An Automatic Curved Surface Reconstruction Method for Vascular Wall Images Based on Centerline Extraction

Abstract

The present application relates to an automatic curved surface reconstruction method for vascular wall images based on centerline extraction, which comprises the following steps: obtaining the coordinates of each point of the vascular centerline by using a vascular centerline extraction method and fitting; calculating the normal plane passing through each vascular center point based on the fitting result, and respectively establishing a spatial coordinate system based on each normal plane, wherein the z-axis is the direction of the normal line of the normal plane; sampling the vascular wall image based on the normal plane to obtain slices; stacking the slices along the z-axis in the order of each point on the vascular centerline to obtain the result of curved surface reconstruction. The present application has the advantages of good vascular wall reconstruction effect, and the blood vessel can be straightened through curved surface reconstruction to reflect the overall characteristics of the blood vessel.

Description

Technical Field

[0001] This application relates to the field of graphic reconstruction, and in particular to an automatic curved surface reconstruction method for vascular wall images based on centerline extraction. Background Art

[0002] Stroke is the disease with the highest mortality and disability rates in China and has become a serious burden on medical expenditures and social families in China. Stroke includes two types: hemorrhagic stroke and ischemic stroke. Ischemic stroke is the main type in China, accounting for 79% of all stroke cases, and there is an increasing trend. Research shows that the rupture of atherosclerotic plaques leading to thrombus formation is the main pathogenesis of ischemic stroke. Among them, 48% of the culprit plaques come from intracranial arteries, 30% come from carotid arteries, and the remaining 22% mainly come from the heart and thoracic aorta. Therefore, timely detection of stroke-related vascular beds, including vulnerable plaques or other wall lesions in intracranial arteries, carotid arteries, and thoracic aorta, is the key to early prevention and precise treatment of ischemic stroke.

[0003] In order to perform automated analysis tasks on stroke-related vascular beds, three-dimensional reconstruction of the relevant vascular walls is required. Due to the curved shape of arteries, especially the tortuous course of intracranial arteries, the entire vascular wall cannot be displayed in the same plane. Therefore, a good method is to extract the vascular centerline coordinates from MR images, use this centerline as a map to identify the positions of cross-sectional images of the vessels, and obtain a series of consecutive slices with the vessels centered, for tasks such as vascular wall segmentation.

[0004] In related technologies, bright blood MR images are usually used to extract the vascular centerline for vascular reconstruction. Bright blood MR images include 3D-TOF MRA, which makes the blood flow appear as bright high signals and is mainly used for screening intracranial vascular stenosis. Since the overly high blood flow signals in bright blood MR images make it difficult to distinguish the outer wall of the blood vessels from other tissues, the vascular walls of the corresponding dark blood MR images usually need to be reconstructed using the extracted vascular centerline. Dark blood MR images refer to MR images obtained by suppressing the flowing blood signals using the characteristics of blood flow and the static vascular wall, making the blood flow appear as low signals, the vessel walls and plaques appear as relatively high signals, and enhancing the contrast of the inner wall of the blood vessels. However, since the blood flow signals are suppressed in dark blood MR images, it is not easy to directly extract the vascular centerline.

[0005] The inventor believes that since the imaging time of a single magnetic resonance sequence is usually about several minutes and can achieve molecular-level resolution, it is easily affected by various factors and generates errors, such as vascular pulsation, muscle movement, human movement, sudden changes in blood flow, etc., and often causes the generation of image artifacts in a single imaging. The above-mentioned black-blood MR and bright-blood MR are imaged independently twice, and it is very easy to have a problem of non-registration when the vascular centerline obtained by the above technology is applied to the black-blood image, especially in the intracranial arteries, especially in the intracranial perforating arteries. Summary of the Invention

[0006] In order to improve the effect of vascular wall reconstruction, the present application provides an automatic curved surface reconstruction method for vascular wall images based on centerline extraction.

[0007] In a first aspect, a method for extracting a vascular centerline for three-dimensional reconstruction of a vascular wall in a black-blood MR image provided by the present application adopts the following technical solutions:

[0008] A method for extracting a vascular centerline for three-dimensional reconstruction of a vascular wall in a black-blood MR image includes the following steps:

[0009] Obtain the image to be detected; wherein, the image to be detected is a black-blood MR image;

[0010] Obtain the first vascular feature from the image to be detected based on the first object detection model, wherein the first object detection model is trained to extract features from a two-dimensional image, and the first vascular feature is used to roughly correspond to the position of the vascular centerline on the image to be detected;

[0011] Superimpose the first vascular feature on the image to be detected and form two-channel data;

[0012] Send the image to be detected superimposed with the first vascular feature into a three-dimensional image segmentation model to obtain the second vascular feature, wherein the three-dimensional image segmentation model is used to extract three-dimensional feature information based on a three-dimensional convolution kernel, and the second vascular feature is used to finely correspond to the three-dimensional spatial position of the vascular centerline on the image to be detected;

[0013] Obtain the vascular centerline coordinates based on the second object detection model and the second vascular feature.

[0014] By adopting the above technical solution, the black blood MR image is a three-dimensional image, which is equivalent to being composed of a series of continuous two-dimensional images. The target detection model samples the image to be detected into multiple two-dimensional images and processes them separately. The target detection model removes the background from the image to be detected to initially obtain the blood vessel region, and calculates the blood vessel centerline based on the blood vessel region and uses it as the first blood vessel feature. Since the target detection model can only extract the features of two-dimensional images, the detection results show that there are many discontinuous blood vessels and isolated data. Therefore, a relatively rough blood vessel centerline is obtained in this step. In addition, since the target detection model can only extract the features of two-dimensional images, it generates fewer intermediate parameters and has the advantages of low computing cost and fast processing speed.

[0015] Then, the first blood vessel feature is input into the image to be detected to form two-channel data, so that the three-dimensional image segmentation model can perform feature recognition to extract the second blood vessel feature. Since the processing by the target detection model provides prior conditions, the three-dimensional image segmentation model uses three-dimensional convolution, which can effectively utilize three-dimensional features and ensure the continuity of the blood vessel prediction results. However, three-dimensional convolution requires higher computing power, so it is used to optimize the rough segmentation result output by the target detection model.

[0016] Finally, although the blood vessel wall contour output by the three-dimensional image segmentation model is fine enough, it cannot provide sufficient information (excluding coordinates) for making blood vessel slices. Therefore, another target detection model needs to be connected to output the coordinates of the blood vessel slices.

[0017] In summary, by continuously processing the black blood MR image through three models, the continuous coordinates of the blood vessel centerline in space can be accurately obtained for subsequent blood vessel wall image reconstruction using the black blood MR image, without the need for registration of MR images of different sequences, avoiding the errors caused thereby; at the same time, the performance of subsequent three-dimensional blood vessel wall reconstruction work can be improved.

[0018] Optionally, the first blood vessel feature is N rectangular frames determined by six dimensions [N, X, Y, W, H, C], where N is the Nth target appearing in the image to be detected, X is the x-axis coordinate of the predicted target center, Y is the y-axis coordinate of the predicted target center, W is the width of the predicted target, H is the height of the predicted target, and C is the category of the recognized target.

[0019] Optionally, the target detection model is a Yolo model, an SSD model, a Mask-R-CNN model or a Fast-R-CNN model.

[0020] By adopting the above technical solutions, the Yolo model, SSD model, Mask-R-CNN model or Fast-R-CNN model are all target detection models for 2D images, which are used for target recognition. They are characterized by lower computing cost and faster processing speed.

[0021] Optionally, the three-dimensional image segmentation model is a V-net model.

[0022] By adopting the above technical solution, the V-net model uses three-dimensional convolution, which can effectively utilize three-dimensional features and ensure the continuity of blood vessel prediction results. However, three-dimensional convolution requires high computing power, so it is used to optimize the rough segmentation results output by Yolo.

[0023] In the second aspect, the present application provides a method for automatic surface reconstruction of vascular wall images based on centerline extraction, which adopts the following technical solution:

[0024] A method for automatic surface reconstruction of a blood vessel wall image based on centerline extraction comprises the following steps:

[0025] Based on the above-mentioned blood vessel centerline extraction method, the coordinates of the blood vessel center point are obtained and fitted;

[0026] Based on the fitting results, the normal plane passing through the center point of each blood vessel is calculated, and a spatial coordinate system is established based on each normal plane, wherein the z axis is the normal direction to the normal plane;

[0027] The vessel wall image is sampled based on the normal plane to obtain slices;

[0028] The slices are superimposed along the z-axis based on the order of the points on the blood vessel centerline to obtain the result of the surface reconstruction.

[0029] By adopting the above technical solution, the blood vessel centerline output by the deep learning model is composed of a series of (three-dimensional) points, so it needs to be fitted into a straight line to characterize the direction of the blood vessel. Due to the tortuous course of the blood vessel, it is difficult to show its full picture on a single section, and even multiple end faces will be generated. The blood vessel can be straightened through surface reconstruction to reflect the overall characteristics of the blood vessel. Therefore, the slices on the normal plane are obtained using the blood vessel center point and the blood vessel centerline, and then the slices are superimposed along the z-axis to complete the purpose of surface reconstruction and straightening the blood vessel.

[0030] Optionally, the method for fitting the coordinates of each point on the blood vessel centerline is B-spline fitting.

[0031] Optionally, before sampling the blood vessel wall image based on the normal plane to obtain a slice, the following steps are also included:

[0032] The sampling range was selected based on the vessel type.

[0033] By adopting the above technical solution, on the premise of not losing the pixels of the blood vessel wall, a sampling range as small as possible is selected, which can effectively reduce the amount of data processing and make the result of surface reconstruction more intuitive.

[0034] In a third aspect, an electronic device provided by the present application adopts the following technical solution:

[0035] A blood vessel centerline extraction system for three-dimensional reconstruction of the blood vessel wall in black blood MR images, comprising:

[0036] An input module for acquiring the image to be detected; wherein, the image to be detected is a black blood MR image;

[0037] A first detection module for obtaining a first blood vessel feature from the image to be detected based on the Yolo model, wherein the first blood vessel feature is used to roughly correspond to the position of the blood vessel centerline on the image to be detected;

[0038] A combination module for superimposing the first blood vessel feature on the image to be detected and forming two-channel data;

[0039] A second detection module for sending the image to be detected superimposed with the first blood vessel feature into the V-net model to obtain a second blood vessel feature, wherein the second blood vessel feature is used to finely correspond to the position of the blood vessel centerline on the image to be detected;

[0040] A third detection module for obtaining the blood vessel centerline coordinates based on the Yolo network and the second blood vessel feature.

[0041] In a fourth aspect, an electronic device provided by the present application adopts the following technical solution:

[0042] An automatic surface reconstruction system for blood vessel wall images based on centerline extraction, comprising:

[0043] The blood vessel centerline extraction system as described above, for obtaining the coordinates of each point of the blood vessel centerline;

[0044] A fitting module for fitting the coordinates of each point of the blood vessel centerline;

[0045] A modeling module for calculating the normal planes passing through the center points of each blood vessel based on the fitting result, and respectively establishing a spatial coordinate system based on each normal plane, wherein the z-axis is the normal direction of the normal plane;

[0046] A sampling module for sampling the blood vessel wall image based on the normal plane to obtain slices;

[0047] A reconstruction module for stacking the slices along the z-axis in the order of the points of the blood vessel centerline to obtain the result of surface reconstruction.

[0048] In a fifth aspect, the present application provides an electronic device, which adopts the following technical solution:

[0049] An electronic device comprising:

[0050] one or more processors;

[0051] Memory;

[0052] one or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs being configured to:

[0053] Execute the above-mentioned blood vessel centerline extraction method;

[0054] And / or, executing the above-mentioned automatic curved surface reconstruction method of blood vessel wall image.

[0055] In a sixth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution:

[0056] A computer-readable storage medium stores a computer program that can be loaded by a processor and execute the above method.

[0057] The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement:

[0058] As mentioned above, the blood vessel centerline extraction method;

[0059] And / or, the above-mentioned method for automatic curved surface reconstruction of vascular wall images. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is used to illustrate the steps of a method for automatic surface reconstruction of blood vessel wall images based on centerline extraction in an embodiment of the present application.

[0061] Figure 2 It is used to illustrate the steps of a method for extracting the blood vessel centerline for three-dimensional reconstruction of blood vessel walls in black blood MR images in an embodiment of the present application.

[0062] Figure 3 It is used to show a black blood MR image as an image to be detected.

[0063] Figure 4 It is used to illustrate the first blood vessel feature obtained from the image to be detected using the first target detection model, wherein the box at the cross is used to illustrate the position of the detected left common carotid artery-internal carotid artery bifurcation.

[0064] Figure 5Used to show the first vascular feature obtained from the image to be detected by the first object detection model, where the square at the crossroads is used to indicate the position of the detected bifurcation of the left common carotid artery - external carotid artery.

[0065] Figure 6 Used to show the filling of the first vascular feature in the image to be detected, where the square is used to indicate the bifurcation of the left common carotid artery - internal carotid artery.

[0066] Figure 7 Used to show the filling of the first vascular feature in the image to be detected, where the square is used to indicate the bifurcation of the left common carotid artery - external carotid artery.

[0067] Figure 8 Used to show the image generated by superimposing the first vascular feature on the image to be detected.

[0068] Figure 9 Used to show the image output after the V-net model trains the superimposed image. Detailed implementation manners

[0069] The following further describes the present application in detail with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0070] In this embodiment, the labels of each step are only for convenience of description and do not represent a limitation on the execution order of each step. In actual applications, the execution order of each step can be adjusted as needed or performed simultaneously, and these adjustments or replacements all fall within the protection scope of the present invention.

[0071] In the following description, for the purpose of explanation, many specific details are set forth in order to provide a thorough understanding of the inventive concept. As part of this specification, some of the drawings in the present disclosure represent structures and devices in block diagram form to avoid complicating the disclosed principles. For clarity, not all features of the actual specific implementation need to be described. In addition, the language used in the present disclosure has been mainly selected for readability and guidance purposes and may not have been selected to delimit or define the subject matter of the present invention, so reference is made to the necessary claims to determine such inventive subject matter. References to "a specific implementation" or "specific implementations" in the present disclosure mean that the specific features, structures, or characteristics described in connection with that specific implementation are included in at least one specific implementation, and multiple references to "a specific implementation" or "specific implementations" should not be construed as necessarily all referring to the same specific implementation.

[0072] An embodiment of the present application discloses an automatic curved surface reconstruction method for vascular wall images based on centerline extraction, referring to Figure 1 , including the following steps:

[0073] S1. Obtain the coordinates of the blood vessel center points and perform fitting. Among them, the fitting method can be methods such as B-spline fitting, etc., as long as it can fit multiple points into a continuous curve.

[0074] Magnetic resonance imaging (MRI) uses the principle of nuclear magnetic resonance (NMR). Based on the different attenuations of the released energy in different structural environments within a substance, by detecting the emitted electromagnetic waves with an externally applied gradient magnetic field, the position and type of the atomic nuclei that make up this object can be known, and based on this, the internal structure image of the object can be drawn. The imaging time of a single magnetic resonance sequence is usually about several minutes, and it can reach molecular-level resolution. Therefore, for images obtained from different nuclear magnetic resonances, such as the black-blood MR image and bright-blood MR image of the same object, it is easy to have the situation where the coordinates of the blood vessel center line extracted from the bright-blood MR image do not match the blood vessel pattern in the black-blood MR image. Therefore, in step S1, a blood vessel center line extraction method for three-dimensional reconstruction of the blood vessel wall in black-blood MR images can be used to directly process the black-blood MR image to obtain the blood vessel center line.

[0075] Specifically, referring to Figure 2 This blood vessel center line extraction method for three-dimensional reconstruction of the blood vessel wall in black-blood MR images includes the following steps:

[0076] S11. Obtain the image to be detected; among them, the image to be detected is a black-blood MR image.

[0077] Referring to Figure 3 This image to be detected is an image obtained by vascular wall magnetic resonance imaging (VW MRI). The image to be detected is a two-dimensional RGB image (three-dimensional matrix), that is, [Channel, Y, X], where Channel is an array, and the size of the dimension is determined to be 3. Specifically, the range of the three-dimensional data corresponding to Channel is usually 0 - 65535, which represents the signal intensity during magnetic resonance imaging of the corresponding acceleration.

[0078] S12. Obtain the first blood vessel feature from the image to be detected based on the first object detection model. Among them, the first object detection model is trained to extract features from two-dimensional images, and the first blood vessel feature is used to roughly correspond to the position of the blood vessel center line on the image to be detected.

[0079] In different embodiments, the first object detection model can be a Yolo model, an SSD model, a Mask-R-CNN model, a Fast-R-CNN model, or other models. These models are all object detection models for 2D images, used for object recognition, and they are characterized by low computing costs and fast processing speeds. As an example, in this embodiment, the first object detection model selects the Yolo model.

[0080] After being processed by the Yolo network, the first vascular feature is output, which is the result of the Yolo network. This result is the coordinates of a rectangle tangent to the blood vessel, including the coordinates, width, and height information. Specifically, the first vascular feature is N rectangular frames determined by six dimensions [N, X, Y, W, H, C], and the N objects recognized by the neural network are framed by them. Among them, N is the Nth object that appears in the image to be detected, X is the x-axis coordinate of the predicted object center, Y is the y-axis coordinate of the predicted object center, W is the width of the predicted object, H is the height of the predicted object, and C is the category of the recognized object. It should be noted that there is only one type of object in this method.

[0081] Since the Yolo model can only extract the features of 2D images, the detection results show many discontinuous blood vessels and isolated data, so there will be N objects. Therefore, a relatively rough blood vessel centerline is obtained in this step.

[0082] For example, referring to Figure 4 and Figure 5 , the first object detection model obtains the first vascular feature from the image to be detected, which are respectively Figure 4 the left common carotid artery-internal carotid artery bifurcation corresponding to Figure 5 the left common carotid artery-external carotid artery bifurcation corresponding to

[0083] S13. Superimpose the first vascular feature on the image to be detected and form two-channel data.

[0084] Convert the output result of Yolo into the same format as the original data and then superimpose it. The conversion method is to assign 1 to the rectangle corresponding to the Yolo detection result and 0 to the rest.

[0085] When performing 3D convolution, five-dimensional data is usually used for operation and reasoning inside the network. The five dimensions are: [Batch, Channel, Z, Y, X]. On the premise of ensuring that the four dimensions of Batch, Z, Y, X are the same, only the second dimension of the matrix needs to be concatenated.

[0086] For example, referring to Figure 6 and Figure 7 ,Figure 6 In the middle, it is filled at the bifurcation of the common carotid artery - internal carotid artery on the left side of the original image based on the first vascular feature (coordinates). Figure 7 It is filled at the bifurcation of the common carotid artery - external carotid artery on the left side of the original image based on the first vascular feature (coordinates).

[0087] S14. Send the image to be detected with the first vascular feature superimposed into a three-dimensional image segmentation model to obtain a second vascular feature. Among them, the three-dimensional image segmentation model is used to extract three-dimensional feature information based on a three-dimensional convolution kernel, and the second vascular feature is used to precisely correspond to the three-dimensional spatial position of the vascular centerline on the image to be detected.

[0088] In this embodiment, the three-dimensional image segmentation model adopts the V-net model. V-net uses three-dimensional convolution, which can effectively utilize three-dimensional features and ensure the continuity of the vascular prediction results. However, three-dimensional convolution requires relatively high computing power, so it is used to optimize the rough segmentation result of the Yolo output. The output of V-net has the same dimension as the original data and is used as an intermediate result.

[0089] For example, superimpose the generated first vascular feature (image) on the original image to obtain an image as shown in Figure 8 and then input it into the V-net model for training to obtain an image as shown in Figure 9 as shown.

[0090] S15. Obtain the coordinates of the vascular centerline based on the second object detection model and the second vascular feature.

[0091] In this step, inputting the intermediate result into the second object detection model, or inputting the intermediate result and the original data superimposed into the network can both obtain accurate vascular centerline coordinates. Similarly, in different embodiments, the second object detection model can be the Yolo model, the SSD model, the Mask-R-CNN model, the Fast-R-CNN model or other models. These models are all object detection models for 2D images and are used for object recognition. Their characteristics are low computing cost and fast processing speed. As an example, in this embodiment, the second object detection model also selects the Yolo model.

[0092] The reason for needing the second Yolo network here is that the vascular wall contour output by V-net is already fine enough, but it cannot provide enough information (coordinates are not included) for making vascular sections. Therefore, another Yolo model needs to be connected to output the coordinates of the vascular sections, that is, the coordinates of the vascular center points.

[0093] S2. Calculate the normal planes passing through the center points of each blood vessel based on the fitting result, and establish a spatial coordinate system based on each normal plane, where the z-axis is the normal direction of the normal plane.

[0094] After fitting the three-dimensional B-spline curve according to the coordinates of the center line of the blood vessel, the normal plane passing through the center point of the blood vessel is obtained. Specifically, the center point of the blood vessel is taken as the origin, the direction orthogonal to the y-axis direction in the original data is defined as the x-axis of the normal plane, the direction orthogonal to the x-axis is defined as the y-axis of the normal plane, and the direction perpendicular to the normal plane is defined as the z-axis.

[0095] S3. Sampling the blood vessel wall image based on the normal plane to obtain slices.

[0096] Under the premise of not losing the pixels where the blood vessel wall is located, a sampling range as small as possible is selected. For example, under the imaging condition of 0.33mm / px, a sampling range of 64x64 centered on the origin can be selected. A slice is obtained according to the sampling range (variable) on the normal plane of each blood vessel center point, which is called a slice.

[0097] S4. Based on the order of the points on the center line of the blood vessel, the slices are superimposed along the z-axis to obtain the result of the surface reconstruction.

[0098] The implementation principle of the method for automatic surface reconstruction of vascular wall images based on centerline extraction in the embodiment of the present application is as follows:

[0099] The blood vessel centerline output by the deep learning model is composed of a series of (three-dimensional) points, which are fitted into a straight line to characterize the direction of the blood vessel. Due to the circuitous course of the blood vessel, it is difficult to show its full picture on a single section, and even multiple end faces will be generated. The blood vessel can be straightened through surface reconstruction to reflect the overall characteristics of the blood vessel. Therefore, the slices on the normal plane are obtained using the blood vessel center point and the blood vessel centerline, and then the slices are superimposed along the z-axis to complete the purpose of surface reconstruction and straightening the blood vessel.

[0100] The present application also discloses an electronic device, including a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executes the above-mentioned blood vessel centerline extraction method for three-dimensional reconstruction of blood vessel walls in black blood MR images, and / or the automatic curved surface reconstruction method of blood vessel wall images based on centerline extraction. The execution subject of the method in this embodiment can be a control device, which is set on an electronic device. The current device can be an electronic device such as a mobile phone, a tablet computer, a laptop computer, etc. with a WIFI function. The execution subject of the method in this embodiment can also be directly a CPU (central processing unit) of the electronic device.

[0101] The embodiments of the present application also disclose a computer-readable storage medium storing a computer program that can be loaded and executed by a processor to perform the above-described method for extracting the centerline of blood vessels and / or the method for automatically reconstructing the curved surface of the blood vessel wall image based on the centerline extraction for three-dimensional reconstruction of the blood vessel wall in black-blood MR images. Through the description of the above embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium as described above (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions for causing a device (which can be a mobile phone, computer, server, controlled terminal, or network device, etc.) to execute the methods of each embodiment of the present application.

[0102] The above are all the preferred embodiments of the present application, and the protection scope of the present application is not limited thereby. Therefore, all equivalent changes made according to the structure, shape, and principle of the present application shall be covered by the protection scope of the present application.

Claims

1. A method for extracting the vascular centerline for three-dimensional reconstruction of the vascular wall in black-blood MR images, characterized in that, it includes the following steps: Obtain the image to be detected; wherein, the image to be detected is a black-blood MR image; Obtain the first vascular feature from the image to be detected based on the first object detection model, wherein the first object detection model is trained to extract features from two-dimensional images, and the first vascular feature is used to roughly correspond to the position of the vascular centerline on the image to be detected; Superimpose the first vascular feature on the image to be detected and form two-channel data; Send the image to be detected with the first vascular feature superimposed into a three-dimensional image segmentation model to obtain the second vascular feature, wherein the three-dimensional image segmentation model is used to extract three-dimensional feature information based on three-dimensional convolution kernels, and the second vascular feature is used to finely correspond to the three-dimensional spatial position of the vascular centerline on the image to be detected; Obtain the vascular centerline coordinates based on the second object detection model and the second vascular feature.

2. The method for extracting the vascular centerline according to claim 1, characterized in that, The first vascular feature is N rectangular frames determined by six dimensions [N, X, Y, W, H, C], where N is the Nth object appearing in the image to be detected, X is the x-axis coordinate of the predicted object center, Y is the y-axis coordinate of the predicted object center, W is the width of the predicted object, H is the height of the predicted object, and C is the category of the recognized object.

3. The method for extracting the vascular centerline according to claim 1, characterized in that, The first object detection model and the second object detection model are Yolo model, SSD model, Mask-R-CNN model or Fast-R-CNN model; and / or, the three-dimensional image segmentation model is a V-net model.

4. An automatic surface reconstruction method for vascular wall images based on centerline extraction, characterized in that, it includes the following steps: Obtain the coordinates of the vascular center points based on the method for extracting the vascular centerline according to any one of claims 1-3 and fit them; Calculate the normal planes passing through each vascular center point based on the fitting result, and establish a spatial coordinate system based on each normal plane, where the z-axis is the normal direction of the normal plane; Sample the vascular wall image based on the normal plane to obtain slices; Stack the slices along the z-axis according to the order of each point on the vascular centerline to obtain the result of surface reconstruction.

5. The automatic surface reconstruction method for vascular wall images according to claim 4, characterized in that, The method for fitting the coordinates of each point on the vascular centerline is B-spline fitting.

6. The automatic surface reconstruction method for vascular wall images according to claim 4, characterized in that, Before sampling the vascular wall image based on the normal plane to obtain slices, it further includes the following steps: Select the sampling range based on the vascular type.

7. A system for extracting the vascular centerline for three-dimensional reconstruction of the vascular wall in black-blood MR images, characterized in that, it includes: An input module for obtaining the image to be detected; wherein, the image to be detected is a black-blood MR image; The first detection module is used to obtain the first blood vessel feature from the image to be detected based on the Yolo model, where the first blood vessel feature is used to roughly correspond to the position of the blood vessel centerline on the image to be detected; The combination module is used to superimpose the first blood vessel feature on the image to be detected and form two-channel data; The second detection module is used to send the image to be detected with the first blood vessel feature superimposed into the V-net model to obtain the second blood vessel feature, where the second blood vessel feature is used to precisely correspond to the position of the blood vessel centerline on the image to be detected; The third detection module is used to obtain the blood vessel centerline coordinates based on the Yolo network and the second blood vessel feature.

8. An automatic curved surface reconstruction system for blood vessel wall images based on centerline extraction, Characterized in that, It includes: The blood vessel centerline extraction system according to claim 7, which is used to obtain the coordinates of each point of the blood vessel centerline; The fitting module is used to fit the coordinates of each point of the blood vessel centerline; The modeling module is used to calculate the normal plane passing through each blood vessel center point based on the fitting result, and respectively establish a spatial coordinate system based on each normal plane, where the z-axis is the normal direction of the normal plane; The sampling module is used to sample the blood vessel wall image based on the normal plane to obtain slices; The reconstruction module is used to stack the slices along the z-axis in the order of each point of the blood vessel centerline to obtain the result of curved surface reconstruction.

9. An electronic device, Characterized in that, It includes: One or more processors; A memory; One or more applications, where the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to: Execute the blood vessel centerline extraction method according to any one of claims 1 to 3, And / or, execute the automatic curved surface reconstruction method for blood vessel wall images according to any one of claims 4 to 6.

10. A computer-readable storage medium, Characterized in that, The storage medium stores at least one instruction, and the at least one instruction is loaded and executed by a processor to implement: The blood vessel centerline extraction method according to any one of claims 1 to 3 and / or the automatic curved surface reconstruction method for blood vessel wall images according to any one of claims 4 to 6.

Images