussian distribution sampling. This is because the scanning directions of the bright blood image and the enhanced black blood image are different.
  • the bright blood image has undergone coordinate transformation and interpolation, so that the Each bright blood image corresponds to the enhanced black blood image of the current layer; at the same time, since the scanning range of the bright blood image and the enhanced black blood image are different, the data of the edge layer of the bright blood image may be incomplete.
  • the Gaussian mean ⁇ is selected as half of the total number of images to be registered, so that the Gaussian randomly selects the middle layer image for registration with the highest probability.
  • S32 for the first bright blood image and the first black blood image in each test image pair, image registration is performed using a registration method based on mutual information and image pyramids, and the first bright blood image in the test image pair is obtained after registration.
  • the rotation matrix corresponding to the blood image; in an optional implementation manner, S32 may specifically include steps S321-S324:
  • a multi-resolution strategy can be used to solve the problem of local extremum.
  • the multi-resolution strategy can meet the conditions of image registration accuracy. It can improve the execution speed of the algorithm and increase the robustness. Building an image pyramid is an effective way to improve the registration accuracy and speed by increasing the complexity of the model, that is, in the registration process, in the order from coarse registration to fine registration, the low-resolution images are first processed. Registration, and then register the high-resolution images based on the registration of the low-resolution images.
  • step S321 may include:
  • the input image of the i-th layer filter the input image of the i-th layer with a Gaussian kernel, and delete the even-numbered rows and even-numbered columns of the filtered image to obtain the image G i of the i-th layer of the Gaussian pyramid.
  • the multiple images in the Gaussian pyramid are images corresponding to different resolutions of the same original image.
  • the Gaussian pyramid obtains images through Gaussian filtering and downsampling.
  • Each layer of its construction steps can be divided into two steps: first, use Gaussian filtering to smooth the image, that is, use a Gaussian kernel for filtering; then delete the even lines of the filtered image.
  • even-numbered columns that is, reducing the width and height of the lower layer image by half to obtain the current layer image, so the current layer image is a quarter of the size of the lower layer image, and by continuously iterating the above steps, the Gaussian pyramid can be finally obtained.
  • Gaussian filtering is actually a low-pass filter.
  • the image frequency range in the Gaussian pyramid is very wide, and the cutoff frequency of the lower layer image is twice the cutoff frequency of the upper layer image.
  • Gaussian filtering first uses a Gaussian function to calculate a weight matrix, and then uses the weight matrix to perform a convolution operation on the original image.
  • a two-dimensional Gaussian template can be used for the above processing.
  • a two-dimensional Gaussian template can be used to achieve the effect of blurring an image, when a point is on the border and there are not enough points around it, the edge image will be missing due to the relationship of the weight matrix. optimization.
  • the two-dimensional Gaussian filter can be split into two independent one-dimensional Gaussian filters, and image filtering is performed in the horizontal and vertical directions respectively. Separating the Gaussian function can not only eliminate the edges generated by the two-dimensional Gaussian template, but also greatly speed up the running speed of the program. Compared with other blur filters, Gaussian filtering can not only achieve the blurring effect of the image, but also better preserve the marginal effect.
  • FIG. 6( a ) is a bright blood Gaussian pyramid and a black blood Gaussian pyramid of an intracranial blood vessel magnetic resonance image according to an embodiment of the present invention.
  • the Si pyramid includes m-1 images with decreasing resolution from bottom to top;
  • the Gaussian pyramid is down-sampling, that is, reducing the image, part of the image data will be lost. Therefore, in the embodiment of the present invention, in order to avoid data loss during the zooming process of the image and restore the detail data, the Laplacian pyramid is used together with the Gaussian pyramid to realize image reconstruction, and the details are highlighted on the basis of the Gaussian pyramid image.
  • step S322 may include:
  • the Laplacian pyramid is the residual between the image and the original image after the downsampling operation, the Laplacian pyramid is one layer less high-level image than the Gaussian pyramid structure compared from the bottom to the top.
  • the mathematical formula for generating the Laplacian pyramid structure is shown in (1), where Li represents the i -th layer of Laplacian pyramids (Bright Blood Laplacian Pyramid or Black Blood Laplacian Pyramid), G i represents the i-th Gaussian pyramid (bright blood Gaussian pyramid or black blood Gaussian pyramid), and the UP operation is up-sampling to enlarge the image, the symbol is the convolution symbol, is the Gaussian kernel used in building the Gaussian pyramid.
  • This formula shows that the Laplacian pyramid is essentially composed of the original image minus the residual data of the image that is first reduced and then enlarged. It is a residual prediction pyramid.
  • the core idea is to store the image after downsampling.
  • the high-frequency information of the image is preserved, in order to completely restore the image before the downsampling operation at each level. Since part of the information lost in the previous downsampling operation cannot be completely recovered by upsampling, that is, downsampling is irreversible, so the image is first downsampled and then upsampled, and the display effect is blurrier than the original image.
  • the residual between the image and the original image after the downsampling operation details can be added to the images of different frequency layers on the basis of the Gaussian pyramid image, and the details can be highlighted.
  • this step can obtain the bright blood Laplacian pyramid and the black blood Laplacian pyramid with 3 image layers.
  • FIG. 6( b ) is a bright blood Laplacian pyramid and a black blood Laplacian pyramid of an intracranial blood vessel magnetic resonance image according to an embodiment of the present invention.
  • the image shows the use of gamma correction to achieve a sharper effect, with a gamma value of 0.5.
  • step S323 register the images of the corresponding layers in the bright blood Laplacian pyramid and the black blood Laplacian pyramid to obtain the registered bright blood Laplacian pyramid; in an optional embodiment, step S323 Can include:
  • the black blood Laplacian image corresponding to the layer is used as a reference image
  • the bright blood Laplacian image corresponding to the layer is used as a reference image.
  • the similarity measure based on mutual information is used, and a predetermined search strategy is adopted to realize image registration, and obtain the registered bright blood Laplacian image of this layer;
  • the registered bright blood Laplacian pyramid is formed from bottom to top; among them, the black blood Laplacian image is black.
  • the image in the blood Laplacian pyramid, the bright blood Laplacian image is the image in the bright blood Laplacian pyramid.
  • the registration process in this step is similar to the above-mentioned pre-registration process.
  • image registration can be achieved, and the registration can be obtained.
  • Post Bright Blood Laplace image Among them, coordinate transformation, image interpolation, similarity measurement and predetermined search strategy will not be repeated.
  • FIG. 7 is the registration result of the Laplacian pyramid image of the intracranial blood vessel magnetic resonance image according to the embodiment of the present invention.
  • the left image is the reference image in the black blood Laplacian pyramid, and the middle image It is a registered image in the Laplacian Pyramid of Bright Blood.
  • the right image is the effect of the left and middle images directly superimposed.
  • the superimposed image is displayed using a montage effect, and the black blood image and the bright image are enhanced by pseudo-color transparency processing.
  • Blood image in which purple is the enhanced black blood Laplacian pyramid image, and green is the bright blood Laplacian pyramid image (the attached image is the grayscale-processed image of the original image, the color is not shown).
  • the embodiment of the present invention is used in low-resolution images.
  • high-resolution images are registered, that is, Gaussian pyramid images are registered from top to bottom, and the registration result of the previous layer image is used as the input of the next layer image registration.
  • the registered bright blood Laplacian pyramid is used as the overlay information to perform top-down registration on the images of each layer in the bright blood Gaussian pyramid and the black blood Gaussian pyramid to obtain:
  • the registered bright blood Gaussian pyramid which can include:
  • the black blood Gaussian image corresponding to this layer is used as the reference image, and the bright blood Gaussian image corresponding to this layer is used as the floating image.
  • the similarity measure of and using a predetermined search strategy to achieve image registration, and obtain the registered jth layer of bright blood Gaussian image;
  • the image registration is realized, and the matching method is obtained.
  • the bright blood Gaussian image of the j+1th layer after calibration; where j 1, 2, ..., m-1, the black blood Gaussian image is the image in the black blood Gaussian pyramid, and the bright blood Gaussian image is the image in the bright blood Gaussian pyramid. image.
  • the above operations are repeated until the high-resolution registration of the underlying Gaussian pyramid image is completed, and the registered bright blood Gaussian pyramid is obtained.
  • the registration process it is similar to the pre-registration process described above. Among them, coordinate transformation, image interpolation, similarity measurement and predetermined search strategy will not be repeated.
  • FIG. 8 is a schematic diagram of the mutual information-based Gaussian pyramid image registration steps of an intracranial blood vessel magnetic resonance image according to an embodiment of the present invention.
  • the registration based on mutual information is performed on the low-resolution black blood Gaussian image of the top layer and the low-resolution bright blood Gaussian image of the top layer; high-frequency information, and the bright blood Laplacian images of the corresponding layers that have been registered according to the above operations are added to serve as the next layer of bright blood Gaussian images; then the bright blood Gaussian images obtained by the above operations are used as input images, and then Register with the black blood Gaussian image of the corresponding layer, and repeat the above operation until the high-resolution registration of the underlying Gaussian pyramid image is completed.
  • the Gaussian pyramid image registration based on mutual information, it is necessary to register the bright blood Gaussian image and the black blood Gaussian image of each layer with the normalized mutual information as the similarity measure, and calculate the NMI of the two images through loop iteration. until the NMI reaches its maximum value.
  • the number of iterations is too small, the accurate registration of images cannot be completed, but when the number of iterations is too large, the amount of calculation will increase sharply.
  • One layer of images that is, the bottom image with the highest resolution in the Gaussian pyramid, stops iterating when the registration reaches the maximum NMI value and the data is stable.
  • a registered bright blood Gaussian pyramid is obtained, in which 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, and the blood vessel image registration process according to the embodiment of the present invention can be completed. . Moreover, after registration, a rotation matrix corresponding to the first bright blood image in the pair of test images after registration can be obtained.
  • the mutual information pyramid method In order to verify the validity and practicability of the registration method based on mutual information and image pyramid (referred to as the mutual information pyramid method) in the embodiment of the present invention, a comparative experiment was also carried out, and a total of five patients' intracranial vascular magnetic resonance images were used.
  • Resonance images including 160 enhanced black blood images and bright blood images of patients A, B, C, and D, respectively, and 150 enhanced black blood images and bright blood images of patient E; at the same time, only use the DICOM image orientation label
  • the algorithm for information registration, and the registration algorithm based on mutual information measurement are compared with the mutual information pyramid method in the embodiment of the present invention, wherein the algorithm based on mutual information measurement is to find reference images and floating images through a multi-parameter optimization method The optimal transformation between the two images maximizes the mutual information value of the two images, and the image pyramid algorithm is not used.
  • the experimental platform is Matlab R2016b.
  • qualitative analysis and quantitative analysis are combined.
  • qualitative analysis due to the large grayscale difference between the multimodal medical images, the difference image obtained by subtracting the registered image from the reference image cannot effectively reflect the registration result of the multimodal medical image.
  • a color overlapping image that can reflect the alignment degree of the registration image and the reference image is obtained, and the registration effect of the multi-modal registration algorithm is qualitatively analyzed through the color overlapping image.
  • FIG. 10 shows the registration result of the multimodal intracranial blood vessel magnetic resonance image
  • FIG. 10 is the registration result of the intracranial blood vessel magnetic resonance image of multiple registration methods including the mutual information pyramid method.
  • the value in a is the mean ⁇ mean square error of the evaluation index based on the registration of multiple images of the patient
  • Quantitative analysis As can be seen from Table 2, from the two evaluation indicators of NCC and NMI, compared with the registration algorithm that only uses the orientation label information of the DICOM image and the registration algorithm based on the mutual information measurement, the embodiment of the present invention The registration accuracy of the mutual information pyramid method is improved, which shows that the registration method based on mutual information and image pyramid proposed in the embodiment of the present invention can well handle the registration of multimodal intracranial blood vessel magnetic resonance images.
  • the rotation matrix corresponding to the first bright blood image after registration in each test image pair can be obtained.
  • the mean value of the rotation matrices of all the test image pairs can be obtained by calculation.
  • the registration process of the registration method based on mutual information and image pyramid needs to iteratively calculate the mutual information of every two images to be registered.
  • the image pairs to be registered all use the registration method based on mutual information and image pyramid, which will result in a slow calculation speed.
  • the inventor considers that intracranial blood vessels can be regarded as a rigid body, which is different from organs such as the heart or lungs, which will change with human breathing and other movements, so the spatial coordinate transformation operation of each bright blood image is almost the same Consistent, that is, almost the same rotation matrix is used. See Table 3.
  • Table 3 shows the rotation matrix mean and mean square error calculated using the mutual information pyramid method, derived from patient A’s 160 400 ⁇ 400 enhanced black blood images and 160 400 ⁇ 400 bright blood images, after mutual
  • the rotation matrix obtained after the registration of the information pyramid method shows that the mean square error of the rotation matrix is very small.
  • the rotation matrix calculated by the registration of a few bright blood images of the patient can be used to perform the same spatial coordinate transformation on all the bright blood images, without the need to calculate the rotation matrix for each bright blood image, thereby speeding up the image matching. standard process.
  • the average value of the rotation matrix is used to perform coordinate transformation on the first bright blood image in the remaining preprocessed image pairs, which can quickly complete the registration of all images, which greatly improves the Registration speed.
  • image interpolation using the similarity measure based on mutual information, and using the predetermined search strategy to realize the process of image registration, please refer to the previous section, and can also be understood in combination with the relevant existing technologies. No longer.
  • FIG. 11 is a registration result of an intracranial blood vessel magnetic resonance image based on a Gaussian distribution sampling-based mutual information and image pyramid registration method and a mutual information pyramid method according to an embodiment of the present invention.
  • the normalized cross-correlation coefficient NCC is used, and the normalized mutual information NMI is used as the evaluation index.
  • Table 4 shows the registration method based on mutual information and image pyramid (referred to as mutual information pyramid method) for all image pairs of one patient, and the mutual information and image pyramid based on Gaussian distribution sampling proposed by the embodiment of the present invention
  • the registration results of the registration method (referred to as the method of the present invention) (the rest of the patient data are not displayed due to space limitations), and the experimental platform is Matlab R2016b. Since the method of the present invention does not need to register all the images, it only needs to randomly select a few images for registration. Therefore, in the experiment, the Gaussian distribution mean ⁇ is set to be half of the total number of images to be registered, and the standard deviation ⁇ is 1, respectively. 2, 3, 4, 5, and 6, 20 enhanced black blood images are randomly selected for registration with the corresponding bright blood images.
  • the value in a is the mean ⁇ mean square error of the evaluation index based on the registration of 160 bright blood images and 160 enhanced black blood images
  • the intracranial blood vessel can be regarded as a rigid body, and the spatial coordinate transformation operations of each bright blood image are almost the same, so the same rotation matrix can be used. Therefore, on the one hand, a small number of image pairs are selected for image registration of the bright blood images, and the average value of the rotation matrix of the registered few bright blood images is used to perform the same spatial coordinate transformation on the bright blood images of the remaining image pairs, without any further Rotation matrices are obtained for each of the remaining bright blood images, thus speeding up the image registration process.
  • the process of image registration based on mutual information as the similarity measure, and using the image pyramid algorithm to increase the model complexity, the registration accuracy and speed can be improved.
  • the image registration method in the embodiment of the present invention can unify the bright blood image and the enhanced black blood image in the same coordinate system, which can facilitate doctors to understand the intracranial blood vessel images corresponding to the black blood sequence and the bright blood sequence, and obtain a diagnosis simply and quickly.
  • the image registration is an important step for the subsequent flow void artifact removal.
  • the registration scheme provided by the embodiment of the present invention can provide a better reference method for the registration of other medical images, which has great clinical application value.
  • the image registration process in the embodiment of the present invention is an important basis for the subsequent elimination of the void artifact.
  • the flow void artifact in the enhanced black blood image after registration can be eliminated.
  • the reason for the flow void artifact is that during the imaging process of the blood vessel wall, the blood flow velocity at the tortuosity is caused by the small blood vessel. Slow, and the surrounding blood and tissue fluid may have signal contamination and other problems, resulting in the image obtained by the black blood sequence scan, the blood information that should be black instead appears bright color, thereby simulating the wall thickening or plaque appearance of normal individuals , exaggerating the degree of vascular stenosis.
  • the blood information in the bright blood image after registration is considered to be used to correct the incorrect blood information in the enhanced black blood image after registration, and the blood information in the bright blood image after registration is embedded in the registration After enhancing the black blood image, to achieve the effect of image fusion. Specifically, it can be achieved through the following steps:
  • S4 may include S41 to S44:
  • the blood in the bright blood image, the blood is high signal, while the surrounding brain tissue is low signal, the bright blood image after registration can be linearly transformed to grayscale, and the grayscale range of the image can be adjusted. , to achieve the purpose of improving image contrast.
  • FIG. 12 is a schematic diagram of the grayscale linear transformation and parameter setting provided by an embodiment of the present invention.
  • the small grayscale value change interval in the original bright blood image f after registration can be expanded into the new bright blood image f1 after registration (contrast-enhanced bright blood image)
  • the larger gray value change interval of the image is adjusted, and the gray scale range of the image is adjusted to achieve the purpose of improving the contrast of the bright blood image after registration.
  • a contrast-enhanced bright blood image can be obtained. Referring to FIG. 13 , FIG.
  • the grayscale linear transformation is a result graph of the grayscale linear transformation according to the embodiment of the present invention, that is, the result graph of the bright blood image after registration after the grayscale linear transformation.
  • the left image is the bright blood image after registration
  • the right image is the result image after grayscale linear transformation. It can be seen that the contrast of the blood part in the right image is significantly enhanced compared with the surrounding pixels. Since the medical image pixel range is large, it may be -1000 to +1000. Through this step, the pixel range can be normalized to 0 to 255, which is a pixel range that conforms to general image processing, which is convenient for subsequent processing.
  • the specific process of the grayscale linear transformation reference may be made to the related prior art, which will not be repeated here.
  • S42 may include the following steps:
  • threshold segmentation The method used in this step is called threshold segmentation.
  • the preset image binarization method is the image binarization process, which can set the grayscale of the point on the image to 0 or 255, that is, the entire image presents an obvious black and white effect. That is, a grayscale image with 256 brightness levels is selected by appropriate thresholds to obtain a binarized image that can still reflect the overall and local characteristics of the image.
  • the blood information in the contrast-enhanced bright blood image can be highlighted in white and irrelevant information in black by the preset image binarization method, so as to extract the bright blood feature map corresponding to the blood information.
  • the preset image binarization method in the embodiment of the present invention may include maximum inter-class variance method OTSU, kittle, and the like.
  • T(x, y) is the gray value of the contrast-enhanced bright blood image
  • F(x, y) is the gray value of the bright blood feature map
  • T is the first threshold
  • the maximum inter-class variance method OTSU is used, and the result is shown in FIG. 14 , which is an image binarization result diagram according to an embodiment of the present invention, wherein the left image is a contrast-enhanced bright blood image, and the right image is a contrast-enhanced bright blood image.
  • the picture shows the blood information after threshold extraction. It can be seen that the brightly colored part in the right image is only information related to blood.
  • S43 Perform image fusion between the bright blood feature map and the enhanced black blood image corresponding to the registered bright blood image according to a preset fusion formula, to obtain a target enhanced black blood image for eliminating flow-space artifacts corresponding to the enhanced black blood image ;
  • this step first establish a spatial mapping relationship between the bright blood feature map and the corresponding enhanced black blood image, map the bright blood feature map to the corresponding enhanced black blood image, and perform image fusion according to a preset fusion formula, wherein
  • the default fusion formula is:
  • F(x,y) is the gray value of the bright blood feature map
  • R(x,y) is the gray value of the corresponding enhanced black blood image
  • g(x,y) is the fused target enhanced black blood The grayscale value of the image.
  • the gray value of the void artifact which should be black but appears bright in the corresponding enhanced black blood image, can be changed to black, so as to achieve the purpose of eliminating the void artifact.
  • FIG. 15 shows the results of eliminating flow void artifacts obtained by different methods for intracranial blood vessels in the embodiment of the present invention, in which the flow void artifact appears at the place indicated by the arrow, and the left picture shows the flow void artifact appearing
  • the original image of the intracranial blood vessel enhanced black blood image the second image on the left is the result obtained by using the flow-space artifact elimination method based on the registration method based on mutual information and image pyramid, and the third image on the left is using the mutual information based on Gaussian distribution sampling.
  • the flow-space artifact elimination method of the registration method of the image pyramid where the standard deviation ⁇ of the Gaussian distribution is 3, it can be seen that the mutual information based on the Gaussian distribution sampling adopted in the embodiment of the present invention and the registration of the image pyramid
  • the elimination effect of the void artifact elimination method of the method is better than that of the void artifact elimination method based on the registration method based on mutual information and image pyramid.
  • the process speed of the flow-space artifact elimination method based on the Gaussian distribution sampling based mutual information and image pyramid registration method is faster than that of the flow-space artifact elimination method based on mutual information and pyramids. Fast, experimentally confirmed to save about 80% of the time.
  • the artifact-removed enhanced black blood image group can be obtained.
  • image registration based on the mutual information of Gaussian distribution sampling and the registration method of image pyramid is performed on the bright blood image and the enhanced black blood image scanned by the magnetic resonance angiography technology, which can improve the registration Quasi-accuracy and registration speed.
  • the blood information is extracted from the registered bright blood image by threshold segmentation, and it is fused into the registered enhanced black blood image, so as to correct the incorrect blood information in the registered enhanced black blood image. , changing the gray value of the flow void artifact that appears as a bright color to black, so as to achieve the purpose of eliminating the flow void artifact and obtain a more accurate and comprehensive intracranial blood vessel image.
  • the solution provided by the embodiment of the present invention is to eliminate the flow void artifact from the perspective of image post-processing, without using new imaging technology, imaging mode or pulse sequence, so the flow void artifact can be eliminated simply, accurately and quickly, and the It can achieve better promotion in clinical application.
  • Each target enhanced black blood image is subtracted from the corresponding black blood image, and a contrast enhanced image with contrast enhancement effect can be obtained.
  • K contrast enhancement images can be obtained. It is understood that the K contrast-enhanced images are all two-dimensional images.
  • the bright blood images after registration are all two-dimensional images.
  • the blood information can be represented as a three-dimensional structure, and a three-dimensional blood model can be established.
  • the process of obtaining a three-dimensional model with a three-dimensional effect through interpolation of a two-dimensional image is called three-dimensional reconstruction.
  • Current 3D reconstruction techniques include the Marching Cubes (MC) method, the Maximum Intensity Projection (MIP) method, the Surface Shadow Covering (SSD), the Volume Walkthrough (VRT), the Surface Reconstruction (CPR), Virtual Endoscopy (VE) and more.
  • MC Marching Cubes
  • MIP Maximum Intensity Projection
  • SSD Surface Shadow Covering
  • VRT Volume Walkthrough
  • CPR Surface Reconstruction
  • VE Virtual Endoscopy
  • any three-dimensional reconstruction method can be used to establish a three-dimensional blood model.
  • the blood three-dimensional model can initially simulate three-dimensional blood vessels, and visually display the direction of blood vessels and the lesion area.
  • S6 may include S61 to S63:
  • the K contrast-enhanced bright blood images obtained in step S41 may be obtained. It can be understood by those skilled in the art that the K contrast-enhanced bright blood images are actually stacked into a three-dimensional cube data. For convenience of distinction, it is named as the first three-dimensional volume data in the embodiment of the present invention.
  • the maximum inter-class variance method OTSU is still used to determine the threshold, but it is different from the method used to determine the first threshold in S421.
  • the maximum inter-class variance method OTSU is used to obtain the large volume located in the first three-dimensional volume data.
  • a threshold corresponding to a plurality of contrast-enhanced bright blood images in a small cube near the middle part is used as the second threshold.
  • the small cube data (second three-dimensional volume data) in the center is selected to determine the second threshold, which can reduce the threshold calculation
  • the second threshold is accurately applicable to all blood information in the first three-dimensional volume data.
  • the center point of the first three-dimensional volume data may be determined first, and then extended in six directions corresponding to the cube with preset side lengths, so as to determine the size of the second three-dimensional volume data, wherein the preset
  • the side length can be determined according to an empirical value including the Willis ring, for example, it is 1/4 of the side length of the cube of the first three-dimensional volume data, and so on.
  • the circle of Willis is the most important intracranial collateral circulation pathway, connecting the two hemispheres with the anterior and posterior circulations.
  • the moving cube method is a three-dimensional reconstruction method, which can directly obtain a three-dimensional blood model by processing the first three-dimensional volume data according to a given input threshold.
  • the moving cube method has the advantage of generating better mesh quality.
  • the specific processing process of the first three-dimensional volume data by the moving cube method please refer to the related prior art, which will not be repeated here.
  • FIG. 16( a ) is an effect diagram of a blood three-dimensional model for intracranial blood vessels according to an embodiment of the present invention.
  • the blood 3D model obtained in step S6 is actually the flow direction and regional distribution of intracranial blood. Since there are blood vessel walls around the blood in practice, the blood 3D model cannot fully represent the real intracranial blood vessels.
  • step S7 the blood boundary in the bright blood image after registration can be expanded, so that the expanded blood boundary can cover the range of the intracranial blood vessel wall, forming the effect of a hollow tube, and then expand the blood boundary.
  • the latter two-dimensional image uses a three-dimensional reconstruction method to generate a three-dimensional model, thereby obtaining a three-dimensional blood vessel model that is closer to the real intracranial blood vessel than the blood three-dimensional model in S5.
  • the expansion of the blood boundary can be achieved by detecting the blood boundary pixel points in the bright blood image after registration, and extending the detected pixel points to a preset direction by a preset number of pixels.
  • the preset pixel points can be based on a large number of intracranial blood vessel diameters. and the empirical value obtained from the blood vessel wall thickness data to select.
  • the manner of expanding the blood boundary in the embodiment of the present invention is not limited to this.
  • S7 may include S71 to S75:
  • the K bright blood feature maps obtained in step S42 are obtained.
  • the dilation operation is a kind of morphological operation.
  • the dilation operation can fill the holes in the image and expand the protruding points of the object at the edge.
  • the final expanded object is larger than the original area.
  • the dilation operation can be written as defined as Among them, B is the structural element, and A is the original image.
  • the original image A here is the bright blood feature map. There are only two pixel values 0 and 255 in the bright blood feature map, 0 corresponds to black, and 255 corresponds to white.
  • Structural elements are also called kernels (referred to as kernels for short), and the kernels can be regarded as a convolution kernel.
  • the expansion operation is to use this convolution kernel B to perform a convolution operation on the original image A to find the local maximum.
  • the convolution kernel B usually has an anchor point, which is usually located in the center of the convolution kernel.
  • the maximization operation causes the bright areas in the image to grow (hence the name dilation).
  • the convolution kernel is used to translate from left to right and top to bottom on the original image. If there is white in the box corresponding to the convolution kernel, then all the colors in the box will become white.
  • the kernel can be rectangle, ellipse, circle.
  • the desired kernel can be obtained by passing the shape and size of the kernel in the OpenCV function cv2.getStructuringElement().
  • a circular kernel with a radius of 1 can be used to expand the bright blood feature map in multiple steps, until the maximum gradient position is reached, and the expansion is stopped, thereby determining the outer wall of the blood vessel, realizing the segmentation of the blood vessel wall, and obtaining bright blood.
  • the related prior art which will not be repeated here.
  • the difference feature map obtained for each bright blood feature map in this step is a two-dimensional plane map similar to a hollow blood vessel. Similarly, the pixel values of the difference feature map are only 0 and 255.
  • one pixel value may be selected as the third threshold value according to the empirical value of all difference feature maps, for example, any value between 100 and 200 may be selected, such as 128, as the third threshold value.
  • the moving cube method uses the third threshold as the input threshold, and can obtain the three-dimensional model of the blood vessel with the expansion of the blood boundary from the K difference feature maps.
  • the specific implementation process of the method for moving the cube will not be repeated here.
  • FIG. 16(b) is an effect diagram of a three-dimensional blood vessel model for intracranial blood vessels according to an embodiment of the present invention.
  • the image is processed in grayscale, and in practice, it can be displayed in colors such as blue.
  • This step can be implemented by using the moving cube method, and details refer to S6 and S7, which will not be repeated here.
  • FIG. 16( c ) is an effect diagram of the contrast-enhanced three-dimensional model for intracranial blood vessels according to an embodiment of the present invention.
  • the image is processed in grayscale, and in practice, it can be displayed in colors such as red.
  • S9 may include the following steps:
  • the search range of the contrast-enhanced characteristics of the vessel wall in the three-dimensional blood vessel model obtained by S7 is used to determine the contrast-enhanced characteristics obtained by S8. Whether the 3D model is located in the blood vessel wall area near the blood, that is, it is judged whether there is an overlapping part with the 3D model of the blood vessel in the contrast-enhanced 3D model. Contrast-enhanced three-dimensional models were obtained after preservation.
  • the blood vessel wall with obvious angiography can be visually displayed, and the area of the intracranial blood vessel where the angiography is enhanced can be clearly seen. The effect is most pronounced, then atherosclerotic or vulnerable plaque may develop in that area.
  • plaque enhancement index CE can be obtained for any point on the blood vessel wall in the three-dimensional model of enhanced intracranial angiography.
  • S preBBMR and S postBBMR are the signal intensities in the black blood image and the contrast-enhanced black blood image, respectively.
  • S preBBMR and S postBBMR are information carried in the images after the black blood image and the contrast-enhanced black blood image are captured, respectively.
  • the above information is used to obtain the plaque enhancement index CE of each point on the edge of the intracranial blood vessel wall, and it is reflected in the three-dimensional model of intracranial angiography enhancement, which can facilitate the doctor to obtain more detailed blood vessel information.
  • a plaque threshold such as 0.5
  • it means that there is plaque on the blood vessel wall. Therefore, by measuring the plaque enhancement index in the blood vessel wall area, it is helpful to identify the responsible intracranial arterial plaque, etc., which can provide Valuable diagnostic aids.
  • FIG. 17 is an effect diagram of an angiography-enhanced three-dimensional model for intracranial blood vessels according to an embodiment of the present invention.
  • the images are grayscaled.
  • different colors can be used to distinguish, for example, blue is the blood vessel part without contrast enhancement, and red is the blood vessel part with contrast enhancement.
  • the bright-colored parts in the white circles in the accompanying drawings are the intracranial blood vessels with contrast enhancement, that is, there may be intracranial atherosclerosis or vulnerable plaques, and the rest are those without contrast enhancement.
  • the angiography-enhanced three-dimensional model can realize basic functions such as rotation, zoom-in and zoom-out, so as to assist doctors in locating the lesion area and make more accurate judgments.
  • the bright blood image and the enhanced black blood image scanned by the magnetic resonance angiography technique are firstly registered by the mutual information and image pyramid registration method based on Gaussian distribution sampling, which can improve the The registration efficiency enables the image to improve the registration accuracy layer by layer from low resolution to high resolution.
  • the bright blood image and the enhanced black blood image can be unified in the same coordinate system.
  • using the registered bright blood image to eliminate the flow-space artifacts on the enhanced black blood image it can display more accurate and comprehensive blood vessel information.
  • the solution provided by the embodiment of the present invention is to eliminate the flow void artifact from the perspective of image post-processing, without using new imaging technology, imaging mode or pulse sequence, so the flow void artifact can be eliminated simply, accurately and quickly, and the It can achieve better promotion in clinical application.
  • a three-dimensional model of blood and a three-dimensional model of blood vessels with expanded blood boundary are established by using the registered bright blood image, and a contrast-enhanced three-dimensional model with contrast-enhanced effect is obtained by subtracting the enhanced black blood image and the black blood image by eliminating the artifacts.
  • the blood three-dimensional model, the blood vessel three-dimensional model, and the contrast-enhanced three-dimensional model are obtained to obtain an angiography-enhanced three-dimensional model corresponding to the blood vessel wall with contrast-enhancing effect.
  • the angiography-enhanced three-dimensional model realizes the three-dimensional visualization of intracranial blood vessels, without the need for doctors to restore the intracranial blood vessel tissue structure and disease characteristics through imagination. It provides realistic three-dimensional space information of blood vessels, which is convenient for intuitively displaying the blood vessel wall with obvious contrast enhancement, and is convenient for locating and displaying the lesion area. In clinical application, the real information of intracranial blood vessels can be easily and quickly obtained to analyze vascular lesions.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Software Systems (AREA)
  • Geometry (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Computer Graphics (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)

Abstract

本发明公开了一种颅内血管造影增强三维模型的建立方法，包括：获取颅内血管部位的亮血图像组、黑血图像组和增强黑血图像组；对每个图像对预处理得到第一亮血图像和第一黑血图像；对每个第一亮血图像以对应的第一黑血图像为基准，利用基于高斯分布抽样的互信息和图像金字塔的配准方法进行图像配准得到配准后亮血图像组；进行流空伪影消除操作得到伪影消除增强黑血图像组；将伪影消除增强黑血图像组和黑血图像组中对应图像相减得到K个造影增强图；利用配准后亮血图像组建立血液三维模型和血液边界扩展的血管三维模型；利用K个造影增强图建立造影增强三维模型；基于血液三维模型、血管三维模型、造影增强三维模型得到颅内血管造影增强三维模型。

Description

一种颅内血管造影增强三维模型的建立方法 技术领域
本发明属于图像处理领域，具体涉及一种颅内血管造影增强三维模型的建立方法。
背景技术
随着我国国民经济的快速发展，人们对健康问题越来越重视。《Lancet》在2019年6月发表的论文分析了从1990年到2017年，中国34个省份(包括港澳台)居民的死亡原因，发现高居中国人死亡原因第一位的是脑卒中。脑卒中是颅内血管破裂、狭窄或堵塞引起脑组织坏死，进而产生的一系列症状，包括脑出血、脑梗死等，如果治疗不及时，患者可能会死亡；而即使治疗及时，也有可能造成患者残疾。
目前临床上对于颅内血管病变程度与血管狭窄化程度的评估，通常使用基于管腔成像的方法，如数字减影血管造影术(Digital Subtraction Angiography,DSA)、CT血管成像(Computed Tomography Angiography,CTA)、磁共振血管成像(Magnetic Resonance Angiography,MRA)以及高分辨率磁共振血管成像(High-Resolution Magnetic Resonance Angiography,HRMRA)等。颅内动脉血管与颈动脉和椎动脉相连，在脑底部形成环状结构，结构形态特殊，走形曲折，且管壁厚度极薄。通过磁共振血管成像技术，能够清晰地描绘出颅内动脉血管的路径。
其中，磁共振血管成像技术(MRA或HRMRA)作为一种对患者无创的成像方法，可以清晰地检测到颅内血管的血管壁结构并进行分析，扫描得到的磁共振图像对于软组织的分辨率高，没有骨伪影，图像质量好，且能够使用多种序列扫描得到具有不同成像特点的组织结构，在颅内血管的显示上具有明显的优越性。
由于磁共振血管成像技术得到的亮血序列、黑血序列对应的图像均为二维图像，在临床上，医生需要凭借经验结合两种图像的信息，来获得颅内血管的综合情况，以进行颅内血管病变分析。但二维图像具有局限性，不利于简便快速地获得颅内血管的真实信息。
发明内容
为了在临床应用上，简便快速地获得颅内血管的真实信息，以进行颅内血管病变分析。本发明实施例提供了一种颅内血管造影增强三维模型的建立方法。包括：
获取颅内血管部位的亮血图像组、黑血图像组和增强黑血图像组；其中，所述亮血图像组、所述黑血图像组、所述增强黑血图像组分别包括K个亮血图像、黑血图像和增强黑血图像；所述亮血图像组、所述黑血图像组、所述增强黑血图像组中的图像一一对应；K为大于2的自然数；
将每个亮血图像和对应的增强黑血图像作为一个图像对，对每个图像对进行预处理，得到该图像对的第一亮血图像和第一黑血图像；
针对每个第一亮血图像，以对应的第一黑血图像为基准，利用基于高斯分布抽样的互信息和图像金字塔的配准方法进行图像配准，得到包括K个配准后亮血图像的配准后亮血图像组；
利用所述配准后亮血图像组，对所述增强黑血图像组中的增强黑血图像进行流空伪影消除操作，得到包括K个目标增强黑血图像的伪影消除增强黑血图像组；
将所述伪影消除增强黑血图像组和所述黑血图像组中对应图像相减，得到K个造影增强图；
利用所述配准后亮血图像组，建立血液三维模型；
利用所述配准后亮血图像组建立血液边界扩展的血管三维模型；
利用所述K个造影增强图建立造影增强三维模型；
基于所述血液三维模型、所述血管三维模型、所述造影增强三维模型，得到颅内血管造影增强三维模型。
本发明实施例所提供的方案中，首先对磁共振血管成像技术扫描得到的亮血图像和增强黑血图像采用基于高斯分布抽样的互信息和图像金字塔的配准方法进行图像配准，可提高配准效率，使图像从低分辨率到高分辨率逐层提高配准精度。通过上述图像配准可以将亮血图像和增强黑血图像统一在相同坐标系下。其次利用配准后亮血图像对增强黑血图像进行流空伪影消除操作，可以显示更准确、全面的血管信息。本发明实施例所提供的方案是从图像后处理的角度对流空伪影进行消除，无需使用新的成像技术、成像模式或脉冲序列，因此可以简便、准确、快速地消除流空伪影，并可以在临床应用实现较好的推广。再次，利用配准后亮血图像建立血液三维模型、血液边界扩展的血管三维模型，通过将伪影消除增强黑血图像和黑血图像相减得到具有造影增强效果的造影增强三维模型；最后基于血液三维模型、血管三维模型、造影增强三维模型，得到具有造影增强效果的血管壁对应的颅内血管造影增强三维模型。该颅内血管造影增强三维模型实现了颅内血管的三维可视化，无需医生通过想象力还原颅内血管组织结构及病症特征等，能够方便医生从任意感兴趣的角度、层次观察和分析血管形态特征，可以提供具有真实感的血管三维空间信息，便于直观显示有明显造影增强的血管壁，便于定位与显示病灶区域。能够在临床应用上，简便快速地获得颅内血管的真实信息，以进行血管病变分析。
当然，实施本发明的任一产品或方法并不一定需要同时达到以上所述的所有优点。
附图说明
图1为本发明实施例提供的一种颅内血管造影增强三维模型的建立方法的流程示意图；
图2为本发明实施例的颅内血管磁共振图像坐标变换示意图；
图3为本发明实施例的两种搜索策略配准对比结果；
图4为本发明实施例的颅内血管磁共振图像经过预配准后的结果图；
图5为本发明实施例的颅内血管磁共振图像的待配准区域示意图；
图6(a)为本发明实施例的颅内血管磁共振图像的亮血高斯金字塔和黑血高斯金字塔；图6(b)为本发明实施例的颅内血管磁共振图像的亮血拉普拉斯金字塔和黑血拉普拉斯金字塔；
图7为本发明实施例的颅内血管磁共振图像的拉普拉斯金字塔图像的配准结果；
图8为本发明实施例的颅内血管磁共振图像基于互信息的高斯金字塔图像配准步骤示意图；
图9为本发明实施例的不同迭代次数下的归一化互信息；
图10为包含互信息金字塔方法的多种配准方法的颅内血管磁共振图像的配准结果；
图11为本发明实施例的基于高斯分布抽样的互信息和图像金字塔的配准方法与互信息金字塔方法的颅内血管磁共振图像的配准结果；
图12为本发明实施例所提供的灰度线性变换及参数设置示意图；
图13为本发明实施例的灰度线性变换结果图；
图14为本发明实施例的图像二值化结果图；
图15为本发明实施例中针对颅内血管用不同方法得到的流空伪影消除结果；
图16中(a)(b)(c)分别为本发明实施例针对颅内血管的血液三维模型效果图、血管三维模型效果图和造影增强三维模型效果图；
图17为本发明实施例的颅内血管造影增强三维模型效果图。
具体实施方式
为了在临床应用上，简便快速地获得颅内血管的真实信息，以进行颅内血管病变分析。本发明实施例提供了一种颅内血管造影增强三维模型的建立方法。
需要说明的是，本发明实施例所提供的一种颅内血管造影增强三维模型的建立方法的执行主体可以为一种颅内血管造影增强三维模型的建立装置，该装置可以运行于电子设备中。其中，该电子设备可以为血管成像设备，或者一图像处理设备，当然并不局限于此。
如图1所示，本发明实施例所提供的一种颅内血管造影增强三维模型的建立方法，可以包括如下步骤：
S1，获取颅内血管部位的亮血图像组、黑血图像组和增强黑血图像组；
其中，亮血图像组、黑血图像组、增强黑血图像组分别包括K个亮血图像、黑血图像和增强黑血图像；亮血图像组、黑血图像组、增强黑血图像组中的图像一一对应；K为大于2的自然数；
亮血图像组是对颅内血管部位使用磁共振血管成像技术进行亮血序列扫描得到的图像组。黑血图像组是对颅内血管部位使用磁共振血管成像技术进行黑血序列扫描得到的图像组。增强黑血图像组是对患者先注入顺磁性造影剂，再对颅内血管部位使用磁共振血管成像技术进行黑血序列扫描得到的图像组。本发明实施例中，磁共振血管成像技术优选为HRMRA。
亮血图像组、黑血图像组、增强黑血图像组中的K个图像是一一对应的，其中对应方式是按照扫描时间形成的图像次序相同。
S2，将每个亮血图像和对应的增强黑血图像作为一个图像对，对每个图像对进行预处理，得到该图像对的第一亮血图像和第一黑血图像；
该步骤可以理解为图像的预处理过程，可选的一种实施方式中，对每个图像对进行预处理，得到该图像对的第一亮血图像和第一黑血图像，可以包括S21和S22：
S21，针对每个图像对，以增强黑血图像为基准，将亮血图像进行坐标变换和图像插值，使用基于互信息的相似性度量，并采用预定搜索策略，得到预配准后的第一亮血图像；
增强黑血图像是按冠状面扫描成像的，而亮血图像却是按轴状面扫描成像的，序列扫描方向的不同导致两者最终的磁共振成像层面不同，因此需要通过坐标变换来实现在一个标准参考坐标系下观察不同成像层面的磁共振图像。
针对血管图像，可以利用医学图像DICOM(Digital Imaging and Communications in Medicine,医学数字成像与通信)文件中的方向信息实现图像的坐标变换。DICOM文件是CT或核磁共振等医疗设备的图像保存格式。DICOM3.0格式影像文件中含有与成像方向有关的方位标签信息，该信息简要地介绍了患者与影像仪器之间的方位关系，通过其中的数据，可以得知在图像中每个像素准确的位置信息。具体的，增强黑血图像和亮血图像为待配准的图像，可以根据亮血图像的DICOM文件中的方位标签信息，以增强黑血图像的坐标系为基准，将增强黑血图像作为参考图像，将亮血图像作为浮动图像，将亮血图像进行坐标变换，实现旋转亮血图像至与增强黑血图像相同坐标系的目的，旋转后亮血图像的扫描方向也变为呈冠状面。
为了便于理解本发明实施例的方法，以下结合图像配准过程进行简要介绍，具体过程可以借鉴相关现有技术进行了解。
对于两幅图像A和B的配准，实际上就是将图像A中的每个坐标位置通过一个映射关系，对应到图像B中。具体的坐标变换方法可以包括刚体变换、仿射变换、投影变换和非线性变换等。由于颅内血管可以看作是一个刚体，因此，本发明实施例选用刚体变换作为坐标变换方法。
但是在坐标变换过程中，浮动图像的坐标系会发生拉伸或形变，经过坐标变换后的图像像素坐 标与原图的采样网格并不会完全重合，即原先为整数的像素坐标点经过坐标变换后可能不再是整数，导致图像的有些区域丢失掉部分像素，因此在图像坐标变换的过程中，需要同时对图像进行重采样插值，来确定经过坐标变换后图像像素坐标点的灰度值，便于后续处理。具体的，就是经过坐标变换后的亮血图像坐标可能被映射到了原图的非整数坐标上，所以需要同时对亮血图像进行图像插值。图像插值方法包括最近邻插值、双线性插值和双三次插值等。本发明实施例对三种插值方法进行了实验，共设定了5项评价指标，分别是均方根误差RMSE、峰值信噪比PSNR、归一化互相关系数NCC、归一化互信息NMI以及耗时Time，其中RMSE越小配准越精确，PSNR、NCC与NMI值越高配准越精确。从整体实验数据上看，双三次插值的精度明显优于最近邻插值和双线性插值，因此，选定双三次插值。
使用图像插值方法将缺失的像素点进行图像恢复之后，还需要使用某种相似性度量来计算参考图像和变化后的浮动图像的相似性，接着利用搜索策略找到最优的相似性度量，循环反复迭代求优，直到两幅图像的相似性度量达到最优时，迭代停止，最后根据确定的空间变换矩阵(旋转矩阵)对浮动图像进行坐标转换，实现图像完全配准。待配准的图像在经历一个迭代算法优化后，能够计算得到两幅图像的空间位置配准关系以及配准图像，使得配准后的浮动图像与参考图像相似度达到最高。
衡量两幅图像之间特征相似性的尺度为相似性度量，选择合适的相似性度量可以提高配准精度，有效抑制噪声等，它在图像的配准中有着非常重要作用。常用的相似性度量主要分为三大类，分别是距离度量、相关度量以及信息熵。在本发明实施例中，针对颅内血管，其可以看作是一个刚体，几乎不会发生形变，不同于心脏或肺部等器官会随着人的呼吸等运动发生改变，因此针对颅内血管可以选用互信息或归一化互信息作为相似性度量，使配准效果更精确。
互信息和归一化互信息是信息熵中的一种。互信息(Mutual Information,MI)，它衡量了两幅图像之间的相关性，或是互相包含的信息量，用来解释两幅图像是否达到了最优配准，互信息的值越大，表示两幅图像越相似。或者也可以选择归一化互信息(Normalization Mutual Information,NMI)，它是互信息度量的改良，当两幅待配准图像的像素灰度级数类似的时候，使用NMI作为相似性度量，得到的配准图像准确度更高，更加可靠。NMI的取值范围为[0,1]，值越接近于1，表明两幅图像越相似。归一化互信息的概念，解决了当两幅图像的重叠部分较小或者重叠区域大部分为背景信息时，基于互信息的图像配准反而精度不高，配准效果不好的问题，降低了互信息对于图像重叠区域的敏感性。
图像配准实质上是一个多参数的优化问题，即通过使用某种搜索策略对图像进行空间坐标变化，最终使得两幅图像的相似性度量达到最优，其中搜索策略与空间坐标变化在实际计算过程中是彼此交叉进行的。算法思想是在每次迭代中计算两幅图像之间的相似性度量，并通过平移或旋转等坐标变换的操作调整浮动图像，同时对图像进行插值，一直到两幅图像的相似性度量最大为止。目前常用的搜索策略包括梯度下降优化器、基于进化策略(Evolution Strategy,ES)的(1+1)-ES等等，本发明实施例中的预定搜索策略可以根据需要选择。
具体的实验结果如下所示，参见图2，图2为本发明实施例的颅内血管磁共振图像坐标变换示意图，其中第一行分别为增强黑血图像和亮血图像，第二行分别为增强黑血图像和经过坐标变换后的亮血图像，可见，经过坐标变换后，亮血图像与增强黑血图像的扫描方向一致，均呈冠状面。
并且使用梯度下降优化器和(1+1)-ES这两种搜索策略分别对160幅亮血图像与相应扫描层面的160幅增强黑血图像进行配准，其中增强黑血图像为参考图像，亮血图像为浮动图像，配准结果显示如图3所示，图3为本发明实施例的两种搜索策略配准对比结果；图3中左图为未使用优化器配准的两 幅图像成对显示结果，中图为使用梯度下降优化器配准的图像成对显示结果，右图为使用(1+1)-ES优化器配准的图像成对显示结果。右图像显示采用蒙太奇效果，使用伪彩色透明处理增强黑血图像与亮血图像，紫色为增强黑血图像，绿色为亮血图像(附图图像是原图灰度处理后的图像，颜色未示出)。从图中可以看出，未使用优化器进行配准的图像中，增强黑血图像与亮血图像并未重合，且阴影较多；当使用梯度下降优化器配准图像时，虽然较左图配准效果较好，但在脑灰质处仍出现了明显的不重合现象；而使用(1+1)-ES优化器的图像中，配准结果精确，图像中不重合的阴影部分完全消失。表1所示数据为配准结果的3项评价指标，分别是归一化互信息NMI、归一化互相关系数NCC与算法耗时Time。从实验结果图上看，(1+1)-ES的配准图像效果显示更清晰，优于梯度下降优化器；从实验数据上看，三项评价指标都表现了(1+1)-ES优化器的良好性能，因此，本发明实施例中，优选的，预定搜索策略为(1+1)-ES。
表1不同搜索策略下的结果分析
Figure PCTCN2021129608-appb-000001
a中的值是基于160幅亮血图像与160幅增强黑血图像配准的评价指标平均值±均方误差
参见图4，图4为本发明实施例的颅内血管磁共振图像经过预配准后的结果图。左图为预配准后的第一亮血图像，其中插值方法采用双三次插值；中图为增强黑血图像，可见两者都为冠状面，右图为两者直接叠加后的效果图，右图可以看到虽然经过了预配准，可在同一冠状面下观测到当前成像层下的亮血图像和增强黑血图像，但两者仍存在不重合现象，因此还需要后续进行图像精配准。
通过该步骤的预配准，可以初步实现在同一坐标系下对比相同扫描层面的磁共振图像，但是由于亮血序列和黑血序列扫描的时间不同，且患者可能在扫描前后发生了轻微的运动，所以上述操作只是一个粗糙的坐标变换，仅通过预配准并不能实现多模态磁共振图像的完全配准，但是该步骤可以为后续的精确配准环节省略不必要的处理过程，提高处理速度。
S22，从增强黑血图像中，提取与第一亮血图像的扫描范围相同的区域内容，形成第一黑血图像。
由于血管成像在不同磁共振序列中的扫描范围不同，当亮血图像经过图像坐标变换后，其冠状面的信息并没有增强黑血图像的信息丰富，因此为能够更快速、准确地配准好两种序列下的同一区域，可以根据第一亮血图像的扫描区域，在增强黑血图像中提取出相同的扫描区域，缩小后续图像的配准范围。可选的，S22可以包括以下步骤：
1.获得第一亮血图像中血管的边缘轮廓信息；
具体可以使用Sobel边缘检测方法等方法获取边缘轮廓信息。边缘轮廓信息包含各个边缘点的坐标值。
2.在边缘轮廓信息中，提取横坐标、纵坐标的最小值和最大值，基于获得的四个坐标值确定初始提取边框；
也就是在边缘轮廓信息中，提取最小横坐标值、最大横坐标值、最小纵坐标值、最大纵坐标值，利用这四个坐标值确定方形边框的四个顶点，从而得到初始提取边框；
3.在第一亮血图像的尺寸范围内，将初始提取边框沿四个方向分别扩大预设数量个像素大小，得到最终提取边框；
其中，四个方向分别是横纵坐标的正负方向；预设数量根据血管图像的类型合理选择，目的是保证扩大后的最终提取边框不超过第一亮血图像的尺寸范围，比如预设数量可以为20等。
4.在增强黑血图像中提取最终提取边框中对应的区域内容，形成第一黑血图像。
依据最终提取边框划定的坐标范围，提取增强黑血图像中对应区域的内容，将提取出的内容形成第一黑血图像。该步骤通过提取待配准区域来获取两种模态下磁共振图像的共同扫描范围，有利于后续的快速配准。
参见图5，图5为本发明实施例的颅内血管磁共振图像的待配准区域示意图，其中左图为预配准后的第一亮血图像，右图为增强黑血图像，方框为增强黑血图像中待提取区域。这个区域包含了在颅内血管磁共振图像中，亮血序列以及黑血序列的共同扫描范围，通过确定待提取区域，能够更快速地关注到有用信息。
本发明实施例的上述两步预处理过程有着非常重要的作用，经过该预处理后的图像可以更多地关注有用信息，排除无关信息，在实际使用中，使用该图像预处理可以提高颅内血管图像配准、识别的可靠性。
S3，针对每个第一亮血图像，以对应的第一黑血图像为基准，利用基于高斯分布抽样的互信息和图像金字塔的配准方法进行图像配准，得到包括K个配准后亮血图像的配准后亮血图像组；
可选的一种实施方式中，S3可以包括S31～S34：
S31，采用高斯分布抽样选取预处理后的部分图像对作为测试图像对；
本发明实施例的测试图像对是待配准的图像对，本发明实施例对待配准图像对的随机选取采用高斯分布抽样。这是由于亮血图像与增强黑血图像的扫描方向不同，在图像预处理过程中，为在同一成像层面上观测亮血图像与增强黑血图像，亮血图像经过了坐标变换与插值，使得每一张亮血图像与当前层面的增强黑血图像对应；同时由于亮血图像与增强黑血图像的扫描范围不同，亮血图像边缘层的数据可能并不完整。综上原因，扫描中间层的亮血数据与增强黑血数据最丰富，因此选取高斯均值μ为待配准图像总数的一半，使得高斯随机选取到中间层图像配准的概率最大。
S32，对每个测试图像对中的第一亮血图像和第一黑血图像，采用基于互信息和图像金字塔的配准方法进行图像配准，得到配准后该测试图像对中第一亮血图像对应的旋转矩阵；可选的一种实施方式中，S32具体可以包括步骤S321～S324：
S321，针对每个测试图像对，基于下采样处理，由第一亮血图像得到亮血高斯金字塔，由第一黑血图像得到黑血高斯金字塔；其中，亮血高斯金字塔和黑血高斯金字塔中包括从下至上分辨率依次变小的m个图像；m为大于3的自然数；
为提高图像配准的准确度，避免图像在配准过程中收敛到局部最大值，可以使用多分辨率策略来解决局部极值的问题，同时多分辨率策略可以在满足图像配准精度的条件下，提高算法执行速度，增加鲁棒性。构建图像金字塔就是一种通过增加模型复杂度来提高配准精度和速度的有效方式，即在配准过程中，按从粗配准到精配准的顺序执行，首先对低分辨率的图像进行配准，接着在低分辨率图像配准完成的基础上，对分辨率高的图像进行配准。
可选的一种实施方式中，S321步骤可以包括：
获取第i层的输入图像，对第i层的输入图像利用高斯内核进行滤波，并删除滤波后图像的偶数行和偶数列，得到高斯金字塔的第i层图像G i，并将第i层图像G i作为第i+1层的输入图像，得到高斯金字塔的第i+1层图像G i+1；其中，i＝1、2，…，m-1；当高斯金字塔为亮血高斯金字塔时，第1层的输入图像为第一亮血图像，当高斯金字塔为黑血高斯金字塔时，第1层的输入图像为第一黑血图像。
具体的，高斯金字塔中的多个图像是同一原图像不同分辨率所对应的图像。高斯金字塔通过高斯滤波和下采样来获取图像，它的每一层构建步骤可以分为两步：首先使用高斯滤波对图像进行平滑滤波，即采用高斯内核进行滤波；接着删除滤波后图像的偶数行和偶数列，即将低一层图像的宽和高缩小一半，得到当前层图像，因此当前层图像为低一层图像大小的四分之一，通过不断地迭代以上步骤，最终可以得到高斯金字塔。
高斯滤波其实是一种低通滤波器，高斯金字塔中的图像频率范围很广，其中低一层图像的截止频率是高一层图像截止频率的2倍。高斯滤波先使用高斯函数来计算得到一个权重矩阵，再使用该权重矩阵对原图像做卷积运算，一般可以使用二维的高斯模板进行上述处理。虽然使用二维的高斯模板可以实现模糊图像的效果，但是当一个点在边界，周围没有足够的点时，会由于权重矩阵的关系导致边缘图像缺失，因此本发明实施例对二维高斯模板进行优化。可以将二维高斯滤波拆分成两个独立的一维高斯滤波，分别在横纵两个方向上进行图像滤波。将高斯函数分离，不仅能够消除二维高斯模板产生的边缘，还可以极大地加快程序的运行速度。相较于其他模糊滤波器，高斯滤波既可以实现图像的模糊效果，又能更好地保留边际效果。
本步骤中通过对经过预处理后的第一亮血图像和第一黑血图像进行上述处理，可以得到亮血高斯金字塔和黑血高斯金字塔。如图6(a)所示，图6(a)为本发明实施例的颅内血管磁共振图像的亮血高斯金字塔和黑血高斯金字塔。
这些分辨率逐渐减小，来源于同一张图像不同分辨率的图像组合，排列起来形似金字塔，因此被称为图像金字塔，其中分辨率最高的图像位于金字塔底部，分辨率最低的图像位于金字塔顶部。在计算机视觉下不同分辨率的图像，恰恰模拟了在不同距离下人眼观测的一幅图像，在图像信息处理上，多分辨率的图像相较于传统的单分辨率图像，更容易获取图像的本质特征。
S322，基于上采样处理，利用亮血高斯金字塔得到亮血拉普拉斯金字塔，利用黑血高斯金字塔得到黑血拉普拉斯金字塔；其中，亮血拉普拉斯金字塔和黑血拉普拉斯金字塔中包括从下至上分辨率依次变小的m-1个图像；
由于高斯金字塔是向下采样，即缩小图像，因此会丢失掉图像的一部分数据。因此，本发明实施例为避免图像在缩放过程中的数据缺失，恢复细节数据，使用拉普拉斯金字塔，配合高斯金字塔一起实现图像重建，在高斯金字塔图像的基础上突出细节。
可选的一种实施方式中，S322步骤可以包括：
对高斯金字塔的第i+1层图像G i+1进行上采样，并用数据0填充新增的行和列，得到填充图像；
对填充图像利用高斯内核进行卷积，获得填充像素的近似值，得到放大图像；
将高斯金字塔的第i层图像G i与放大图像相减，得到拉普拉斯金字塔的第i层图像L i；其中，当高斯金字塔为亮血高斯金字塔时，拉普拉斯金字塔为亮血拉普拉斯金字塔，当高斯金字塔为黑血高斯金字塔时，拉普拉斯金字塔为黑血拉普拉斯金字塔。
由于拉普拉斯金字塔是图像经过下采样操作后与原图的残差，因此从下至上对比，拉普拉斯金字塔比高斯金字塔结构少一层高层图像。
具体的，生成拉普拉斯金字塔结构的数学公式如(1)所示，其中L i表示第i层拉普拉斯金字塔(亮血拉普拉斯金字塔或黑血拉普拉斯金字塔)，G i表示第i层高斯金字塔(亮血高斯金字塔或黑血高斯金字塔)，而UP操作为向上采样放大图像，符号
Figure PCTCN2021129608-appb-000002
是卷积符号，
Figure PCTCN2021129608-appb-000003
是在构建高斯金字塔中使用的高斯内核。此公式表明了拉普拉斯金字塔实质上是使用原图像减去先缩小、再放大的图像的残差数据构成的，是一种残差预测金字塔，核心思想是用来存储图像经过下采样操作后与原图的差异， 保留图像的高频信息，目的是为能够完整地恢复出每一层级进行下采样操作前的图像。由于之前下采样操作丢失的一部分信息并不能通过上采样来完全恢复，即下采样是不可逆的，所以图像先经过下采样，再进行上采样后的显示效果比原图模糊。通过存储图像经过下采样操作后与原图的残差，能够在高斯金字塔图像的基础上给不同频率层的图像增加细节，对细节等进行突出。
Figure PCTCN2021129608-appb-000004
对应于4层的高斯金字塔，本步骤可以得到图像层数为3的亮血拉普拉斯金字塔和黑血拉普拉斯金字塔。如图6(b)所示，图6(b)为本发明实施例的颅内血管磁共振图像的亮血拉普拉斯金字塔和黑血拉普拉斯金字塔。图像显示使用了伽马矫正实现更清晰的效果，伽马值为0.5。
S323，对亮血拉普拉斯金字塔和黑血拉普拉斯金字塔中对应层的图像进行配准，得到配准的亮血拉普拉斯金字塔；可选的一种实施方式中，S323步骤可以包括：
针对亮血拉普拉斯金字塔和黑血拉普拉斯金字塔中的每一层，将该层对应的黑血拉普拉斯图像作为参考图像，将该层对应的亮血拉普拉斯图像作为浮动图像，使用基于互信息的相似性度量，并采用预定搜索策略，实现图像配准，得到配准后的该层亮血拉普拉斯图像；
由配准后的多层亮血拉普拉斯图像，依据分辨率依次减小的顺序，从下至上构成配准的亮血拉普拉斯金字塔；其中，黑血拉普拉斯图像为黑血拉普拉斯金字塔中的图像，亮血拉普拉斯图像为亮血拉普拉斯金字塔中的图像。
该步骤中的配准过程类似于前述的预配准过程，通过对亮血拉普拉斯图像进行坐标变换、图像插值，使用相似性度量及预定搜索策略，实现图像配准，可以得到配准后的亮血拉普拉斯图像。其中，坐标变换、图像插值、相似性度量及预定搜索策略不再赘述。
结果如图7所示，图7为本发明实施例的颅内血管磁共振图像的拉普拉斯金字塔图像的配准结果，左图为黑血拉普拉斯金字塔中的参考图像，中图为亮血拉普拉斯金字塔中已配准好的图像，右图为左、中两幅图像直接叠加后的效果图，叠加图像显示采用蒙太奇效果，使用伪彩色透明处理增强黑血图像与亮血图像，其中紫色为增强黑血拉普拉斯金字塔图像，绿色为亮血拉普拉斯金字塔图像(附图图像是原图经过灰度处理的图像，颜色未示出)。
S324，利用配准的亮血拉普拉斯金字塔作为叠加信息，对亮血高斯金字塔和黑血高斯金字塔中各层图像进行自上而下的配准，得到配准的亮血高斯金字塔，并得到配准后该测试图像对中第一亮血图像对应的旋转矩阵。
在该步骤的配准中，需要对高斯金字塔中不同分辨率的图像进行配准，由于低分辨率图像的配准可以更容易把握住图像的本质特征，因此本发明实施例在低分辨率图像配准的基础上配准高分辨率图像，即将高斯金字塔图像从上到下配准，将上一层图像的配准结果作为下一层图像配准的输入。
可选的一种实施方式中，S324中利用配准的亮血拉普拉斯金字塔作为叠加信息，对亮血高斯金字塔和黑血高斯金字塔中各层图像进行自上而下的配准，得到配准的亮血高斯金字塔，可以包括：
对亮血高斯金字塔和黑血高斯金字塔中自上而下的第j层，将该层对应的黑血高斯图像作为参考图像，将该层对应的亮血高斯图像作为浮动图像，使用基于互信息的相似性度量，并采用预定搜索策略，实现图像配准，得到配准后的第j层亮血高斯图像；
将配准后的第j层亮血高斯图像进行上采样操作，并与配准后的对应层亮血拉普拉斯图像相加，利用相加后的图像替换亮血高斯金字塔中第j+1层的亮血高斯图像；
将第j+1层的黑血高斯图像作为参考图像，将替换后的第j+1层的亮血高斯图像作为浮动图像，使用预定相似性度量及预定搜索策略，实现图像配准，得到配准后的第j+1层亮血高斯图像；其中j＝1， 2，…，m-1，黑血高斯图像为黑血高斯金字塔中的图像，亮血高斯图像为亮血高斯金字塔中的图像。
重复以上操作，直至完成底层高斯金字塔图像的高分辨率配准，得到配准的亮血高斯金字塔。在配准过程中，类似于前述的预配准过程。其中，坐标变换、图像插值、相似性度量及预定搜索策略不再赘述。
基于互信息的高斯金字塔图像配准具体步骤如图8所示，图8为本发明实施例的颅内血管磁共振图像基于互信息的高斯金字塔图像配准步骤示意图。首先对顶层的低分辨率的黑血高斯图像和顶层的低分辨率亮血高斯图像进行基于互信息的配准；接着将已配准好的亮血高斯图像进行上采样操作，并与保留了高频信息，且根据上述操作已配准好的对应层的亮血拉普拉斯图像相加，作为下一层亮血高斯图像；接着将上述操作得到的亮血高斯图像作为输入图像，再与对应层的黑血高斯图像进行配准，重复以上操作，直至完成底层高斯金字塔图像的高分辨率配准。
在基于互信息的高斯金字塔图像配准中，需要对每一层亮血高斯图像和黑血高斯图像以归一化互信息作为相似性度量进行配准，通过循环迭代计算两幅图的NMI，直至NMI达到最大。其中当迭代次数太小时，不能完成图像的准确配准，但是当迭代次数太大时，计算量会急剧增加，图9为本发明实施例的不同迭代次数下的归一化互信息，当第一层图像，也就是高斯金字塔中分辨率最高的底层图像配准达到最大NMI值且数据稳定时，停止迭代。
至此，得到配准的亮血高斯金字塔，其中的亮血图像的坐标系与增强黑血图像的坐标系一致，且图像具备较高的相似性，可以完成本发明实施例的血管图像配准过程。并且，配准后可以得到配准后该测试图像对中第一亮血图像对应的旋转矩阵。
为验证本发明实施例中的基于互信息和图像金字塔的配准方法(简称为互信息金字塔方法)的有效性与实用性，还进行了对比实验，共使用了五位患者的颅内血管磁共振图像，其中患者A、B、C、D的增强黑血图像与亮血图像分别为160张，患者E的增强黑血图像与亮血图像分别为150张；同时选取仅使用DICOM图像方位标签信息进行配准的算法，以及基于互信息度量的配准算法，与本发明实施例中的互信息金字塔方法进行对比，其中基于互信息度量的算法是通过多参数优化方法寻找参考图像和浮动图像间的最佳变换，使得两幅图像的互信息值最大，并没有使用到图像金字塔算法。
实验平台是Matlab R2016b。针对实验的图像配准结果，采用定性分析与定量分析相结合。在定性分析方面，由于多模态医学图像间存在较大的灰度差异，将配准图像与参考图像相减得到的差值图像无法有效地反映出多模态医学图像的配准结果，因此本发明实施例通过将配准图像与参考图像进行重叠，获得可以反映出配准图像和参考图像对齐程度的彩色重叠图像，通过彩色重叠图像对多模态配准算法的配准效果进行定性分析，图10显示了多模态颅内血管磁共振图像的配准结果，图10为包含互信息金字塔方法的多种配准方法的颅内血管磁共振图像的配准结果。其中，(a)为参考图像；(b)为浮动图像；(c)为基于图像方位标签信息的重叠图像；(d)为基于互信息度量的重叠图像；(e)为本发明实施例的互信息金字塔方法的重叠图像。其中附图均为原图的灰度图，彩色未示出。在定量分析方面，由于评价指标均方根误差RMSE和峰值信噪比PSNR并不适用于对灰度变化较大的图像进行评价，所以为了更好地对多模态医学图像的配准结果进行评价，采用归一化互相关系数NCC，归一化互信息NMI作为评价指标，当归一化互相关系数NCC和归一化互信息NMI的值越大时，表示图像配准精度越高，表2显示了不同配准算法的评价指标结果分析。
表2不同配准方法的结果分析
患者数据 配准算法 NCC a NMI a
Figure PCTCN2021129608-appb-000005
a中的值是基于患者多幅图像配准的评价指标平均值±均方误差
定性分析：从图10的重叠图像可以明显看出，基于互信息度量的方法出现了较大的配准偏移，分析原因可能是因为仅使用基于互信息度量的方法容易陷入局部最优值，而非全局最优值；基于图像方位标签信息的配准效果表现同样欠佳，图像出现了部分不重叠情况；而互信息金字塔方法的配准图像效果表现良好，图像显示更加清晰，图像几乎完全重叠在一起。
定量分析：由表2可知，从NCC与NMI这两个评价指标来看，相较于只使用DICOM图像的方位标签信息的配准算法，以及基于互信息度量的配准算法，本发明实施例的互信息金字塔方法在配准精度上有所提高，说明本发明实施例提出的基于互信息和图像金字塔的配准方法可以良好地处理多模态颅内血管磁共振图像的配准。
S33，获得所有测试图像对的旋转矩阵的均值；
上一步中，可以获得每个测试图像对中，配准后第一亮血图像对应的旋转矩阵，那么，可以计算获得所有测试图像对的旋转矩阵的均值。
S34，利用旋转矩阵的均值，对除测试图像对之外的、其余预处理后的图像对中的第一亮血图像进行坐标变换，完成图像配准，得到若干个配准后图像对。
考虑到患者在使用磁共振亮血序列扫描时，如果发生了轻微的运动，那么亮血序列扫描得到的颅内血管图像坐标位置就会发生轻微的变化，这时需要将每一张亮血图像都进行空间坐标变换操作，才能保持与增强黑血图像具有相同的坐标位置。基于互信息和图像金字塔的配准方法的配准过程中需要不停地迭代计算每两幅待配准图像的互信息，当图像尺寸和数量较大时，耗时会过大，如果对所有待配准的图像对均使用基于互信息和图像金字塔的配准方法会导致计算速度较慢。而发明人考虑到颅内血管可以看作是一个刚体，它不同于心脏或肺部等器官会随着人的呼吸等运动发生改变，所以每一张亮血图像的空间坐标变换操作几乎都是一致的，即几乎使用的是同一个旋转矩阵。参见表3，表3显示了使用互信息金字塔方法计算得到的旋转矩阵均值与均方误差，源于患者A的160 张400×400增强黑血图像与160张400×400亮血图像，经过互信息金字塔方法的配准之后得到的旋转矩阵，由数据可知旋转矩阵的均方误差很小。那么，可以使用患者的少量几张亮血图像配准计算得到的旋转矩阵，来对所有亮血图像进行相同的空间坐标变换，而无需再对每张亮血图像求旋转矩阵，从而加速图像配准过程。
表3旋转矩阵的均值与均方误差
Figure PCTCN2021129608-appb-000006
该步骤中，利用旋转矩阵的均值，也就是一个新的旋转矩阵，对其余预处理后的图像对中的第一亮血图像进行坐标变换，可以快速完成所有图像的配准，极大地提高了配准速度。关于利用旋转矩阵实现亮血图像的坐标变换、图像插值、使用基于互信息的相似性度量，并采用预定搜索策略实现图像配准的过程可以参见前文，也可以结合相关现有技术理解，在此不再赘述。
在该步骤后，可以得到配准后的若干个亮血图像，这些亮血图像和对应的增强黑血图像在同一坐标系下，且相似度较高。
为验证本发明实施例的基于高斯分布抽样的互信息和图像金字塔的配准方法的可行性，共使用了五位患者的颅内血管磁共振图像进行配准，其中患者A、B、C、D的增强黑血图像与亮血图像分别为160张，患者E的增强黑血图像与亮血图像分别为150张。针对实验的图像配准结果，由于多模态磁共振图像的获取原理不同，呈现信息不同，现阶段仍没有一个统一的金标准来评价哪一种配准算法最好，应该从具体的配准对象以及应用目的对配准结果的好坏进行评价，所以采用定性分析与定量分析相结合。在定性分析方面，通过对能反映出配准图像和参考图像对齐程度的彩色重叠图像，对配准算法结果进行定性分析，图11显示了多模态颅内血管磁共振图像的配准结果对比。图11为本发明实施例的基于高斯分布抽样的互信息和图像金字塔的配准方法与互信息金字塔方法的颅内血管磁共振图像的配准结果。其中(a)为参考图像；(b)为浮动图像；(c)为互信息金字塔方法的重叠图像；(d)为本发明方法的重叠图像，标准差σ取1；(e)为本发明方法的重叠图像，标准差σ取2；(f)为本发明方法的重叠图像，标准差σ取3；(g)为本发明方法的重叠图像，标准差σ取4；(h)为本发明方法的重叠图像，标准差σ取5；(i)为本发明方法的重叠图像，标准差σ取6。各图像为处理后的灰度图，彩色未示出。在定量分析方面，采用归一化互相关系数NCC，归一化互信息NMI作为评价指标，NCC与NMI的值越大时，表示图像配准精度越高。表4为对其中一位患者所有图像对均使用基于互信息和图像金字塔的配准方法(简称为互信息金字塔方法)，以及使用本发明实施例提出的基于高斯分布抽样的互信息和图像金字塔的配准方法(简称为本发明方法)的配准结果(其余患者数据受篇幅限制不予显示)，实验平台是Matlab R2016b。由于本发明方法不需要对所有图像进行配准，只需要随机选取少量几幅图像进行配准即可，因此实验设置高斯分布均值μ为待配准图像总数的一半，标准差σ分别为1，2，3，4，5，6时，随机抽取20幅增强黑血图像与对应的亮血图像进行配准。
表4不同配准算法的结果分析
Figure PCTCN2021129608-appb-000007
Figure PCTCN2021129608-appb-000008
a中的值是基于160幅亮血图像与160幅增强黑血图像配准的评价指标平均值±均方误差
从图11的重叠图像可以明显看出，互信息金字塔方法的配准图像与本发明方法的配准图像效果表现均良好，图像都几乎完全重叠在一起；且由表4可知，从NCC与NMI这两个评价指标来看，虽然本发明方法相较于互信息金字塔方法精度较低，但在不同的高斯分布函数设置下，均没有相差太多。当参与配准的图像尺寸较大，数量较多时，互信息金字塔方法计算量较大。本发明方法针对颅内血管的刚性空间变换特征对算法进行改进，利用变换矩阵的相似性对图像配准加速，实验结果证明改进后的算法配准耗时仅为互信息金字塔方法配准时间的五分之一，可大幅度提高配准速度，可以良好地实现多模态颅内血管磁共振图像的配准。
本发明实施例提供的配准方案中，颅内血管可视为一个刚体，每一张亮血图像的空间坐标变换操作几乎一致，因此可以使用同一个旋转矩阵。因此，一方面选取少量图像对进行亮血图像的图像配准，利用已配准的少量亮血图像的旋转矩阵的均值，对其余图像对的亮血图像进行相同的空间坐标变换，而无需再对其余每张亮血图像求旋转矩阵，因此可以加速图像配准过程。另一方面，在图像配准过程中，基于互信息作为相似性度量，并采用图像金字塔算法增加模型复杂度，可以提高配准精度和速度。相比现有技术中，观察颅内血管的亮血图像和黑血图像需要医生凭着空间想象和主观经验。本发明实施例采用图像配准方法可以将亮血图像和增强黑血图像统一在相同坐标系下，可以方便医生理解黑血序列和亮血序列对应的颅内血管图像，简便、快速地得到诊断所需的综合信息，为后续的医疗诊断、制定手术计划、放射治疗计划等提供准确可靠的参考信息。同时，该图像配准是后续流空伪影消除的重要步骤。通过本发明实施例提供的配准方案，可以给其余医学图像配准提供一种较好的参考方式，具有很大的临床应用价值。同时，本发明实施例的图像配准过程是后续消除流空伪影的重要基础。
在图像配准后，可以对配准后增强黑血图像中的流空伪影进行消除，其中流空伪影出现的原因是在血管壁成像过程中，由于血管太细小，走向迂曲处血液流速较慢，以及周围血液和组织液可能有信号污染等问题，导致在黑血序列扫描得到的图像中，本应该为黑色的血液信息反而表现为亮色，从而模拟正常个体的壁增厚或斑块外观，夸大血管狭窄程度。本发明实施例考虑利用配准后亮血图像中的血液信息，对配准后增强黑血图像中信号显示不正确的血液信息进行修正，将配准后亮血图像中的血液信息嵌入配准后增强黑血图像中，以达到图像融合的效果。具体可以通过以下步骤实现：
S4，利用配准后亮血图像组，对增强黑血图像组中的增强黑血图像进行流空伪影消除操作，得到包括K个目标增强黑血图像的伪影消除增强黑血图像组；
可选的一种实施方式中，S4可以包括S41～S44：
S41，针对每一个配准后亮血图像，提高该配准后亮血图像的对比度，得到对比增强亮血图像；
步骤可选的一种实施方式中，可以根据在亮血图像中血液呈高信号，而周围脑组织呈低信号的特点，对配准后亮血图像进行灰度线性变换，调整图像灰度范围，实现提高图像对比度的目的。
比如，配准后亮血图像所使用的一种具体的灰度线性变换及参数设置如图12所示，图12为本发明实施例所提供的灰度线性变换及参数设置示意图。利用图12所示的灰度线性变换，可以将原来的配准后亮血图像f中较小的灰度值变化区间扩展为新的配准后亮血图像f1(对比增强亮血图像)中的较大灰度值变化区间，调整图像灰度范围，实现提高配准后亮血图像的对比度的目的。通过该步骤，可以得到对比增强亮血图像。参见图13，图13为本发明实施例的灰度线性变换结果图，即配准后亮血图像经过灰度线性变换后的结果图。其中，左图为配准后亮血图像，右图为其经过灰度线性变换后的结果图，可以看到右图中血液部分与周围像素相比对比度明显增强。由于医学图像像素范围较大，可能是-1000～+1000，通过该步骤，可以将像素范围归一化为0～255，成为符合一般图像处理的像素范围，可便于后续处理。关于灰度线性变换的具体过程可以参见相关现有技术，在此不再赘述。
S42，从对比增强亮血图像中提取出血液信息，得到亮血特征图；
可选的一种实施方式中，S42可以包括以下步骤：
S421，利用预设图像二值化方法确定第一阈值；
S422，利用第一阈值，从对比增强亮血图像中提取出血液信息；
该步骤使用的方法称为阈值分割。
S423，由提取出的血液信息得到亮血特征图。
预设图像二值化方法即图像的二值化处理，可以将图像上的点的灰度置为0或255，也就是将整个图像呈现出明显的黑白效果。即将256个亮度等级的灰度图像通过适当的阈值选取而获得仍然可以反映图像整体和局部特征的二值化图像。本发明实施例通过预设图像二值化方法可以将对比增强亮血图像中的血液信息突出显示为白色，将无关信息显示为黑色，以便于提取出血液信息对应的亮血特征图。本发明实施例中的预设图像二值化方法可以包括最大类间方差法OTSU、kittle等等。
血液信息的提取公式如(2)所示，其中T(x,y)为对比增强亮血图像灰度值，F(x,y)为亮血特征图灰度值，T为第一阈值。
Figure PCTCN2021129608-appb-000009
可选的一种实施方式中，采用最大类间方差法OTSU，结果如图14所示，图14为本发明实施例的图像二值化结果图，其中左图为对比增强亮血图像，右图为其经过阈值提取后的血液信息。可以看到，右图中显示为亮色的部分仅为与血液相关的信息。
S43，将亮血特征图与该配准后亮血图像对应的增强黑血图像，依据预设融合公式进行图像融合，得到该增强黑血图像对应的流空伪影消除的目标增强黑血图像；
在该步骤中，首先建立亮血特征图与对应的增强黑血图像之间的空间映射关系，将亮血特征图映射到对应的增强黑血图像中，依据预设融合公式进行图像融合，其中预设融合公式为：
Figure PCTCN2021129608-appb-000010
其中，F(x,y)为亮血特征图的灰度值，R(x,y)为对应的增强黑血图像的灰度值，g(x,y)为融合后的目标增强黑血图像的灰度值。
经过以上操作，可以将对应的增强黑血图像中本应该为黑色，却表现为亮色的流空伪影灰度值 更改为黑色，从而实现消除流空伪影的目的。
参见图15，图15为本发明实施例中针对颅内血管用不同方法得到的流空伪影消除结果，其中箭头所示处出现了流空伪影，左一图为出现了流空伪影的颅内血管增强黑血图像的原图，左二图为使用基于互信息和图像金字塔的配准方法的流空伪影消除方法得到的结果，左三图为使用基于高斯分布抽样的互信息和图像金字塔的配准方法的流空伪影消除方法得到的结果，其中高斯分布标准差σ为3，可以看出本发明实施例所采用的基于高斯分布抽样的互信息和图像金字塔的配准方法的流空伪影消除方法的消除效果与基于互信息和图像金字塔的配准方法的流空伪影消除方法的消除效果都较好。但可以理解的是，本发明实施例所采用的基于高斯分布抽样的互信息和图像金字塔的配准方法的流空伪影消除方法比基于互信息和金字塔的流空伪影消除方法处理速度更快，实验证实可节省约80％的时间。
S44，由K个增强黑血图像对应的目标增强黑血图像，得到伪影消除增强黑血图像组。
所有增强黑血图像均完成流空伪影消除后，可以得到伪影消除增强黑血图像组。
本发明实施例所提供的方案中，对磁共振血管成像技术扫描得到的亮血图像和增强黑血图像进行基于高斯分布抽样的互信息和图像金字塔的配准方法的图像配准，可以提高配准精度和配准速度。通过阈值分割从配准后的亮血图像中提取血液信息，将其融合进配准后的增强黑血图像中，从而对配准后的增强黑血图像中信号显示不正确的血液信息进行修正，将表现为亮色的流空伪影灰度值更改为黑色，从而实现消除流空伪影的目的，得到显示更准确、全面的颅内血管图像。本发明实施例所提供的方案是从图像后处理的角度对流空伪影进行消除，无需使用新的成像技术、成像模式或脉冲序列，因此可以简便、准确、快速地消除流空伪影，并可以在临床应用实现较好的推广。
S5，将伪影消除增强黑血图像组和黑血图像组中对应图像相减，得到K个造影增强图；
每个目标增强黑血图像和对应的黑血图像相减，可以得到具有造影增强效果的造影增强图，当所有目标增强黑血图像和对应的黑血图像均相减后，可以得到K个造影增强图，可以理解的是，这K个造影增强图均是二维图。
S6，利用配准后亮血图像组，建立血液三维模型；
可以理解的是，配准后亮血图像均为二维图，通过对配准后亮血图像进行三维重建，可以将血液信息表现为三维结构，建立出血液三维模型。其中，二维图像通过插值得到具有立体效果的三维模型的过程称为三维重建。目前的三维重建技术包括移动立方体(Marching Cubes,MC)方法、最大密度投影(Maximum Intensity Projection,MIP)方法、表面阴影遮盖方法(SSD)、容积漫游技术(VRT)、曲面重建方法(CPR)、虚拟内镜技术(VE)等等。本发明实施例可以采用任意一种三维重建方法建立血液三维模型。该血液三维模型可以初步模拟三维血管，直观地显示血管的走向及病灶区域等。
可选的一种实施方式中，S6可以包括S61～S63：
S61，获取K个对比增强亮血图像构成的第一三维体数据；
可以获取S41步骤得到的K个对比增强亮血图像。本领域技术人员可以理解的是，K个对比增强亮血图像其实是层叠为一个三维立方体数据的。为了方便区分，本发明实施例中将其命名为第一三维体数据。
S62，利用最大类间方差法计算第一三维体数据中，居中的第二三维体数据对应的第二阈值；
该步骤仍是利用最大类间方差法OTSU确定阈值，但和S421中利用该方法确定第一阈值有所不同，本步骤中是使用最大类间方差法OTSU求出位于第一三维体数据这个大的三维立方体中，靠近中间部分的一个小立方体(称为第二三维体数据)中的多个对比增强亮血图像所对应的一个阈值作为第二 阈值。
因为在对比增强图像中，血液信息基本集中于图像的中部，那么，针对第一三维体数据中，选取居中的小的立方体数据(第二三维体数据)确定第二阈值，可以减小阈值计算量，提高计算速度，且该第二阈值准确适用于第一三维体数据中所有血液信息。
对于第二三维体数据的大小，可以首先确定第一三维体数据的中心点，然后以预设边长在立方体对应的六个方向延伸，从而确定第二三维体数据的大小，其中，预设边长可以根据包含Willis环的经验值确定，比如为第一三维体数据这个立方体的边长的1/4等。其中Willis环是颅内最重要的侧支循环途径，将两侧半球和前、后循环联系起来。
S63，将第二阈值作为移动立方体方法的输入阈值，利用移动立方体方法对第一三维体数据进行处理，得到血液三维模型。
如前，移动立方体方法(简称MC)是一种三维重建方法，可以依据给定的输入阈值，对第一三维体数据进行处理，直接得到血液三维模型。
移动立方体方法相较于其他的面绘制算法，具有生成网格质量好的优点。关于移动立方体方法对第一三维体数据的具体处理过程，请参见相关的现有技术，在此不做赘述。
具体结果参见图16(a)，图16(a)为本发明实施例针对颅内血管的血液三维模型效果图。
S7，利用配准后亮血图像组建立血液边界扩展的血管三维模型；
S6步骤中得到的是血液三维模型，其表征的其实是颅内血液的流向和区域分布，由于实际中血液外围存在有血管壁，因此血液三维模型其实并不能完全代表真实的颅内血管情况。
因此，在S7步骤中，可以对配准后亮血图像中的血液边界进行扩展，使得扩展后的血液边界能够涵盖颅内血管壁的范围，形成一个中空管的效果，再对扩展血液边界后的二维图像利用三维重建方法生成三维模型，进而得到比S5中的血液三维模型更接近真实颅内血管情况的血管三维模型。
关于血液边界的扩展可以通过检测配准后亮血图像中血液边界像素点，将检测到的像素点向预设方向扩展预设个像素点实现，预设个像素点可以根据大量颅内血管直径及血管壁厚度数据所得到的经验值来选取。当然，本发明实施例中扩展血液边界的方式不限于此。
可选的一种实施方式中，S7可以包括S71～S75：
S71，获取K个亮血特征图；
即获取S42步骤得到的K个亮血特征图。
S72，针对每个亮血特征图，利用膨胀操作扩大该亮血特征图中血液的边界，得到该亮血特征图对应的扩展亮血特征图；
膨胀操作是形态学运算的一种，膨胀操作可以填充图像中的空洞，并使物体位于边缘的凸出点向外扩张，最终膨胀后的物体比原先的面积更大。膨胀运算可以记为
Figure PCTCN2021129608-appb-000011
定义为
Figure PCTCN2021129608-appb-000012
其中B为结构元素，A为原图。这里的原图A是亮血特征图，亮血特征图中仅有0和255两种像素值，0对应黑色，255对应白色。
结构元素也称为内核(简称为kernel)，内核可视为一个卷积核。膨胀操作就是利用这个卷积核B对原图A进行卷积运算求局部最大值，卷积核B通常有个锚点，通常位于卷积核的中央位置。随着卷积核扫描原图A，计算叠加区域的最大像素值，并将锚点的位置用最大值替换。也就是最大化操作导致图片中亮的区域增长(所以叫做膨胀)。简单来说就是利用卷积核在原图上进行从左到右，从上到下的平移，如果卷积核对应的框中存在白色，那么这个框内所有的颜色都变为白色。
内核可以为矩形、椭圆、圆形。具体可以在OpenCV的函数cv2.getStructuringElement()中，通 过传递内核的形状和大小，即可获得所需的内核。
一种可选的实施方式中，可以利用半径为1的圆形内核对亮血特征图进行多步膨胀，直到达到最大梯度位置停止膨胀，从而确定血管外壁边界，实现血管壁的分割，得到亮血特征图对应的扩展亮血特征图。由于血管壁紧贴血液，且管壁极薄，假设膨胀后的范围就是血管壁的所在范围，这步操作即可将血液附近的血管壁所在区域包括进来，作为血管壁造影增强特性的搜索范围。关于膨胀操作的具体实施过程可以参见相关现有技术，在此不再赘述。
S73，将该亮血特征图对应的扩展亮血特征图与该亮血特征图求差，得到该亮血特征图对应的差值特征图；
该步骤针对每个亮血特征图得到的差值特征图是一个类似于中空血管的二维平面图。同样的，该差值特征图的像素值也仅有0和255。
S74，确定第三阈值；
该步骤可以根据经验值为所有差值特征图选定一个像素值作为第三阈值，比如可以选取100～200之间的任意一个值，如128作为第三阈值。
S75，将第三阈值作为移动立方体方法的输入阈值，利用移动立方体方法对K个差值特征图进行处理，得到血液边界扩展的血管三维模型。
移动立方体方法利用第三阈值作为输入阈值，可以由K个差值特征图得到血液边界扩展的血管三维模型。关于移动立方体方法的具体实施过程在此不再赘述。
具体结果参见图16(b)，图16(b)为本发明实施例针对颅内血管的血管三维模型效果图。其中图像进行了灰度处理，在实际中，可以以蓝色等颜色进行显示。
S8，利用K个造影增强图建立造影增强三维模型；
该步骤可以利用移动立方体方法实现，具体参见S6和S7，在此不再赘述。
具体结果参见图16(c)，图16(c)为本发明实施例针对颅内血管的造影增强三维模型效果图。其中图像进行了灰度处理，在实际中，可以以红色等颜色进行显示。
S9，基于血液三维模型、血管三维模型、造影增强三维模型，得到血管造影增强三维模型。
可选的一种实施方式中，S9可以包括以下步骤：
S91，保留造影增强三维模型中与血管三维模型的重叠部分，得到保留后造影增强三维模型；
由于S8得到的造影增强三维模型并非只包含了血管的造影增强，需要排除无关组织的增强特性，因此使用S7得到的血管三维模型中血管壁造影增强特性的搜索范围，来判断S8得到的造影增强图三维模型是否位于血液附近的血管壁区域，即判断造影增强三维模型中是否有与血管三维模型的重叠部分，如果是，则表明重叠部分位于搜索范围之内，则需要保留该重叠部分，因此得到保留后造影增强三维模型。
S92，将保留后造影增强三维模型与血液三维模型融合，得到颅内血管造影增强三维模型。
将表征血管造影增强的保留后造影增强三维模型，与表征血液信息的血液三维模型进行融合，可以直观显示有明显造影增强的血管壁，可以清晰地看到颅内血管的哪个部位范围内造影增强效果最为明显，那么该区域可能出现粥样硬化或易损性斑块。
可选的一种实施方式中，颅内血管造影增强三维模型中可以获得造影增强定量分析，具体的，可以针对颅内血管造影增强三维模型中血管壁上任意一个点，得到斑块强化指数CE，其中CE定义为：
Figure PCTCN2021129608-appb-000013
其中，S preBBMR和S postBBMR分别为黑血图像和造影增强黑血图像中的信号强度。
本领域技术人员可以理解的是，S preBBMR和S postBBMR分别是拍摄黑血图像和造影增强黑血图像后，图像中携带的信息。本发明实施例利用上述信息得到颅内血管壁边沿各个点的斑块强化指数CE，并将其体现在颅内血管造影增强三维模型中，可以方便医生获取更为详细的血管信息，具体的，当CE大于一个斑块阈值，比如0.5时，表示该处血管壁上出现了斑块，因此，通过测量血管壁区域的斑块强化指数，有助于鉴别责任颅内动脉斑块等，可以提供有价值的诊断辅助信息。关于两个三维模型的融合技术可以采用现有技术实现，在此不做赘述。具体结果参见图17，图17为本发明实施例针对颅内血管的血管造影增强三维模型效果图。其中图像进行了灰度处理。在实际中，图17中可以用不同颜色进行区分，比如蓝色为没有出现造影增强的血管部位，红色处为出现了造影增强的血管部位。说明书附图中白色线圈内的亮色部分为出现了造影增强的颅内血管部位，即该处可能出现了颅内动脉粥样硬化的病症或者易损性斑块，其余部分为没有出现造影增强的血管部位，并且该血管造影增强三维模型可以实现旋转、放大缩小等基本功能，从而辅助医生定位病灶区域，作出更精准的判断。
本发明实施例所提供的方案中，首先对磁共振血管成像技术扫描得到的亮血图像和增强黑血图像采用基于高斯分布抽样的互信息和图像金字塔的配准方法进行图像配准，可提高配准效率，使图像从低分辨率到高分辨率逐层提高配准精度。通过上述图像配准可以将亮血图像和增强黑血图像统一在相同坐标系下。其次利用配准后亮血图像对增强黑血图像进行流空伪影消除操作，可以显示更准确、全面的血管信息。本发明实施例所提供的方案是从图像后处理的角度对流空伪影进行消除，无需使用新的成像技术、成像模式或脉冲序列，因此可以简便、准确、快速地消除流空伪影，并可以在临床应用实现较好的推广。再次，利用配准后亮血图像建立血液三维模型、血液边界扩展的血管三维模型，通过将伪影消除增强黑血图像和黑血图像相减得到具有造影增强效果的造影增强三维模型；最后基于血液三维模型、血管三维模型、造影增强三维模型，得到具有造影增强效果的血管壁对应的血管造影增强三维模型。该血管造影增强三维模型实现了颅内血管的三维可视化，无需医生通过想象力还原颅内血管组织结构及病症特征等，能够方便医生从任意感兴趣的角度、层次观察和分析血管形态特征，可以提供具有真实感的血管三维空间信息，便于直观显示有明显造影增强的血管壁，便于定位与显示病灶区域。能够在临床应用上，简便快速地获得颅内血管的真实信息，以进行血管病变分析。
注：本发明实施例中的患者实验数据均来源于陕西省人民医院，图像可用作一般的科学研究。以上仅为本发明的较佳实施例而已，并非用于限定本发明的保护范围。凡在本发明的精神和原则之内所作的任何修改、等同替换、改进等，均包含在本发明的保护范围内。

Claims (10)

  1. 一种颅内血管造影增强三维模型的建立方法，其特征在于，包括：
    获取颅内血管部位的亮血图像组、黑血图像组和增强黑血图像组；其中，所述亮血图像组、所述黑血图像组、所述增强黑血图像组分别包括K个亮血图像、黑血图像和增强黑血图像；所述亮血图像组、所述黑血图像组、所述增强黑血图像组中的图像一一对应；K为大于2的自然数；
    将每个亮血图像和对应的增强黑血图像作为一个图像对，对每个图像对进行预处理，得到该图像对的第一亮血图像和第一黑血图像；
    针对每个第一亮血图像，以对应的第一黑血图像为基准，利用基于高斯分布抽样的互信息和图像金字塔的配准方法进行图像配准，得到包括K个配准后亮血图像的配准后亮血图像组；
    利用所述配准后亮血图像组，对所述增强黑血图像组中的增强黑血图像进行流空伪影消除操作，得到包括K个目标增强黑血图像的伪影消除增强黑血图像组；
    将所述伪影消除增强黑血图像组和所述黑血图像组中对应图像相减，得到K个造影增强图；
    利用所述配准后亮血图像组，建立血液三维模型；
    利用所述配准后亮血图像组建立血液边界扩展的血管三维模型；
    利用所述K个造影增强图建立造影增强三维模型；
    基于所述血液三维模型、所述血管三维模型、所述造影增强三维模型，得到颅内血管造影增强三维模型。
  2. 根据权利要求1所述的方法，其特征在于，所述对每个图像对进行预处理，得到该图像对的第一亮血图像和第一黑血图像，包括：
    针对每个图像对，以所述增强黑血图像为基准，将所述亮血图像进行坐标变换和图像插值，使用基于互信息的相似性度量，并采用预定搜索策略，得到预配准后的第一亮血图像；
    从所述增强黑血图像中，提取与所述第一亮血图像的扫描范围相同的区域内容，形成第一黑血图像。
  3. 根据权利要求1或2所述的方法，其特征在于，所述针对每个第一亮血图像，以对应的第一黑血图像为基准，利用基于高斯分布抽样的互信息和图像金字塔的配准方法进行图像配准，得到包括K个配准后亮血图像的配准后亮血图像组，包括：
    采用高斯分布抽样选取预处理后的部分图像对作为测试图像对；
    对每个测试图像对中的所述第一亮血图像和所述第一黑血图像，采用基于互信息和图像金字塔的配准方法进行图像配准，得到配准后该测试图像对中所述第一亮血图像对应的旋转矩阵；
    获得所有测试图像对的旋转矩阵的均值；
    利用所述旋转矩阵的均值，对除所述测试图像对之外的、其余预处理后的图像对中的所述第一亮血图像进行坐标变换，完成图像配准，得到包括K个配准后亮血图像的配准后亮血图像组。
  4. 根据权利要求3所述的方法，其特征在于，所述对每个测试图像对中的所述第一亮血图像和所述第一黑血图像，采用基于互信息和图像金字塔的配准方法进行图像配准，得到配准后该测试图像对中所述第一亮血图像对应的旋转矩阵，包括：
    针对每个测试图像对，基于下采样处理，由所述第一亮血图像得到亮血高斯金字塔，由所述第 一黑血图像得到黑血高斯金字塔；其中，所述亮血高斯金字塔和所述黑血高斯金字塔中包括从下至上分辨率依次变小的m个图像；m为大于3的自然数；
    基于上采样处理，利用所述亮血高斯金字塔得到亮血拉普拉斯金字塔，利用所述黑血高斯金字塔得到黑血拉普拉斯金字塔；其中，所述亮血拉普拉斯金字塔和所述黑血拉普拉斯金字塔中包括从下至上分辨率依次变小的m-1个图像；
    对所述亮血拉普拉斯金字塔和所述黑血拉普拉斯金字塔中对应层的图像进行配准，得到配准的亮血拉普拉斯金字塔；
    利用所述配准的亮血拉普拉斯金字塔作为叠加信息，对所述亮血高斯金字塔和所述黑血高斯金字塔中各层图像进行自上而下的配准，得到配准的亮血高斯金字塔，并得到配准后该测试图像对中所述第一亮血图像对应的旋转矩阵。
  5. 根据权利要求4所述的方法，其特征在于，所述利用所述配准的亮血拉普拉斯金字塔作为叠加信息，对所述亮血高斯金字塔和所述黑血高斯金字塔中各层图像进行自上而下的配准，得到配准的亮血高斯金字塔，包括：
    对所述亮血高斯金字塔和所述黑血高斯金字塔中自上而下的第j层，将该层对应的黑血高斯图像作为参考图像，将该层对应的亮血高斯图像作为浮动图像，使用基于互信息的相似性度量，并采用预定搜索策略，实现图像配准，得到配准后的第j层亮血高斯图像；
    将所述配准后的第j层亮血高斯图像进行上采样操作，并与配准后的对应层亮血拉普拉斯图像相加，利用相加后的图像替换所述亮血高斯金字塔中第j+1层的亮血高斯图像；
    将第j+1层的黑血高斯图像作为参考图像，将替换后的第j+1层的亮血高斯图像作为浮动图像，使用预定相似性度量及预定搜索策略，实现图像配准，得到配准后的第j+1层亮血高斯图像；
    其中j＝1，2，…，m-1，所述黑血高斯图像为所述黑血高斯金字塔中的图像，所述亮血高斯图像为所述亮血高斯金字塔中的图像。
  6. 根据权利要求1或5所述的方法，其特征在于，所述利用所述配准后亮血图像组，对所述增强黑血图像组中的增强黑血图像进行流空伪影消除操作，得到包括K个目标增强黑血图像的伪影消除增强黑血图像组，包括：
    针对每一个配准后亮血图像，提高该配准后亮血图像的对比度，得到对比增强亮血图像；
    从所述对比增强亮血图像中提取出血液信息，得到亮血特征图；
    将所述亮血特征图与该配准后亮血图像对应的增强黑血图像，依据预设融合公式进行图像融合，得到该增强黑血图像对应的流空伪影消除的目标增强黑血图像；
    由K个增强黑血图像对应的目标增强黑血图像，得到伪影消除增强黑血图像组。
  7. 根据权利要求6所述的方法，其特征在于，所述从所述对比增强亮血图像中提取出血液信息，得到亮血特征图，包括：
    利用预设图像二值化方法确定第一阈值；
    利用所述第一阈值，从所述对比增强亮血图像中提取出血液信息；
    由提取出的所述血液信息得到亮血特征图。
  8. 根据权利要求7所述的方法，其特征在于，所述利用所述配准后亮血图像组，建立血液三维模型，包括：
    获取K个对比增强亮血图像构成的第一三维体数据；
    利用最大类间方差法计算所述第一三维体数据中，居中的第二三维体数据对应的第二阈值；
    将所述第二阈值作为移动立方体方法的输入阈值，利用所述移动立方体方法对所述第一三维体数据进行处理，得到血液三维模型。
  9. 根据权利要求8所述的方法，其特征在于，所述利用所述配准后亮血图像组建立血液边界扩展的血管三维模型，包括：
    获取K个所述亮血特征图；
    针对每个亮血特征图，利用膨胀操作扩大该亮血特征图中血液的边界，得到该亮血特征图对应的扩展亮血特征图；
    将该亮血特征图对应的扩展亮血特征图与该亮血特征图求差，得到该亮血特征图对应的差值特征图；
    确定第三阈值；
    将所述第三阈值作为移动立方体方法的输入阈值，利用所述移动立方体方法对K个所述差值特征图进行处理，得到血液边界扩展的血管三维模型。
  10. 根据权利要求1或9所述的方法，其特征在于，所述基于所述血液三维模型、所述血管三维模型、所述造影增强三维模型，得到血管造影增强三维模型，包括：
    保留所述造影增强三维模型中与所述血管三维模型的重叠部分，得到保留后造影增强三维模型；
    将所述保留后造影增强三维模型与所述血液三维模型融合，得到血管造影增强三维模型。
PCT/CN2021/129608 2020-11-23 2021-11-09 一种颅内血管造影增强三维模型的建立方法 WO2022105647A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/552,282 US12112489B2 (en) 2020-11-23 2021-12-15 Method of establishing an enhanced three-dimensional model of intracranial angiography

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011324154.8A CN112634197A (zh) 2020-11-23 2020-11-23 一种颅内血管造影增强三维模型的建立方法
CN202011324154.8 2020-11-23

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/552,282 Continuation US12112489B2 (en) 2020-11-23 2021-12-15 Method of establishing an enhanced three-dimensional model of intracranial angiography

Publications (1)

Publication Number Publication Date
WO2022105647A1 true WO2022105647A1 (zh) 2022-05-27

Family

ID=75303692

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/129608 WO2022105647A1 (zh) 2020-11-23 2021-11-09 一种颅内血管造影增强三维模型的建立方法

Country Status (2)

Country Link
CN (1) CN112634197A (zh)
WO (1) WO2022105647A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116309922A (zh) * 2023-05-22 2023-06-23 杭州脉流科技有限公司 Ct灌注影像的去伪影方法、装置、设备和存储介质
CN116863146A (zh) * 2023-06-09 2023-10-10 强联智创（北京）科技有限公司 用于对血管瘤特征进行提取的方法、设备及存储介质
CN116942104A (zh) * 2023-09-21 2023-10-27 首都医科大学附属北京儿童医院 一种用于测量在体局部循环血管的智能观测方法
CN117830308A (zh) * 2024-03-05 2024-04-05 西安交通大学医学院第一附属医院 一种介入手术前后血管造影智能对比分析方法
CN118298069A (zh) * 2024-06-06 2024-07-05 烟台大学 一种pet合成图的获取方法、系统、设备和存储介质

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112634197A (zh) * 2020-11-23 2021-04-09 西安科锐盛创新科技有限公司 一种颅内血管造影增强三维模型的建立方法
CN113222987B (zh) * 2021-06-04 2022-10-25 清华大学 磁共振成像血管壁增强强度映射方法、装置、设备和介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103110420A (zh) * 2013-01-31 2013-05-22 深圳先进技术研究院 磁共振血管成像方法和系统
CN103635825A (zh) * 2011-04-21 2014-03-12 皇家飞利浦有限公司 交错黑血和亮血动态对比增强（dce）mri
US20150323636A1 (en) * 2012-12-14 2015-11-12 Koninklijke Philips N.V. Interleaved black and bright blood imaging for interleaved dynamic contrast enhanced magnetic resonance imaging
CN111681226A (zh) * 2020-06-09 2020-09-18 上海联影医疗科技有限公司 基于血管识别的目标组织定位方法和装置
CN112634197A (zh) * 2020-11-23 2021-04-09 西安科锐盛创新科技有限公司 一种颅内血管造影增强三维模型的建立方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103635825A (zh) * 2011-04-21 2014-03-12 皇家飞利浦有限公司 交错黑血和亮血动态对比增强（dce）mri
US20150323636A1 (en) * 2012-12-14 2015-11-12 Koninklijke Philips N.V. Interleaved black and bright blood imaging for interleaved dynamic contrast enhanced magnetic resonance imaging
CN103110420A (zh) * 2013-01-31 2013-05-22 深圳先进技术研究院 磁共振血管成像方法和系统
CN111681226A (zh) * 2020-06-09 2020-09-18 上海联影医疗科技有限公司 基于血管识别的目标组织定位方法和装置
CN112634197A (zh) * 2020-11-23 2021-04-09 西安科锐盛创新科技有限公司 一种颅内血管造影增强三维模型的建立方法

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116309922A (zh) * 2023-05-22 2023-06-23 杭州脉流科技有限公司 Ct灌注影像的去伪影方法、装置、设备和存储介质
CN116309922B (zh) * 2023-05-22 2023-08-11 杭州脉流科技有限公司 Ct灌注影像的去伪影方法、装置、设备和存储介质
CN116863146A (zh) * 2023-06-09 2023-10-10 强联智创（北京）科技有限公司 用于对血管瘤特征进行提取的方法、设备及存储介质
CN116863146B (zh) * 2023-06-09 2024-03-08 强联智创（北京）科技有限公司 用于对血管瘤特征进行提取的方法、设备及存储介质
CN116942104A (zh) * 2023-09-21 2023-10-27 首都医科大学附属北京儿童医院 一种用于测量在体局部循环血管的智能观测方法
CN116942104B (zh) * 2023-09-21 2024-01-02 首都医科大学附属北京儿童医院 一种用于测量在体局部循环血管的智能观测方法
CN117830308A (zh) * 2024-03-05 2024-04-05 西安交通大学医学院第一附属医院 一种介入手术前后血管造影智能对比分析方法
CN117830308B (zh) * 2024-03-05 2024-05-10 西安交通大学医学院第一附属医院 一种介入手术前后血管造影智能对比分析方法
CN118298069A (zh) * 2024-06-06 2024-07-05 烟台大学 一种pet合成图的获取方法、系统、设备和存储介质

Also Published As

Publication number Publication date
CN112634197A (zh) 2021-04-09

Similar Documents

Publication Publication Date Title
WO2022105647A1 (zh) 一种颅内血管造影增强三维模型的建立方法
WO2022105623A1 (zh) 一种基于迁移学习的颅内血管病灶识别方法
CN112562058B (zh) 一种基于迁移学习的颅内血管模拟三维模型快速建立方法
US11830193B2 (en) Recognition method of intracranial vascular lesions based on transfer learning
CN114170152A (zh) 一种模拟化三维颅内血管狭窄分析模型的建立方法
CN112598619A (zh) 基于迁移学习的颅内血管模拟三维狭窄化模型的建立方法
CN112634196A (zh) 一种基于智慧医疗的医学图像分割与显示方法
CN112508873A (zh) 基于迁移学习的颅内血管模拟三维狭窄化模型的建立方法
CN112509075A (zh) 基于智慧医疗的颅内血管病灶标记与三维显示方法
CN112509079A (zh) 颅内血管造影增强三维狭窄化分析模型的建立方法
CN114170151A (zh) 一种基于迁移学习的颅内血管病灶识别方法
CN112508869A (zh) 颅内血管图像融合方法和计算机可读存储介质
CN112508868A (zh) 一种颅内血管综合图像生成方法
US12112489B2 (en) Method of establishing an enhanced three-dimensional model of intracranial angiography
CN112509076A (zh) 基于智慧医疗的颅内血管病灶标记与三维显示系统
CN114240841A (zh) 一种模拟化三维血管狭窄分析模型的建立方法
CN112509077A (zh) 一种基于智慧医疗的颅内血管图像分割与显示方法
CN112508877A (zh) 一种血管综合图像生成方法
CN112508881A (zh) 一种颅内血管图像配准方法
CN112634386A (zh) 一种血管造影增强三维狭窄化分析模型的建立方法
CN114170337A (zh) 一种基于迁移学习的颅内血管增强三维模型的建立方法
CN112509080A (zh) 一种基于迁移学习的颅内血管模拟三维模型的建立方法
CN112669256B (zh) 一种基于迁移学习的医学图像分割与显示方法
CN112509081A (zh) 颅内血管造影增强三维狭窄化分析模型的建立方法
CN112508874A (zh) 基于智慧医疗的脑血管病变标记与三维显示系统

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21893794

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21893794

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 21893794

Country of ref document: EP

Kind code of ref document: A1