Blood vessel segmentation method, device and storage medium

Abstract

The present invention provides a blood vessel segmentation method, device and storage medium, including: obtaining aneurysm image data through 3D-CTA scanning, extracting the blood vessel network in the aneurysm image, and optimizing the blood vessel network; reconstructing a polygonal blood vessel grid based on the aneurysm image data, determining the aneurysm orifice and the center line of the adjacent blood vessel that separate the aneurysm sac and the adjacent blood vessel; analyzing and processing the adjacent blood vessels, obtaining a transition area that separates the proximal blood vessel area from the distal blood vessel area, and obtaining an axial plane; calculating the axial plane so that the direction of the axial plane is at a minimum distance from the center point of the orifice contour and the adjacent blood vessel transition area; and obtaining a blood vessel segmentation image after blood vessel segmentation based on the calculation result of the axial plane. The adjacent blood vessels are decomposed into proximal blood vessel areas and distal blood vessel areas, and other geometric description values ​​are introduced, so that the blood vessels can be better distinguished for the flow field segments related to the function, so that the blood vessel segmentation is more accurate.

Description

Technical Field

[0001] The present invention relates to image recognition and processing technology, and in particular to a blood vessel segmentation method, device and storage medium. Background Art

[0002] Intracranial aneurysm (IA) is a cerebrovascular disease that occurs primarily in cerebral arteries and is characterized by pathological dilatation of the vessel. Its worldwide incidence is approximately 1% to 3%, accounting for more than 80% of non-traumatic life-threatening subarachnoid hemorrhage. Moreover, current theories on the development of intracranial aneurysms suggest that there is a general weakness in the vascular connective tissue. This is a long-term injury to patients with vascular aneurysms. Observation and identification of aneurysms and parent vessels are crucial for the treatment of aneurysms. The successful treatment of intracranial aneurysms depends on a full understanding of the anatomical relationship of a specific aneurysm to its parent vessel and nearby branches. In addition, new endovascular techniques are often limited by size limitations. It is necessary to provide an effective method for vessel segmentation in subjects at high risk for intracranial aneurysms to provide adequate radiological screening guidelines and effectively allocate medical resources.

[0003] In the current situation, clinicians usually rely only on computed tomography angiography (CTA) results. In a past study, doctors who only used traditional CTA to observe the parent vessel had a misdiagnosis rate of about 16.2%. Misdiagnosis of the parent vessel in the original CTA image is a risk factor for arrhythmia (HT) and poor clinical prognosis after intra-arterial reperfusion therapy. Classical feature extraction methods are severely hampered by the complex vascular distribution.

[0004] Previous studies have shown that the volume of aneurysm models depends largely on different vessel segmentation methods. The vessel segmentation method also affects the local geometry of the aneurysm. Therefore, it is necessary to validate, compare segmentation methods and adjust the parameters of these segmentation techniques to ensure the quality of patient-specific cerebral vascular hemodynamic analysis.

[0005] However, when dealing with vascular systems with varying intensities and low-contrast aneurysm regions affected by turbulence, the accuracy of vessel segmentation is low. Summary of the invention

[0006] The embodiments of the present invention provide a blood vessel segmentation method, device and storage medium, which decompose adjacent blood vessels into near-vessel regions and far-vessel regions, and introduce other geometric description values, thereby better distinguishing function-related flow field segments of blood vessels, making blood vessel segmentation more accurate.

[0007] A first aspect of an embodiment of the present invention provides a blood vessel segmentation method, comprising:

[0008] Acquire aneurysm image data through 3D-CTA scanning, extract the vascular network in the aneurysm image, and optimize the vascular network;

[0009] reconstructing a polygonal blood vessel mesh based on the aneurysm image data, and determining the aneurysm ostium and the centerline of the adjacent blood vessels separating the aneurysm sac from the adjacent blood vessels;

[0010] Analyze and process the adjacent blood vessels to obtain a transition area separating the proximal blood vessel area from the distal blood vessel area, and obtain an axial plane;

[0011] The axial plane is calculated so that the distance between the direction of the axial plane and the center point of the orifice contour and the transition zone of the adjacent blood vessel is minimized;

[0012] A blood vessel segmentation image is obtained after the blood vessel segmentation based on the calculation result of the axial plane.

[0013] Optionally, in a possible implementation manner of the first aspect, obtaining aneurysm image data through 3D-CTA scanning, extracting a vascular network in the aneurysm image, and optimizing the vascular network includes:

[0014] Determine the starting point of the vascular network;

[0015] Search all vascular networks connected to the starting point;

[0016] The searched vascular network is optimized.

[0017] Optionally, in a possible implementation of the first aspect, the method further includes:

[0018] The geodesic active contour model and Chan-Vese model are used to integrate the region and boundary information, and the velocity function is generated to segment the blood vessels by using the overall threshold and gradient amplitude;

[0019] The initial threshold is calculated based on the Chan-Vese model;

[0020] The Chan-Vese model is continuously trained during the process of segmenting blood vessels.

[0021] Optionally, in a possible implementation of the first aspect, the Chan-Vese model is based on the Mumford-Shah function, and the associated evolution partial differential equation in the level set framework is shown in the following formula:

[0022]

[0023] where μ in is the average intensity of the target object, μ out represents the mean intensity, λ1, λ2, α and β are positive constants.

[0024] Optionally, in a possible implementation of the first aspect, whether the blood vessel segmentation is completed is determined based on the following formula:

[0025] I(X)∈[X-T1,X+T2]

[0026] Where T1 and T2 represent the low and high thresholds for region intensity, l(X) represents the image, and X represents the location of a specific neighboring pixel to be considered for inclusion in the region;

[0027] When there are no more neighboring pixels that meet the criteria of the above formula, the vessel segmentation is considered to be completed.

[0028] Optionally, in a possible implementation of the first aspect, the method further includes:

[0029] Get three consecutive centerline vertices p1, p2 and p3, where p2 is the origin of the plane; calculate the normal vector n orthogonal to the cutting plane by the following formula

[0030]

[0031] The set closest to the centerline represents the cross section of the current centerline container, and the cross section is calculated based on the ellipse.

[0032] A second aspect of an embodiment of the present invention provides a blood vessel segmentation device, comprising:

[0033] A processing module, used for obtaining aneurysm image data through 3D-CTA scanning, extracting a vascular network in the aneurysm image, and optimizing the vascular network;

[0034] a determination module for reconstructing a polygonal blood vessel mesh based on the aneurysm image data, and determining the aneurysm ostium and the centerline of the adjacent blood vessel separating the aneurysm sac from the adjacent blood vessel;

[0035] An acquisition module is used to analyze and process adjacent blood vessels, obtain a transition area separating a proximal blood vessel area from a distal blood vessel area, and obtain an axial plane;

[0036] a calculation module, used for calculating the axial plane so as to minimize the distance between the direction of the axial plane and the center point of the orifice contour and the transition zone of the adjacent blood vessel;

[0037] The segmentation module is used to obtain a blood vessel segmentation image after blood vessel segmentation based on the calculation result of the axial plane.

[0038] Optionally, in a possible implementation manner of the second aspect, the processing module is further configured to perform the following steps, including:

[0039] Determine the starting point of the vascular network;

[0040] Search all vascular networks connected to the starting point;

[0041] The searched vascular network is optimized.

[0042] Optionally, in a possible implementation manner of the second aspect, a cutting module is further included, and the cutting unit performs the following steps, including:

[0043] The geodesic active contour model and Chan-Vese model are used to integrate the region and boundary information, and the velocity function is generated to segment the blood vessels by using the overall threshold and gradient amplitude;

[0044] The initial threshold is calculated based on the Chan-Vese model;

[0045] The Chan-Vese model is continuously trained during the process of segmenting blood vessels.

[0046] According to a third aspect of an embodiment of the present invention, a readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, it is used to implement the method of the first aspect of the present invention and various possible designs of the first aspect.

[0047] The present invention provides a blood vessel segmentation method, device and storage medium. In order to create a more reliable blood vessel segmentation to identify the cause of disease, the present invention adopts a solution of decomposing adjacent blood vessels into near blood vessel and far blood vessel regions. The length and width of the blood vessels are obtained in the quantitative evaluation of the microvascular network, and the decomposition is further applied to calculate the characteristic axial plane. The present invention can handle vascular systems with varying intensity and low-contrast aneurysm areas affected by turbulence, and improve the accuracy of blood vessel segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 is a flow chart of a first embodiment of a blood vessel segmentation method;

[0049] Figure 2 is a flow chart of a second embodiment of the blood vessel segmentation method;

[0050] Figure 3 Schematic diagram of blood vessel segmentation;

[0051] Figure 4 Schematic diagram of blood vessel segmentation;

[0052] Figure 5 This is a structural diagram of a first embodiment of a blood vessel segmentation device. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in sequences other than those illustrated or described herein.

[0055] It should be understood that in various embodiments of the present invention, the size of the sequence number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0056] It should be understood that in the present invention, "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0057] It should be understood that in the present invention, "plurality" refers to two or more than two. "And / or" is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "Contains A, B and C", "Contains A, B, C" means that A, B, and C are all included, "Contains A, B or C" means that one of A, B, and C is included, and "Contains A, B and / or C" means that any one, any two, or any three of A, B, and C are included.

[0058] It should be understood that in the present invention, "B corresponding to A", "B corresponding to A", "A corresponds to B" or "B corresponds to A" means that B is associated with A and B can be determined based on A. Determining B based on A does not mean determining B based only on A, but B can also be determined based on A and / or other information. A and B match when the similarity between A and B is greater than or equal to a preset threshold.

[0059] Depending on the context, "if" as used herein may be interpreted as "when" or "when" or "in response to determining" or "in response to detecting."

[0060] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0061] The present invention provides a blood vessel segmentation method, such as Figure 1 and Figure 2 As shown, including:

[0062] Step S10, obtaining aneurysm image data through 3D-CTA scanning, extracting a vascular network in the aneurysm image, and optimizing the vascular network;

[0063] Step S20, reconstructing a polygonal blood vessel mesh based on the aneurysm image data, and determining the aneurysm ostium and the center line of the adjacent blood vessel that separates the aneurysm sac from the adjacent blood vessel;

[0064] Step S20, analyzing and processing the adjacent blood vessels to obtain a transition region separating the proximal blood vessel region from the distal blood vessel region, and obtaining an axial plane;

[0065] Step S30, calculating the axial plane so that the distance between the direction of the axial plane and the center point of the orifice contour and the transition zone of the adjacent blood vessel is minimized;

[0066] A blood vessel segmentation image after blood vessel segmentation is obtained based on the calculation result of the axial plane.

[0067] The algorithm was performed using 3D CTA scans, and the images were acquired with a CT angiography scanner with a table speed of 9 mm / s and a zero-degree table (gantry tilt). The scan started from the common carotid artery and continued parallel to the orbital line until the level of the circle of Willis, during which contrast was injected intravenously at a rate of 3.5 mls / s. The aneurysm image was a 512×512 pixel field, and slices of continuous thickness were used for segmentation and reconstruction of the 3D vessel geometry. Pixels were expressed in Hounsfield units (HU). The initial zero level set was a rectangular prism surface, which was constructed by subtracting two pixels from both sides of the CTA image. In the third vessel quantification part using a morphological characterization method, the length and width of the vessel were quantified. The total area and length of the vessel were obtained by pixel accumulation from the vessel segment image, respectively, and therefore, the width was defined as the ratio of the area to the length.

[0068] In the vessel extraction part, the connection area is first determined from the starting point of the vessel (as a microvascular network). In order to obtain an accurate and optimized set of vascular networks with lower noise and fewer artifacts, an improved method is proposed here. Three main processing steps are listed here.

[0069] In one embodiment, in step S10, it further includes:

[0070] Determine the starting point of the vascular network. Detect the matrix of the CTA image by local line transection and record the number of lines without valid points. If the ratio of the detection value is lower than the preset threshold, the point is considered as the starting point. Then perform Z-shaped transection detection and continue to search for other starting points of the qualified vascular network until all transection detections are completed.

[0071] All vascular networks connected to the starting point are searched. First, all starting points are marked as part of the initial vascular network and restored in the network point set. Then, the detection coordinates are moved from the starting point to the next position in the optical coherence tomography angiography (OCTA) image, and the minimum distance d between the detection point and all points in the set is calculated. min If d min If the point is small enough (within 2 pixels), that is, the detected point is close enough or connected to some blood vessels recovered in the network point set, it will be marked as part of the initial blood vessel network and stored in the set. Through global image cross-section detection, all point sets that meet the requirements are stored to obtain the initial blood vessel network. However, since the blood vessels extracted by the regional connectivity method are sensitive to the starting position, the original image will be rotated 90°, 180°, and 270° respectively to search for different starting points and initial blood vessels. Finally, the blood vessel networks obtained from different starting points are merged to obtain the complete blood vessel network.

[0072] The searched vascular network is optimized. Binary image microvessels can be achieved by the maximum between-class variance method (OTSU) image binarization method. In the initial vascular network, some noise and artifacts may be mistakenly marked as blood vessels because the distance or gray value is too close to the actual blood vessel. Therefore, this algorithm applies the noise and artifact reduction method (combining morphology and segment-by-segment analysis methods) to vacuum closure assistance (VCA). Here, the skeleton of the microvessels in the OCTA image is extracted to obtain a thinner blood vessel map showing only the vascular skeleton. The segment-by-segment analysis method is used to evaluate the correctness of the extracted vascular skeleton to further reduce noise and artifacts. The branch and breakpoint information of each vascular curve is used to obtain the branch length and total length of each blood vessel. If the branch length is too short, or the ratio of the total length to the number of bifurcation points plus the number of breakpoints is lower than a preset threshold, it will be regarded as noise or an artifact to be eliminated. Therefore, noise and artifact pixels can be easily distinguished from vascular pixels. After performing all the above steps, a fully optimized vascular network is finally obtained.

[0073] The Chan-Vese model is based on the Mumford-Shah function. The associated evolution partial differential equation (PDE) in the level set framework is

[0074]

[0075] Where μin is the average intensity of the target object, μout represents the mean intensity, λ1, λ2, α and β are positive constants. The Chan-Vese model does not require terms related to image gradients. Instead, the regional intensity information is used to segment the target object. The model shows significant effectiveness in image segmentation with blurred boundaries.

[0076] The 3D surface models were imported into the system to generate the volume finite element mesh for CTA calculations. An unstructured grid of tetrahedral and prismatic pixels was used for volume meshing with a consistent mesh density. For the patient-specific truncated models, the number of pixels varied according to the volume differences of the truncated models, but the models occupied the same pixel density of approximately 2800 pixels / mm3. The total number of pixels varied from 1.8 million to 8.34 million, which was sufficient to obtain meshless computational fluid dynamics (CFD) simulation results. For the ideal model, 3 million tetrahedral and prismatic pixels were generated with a density of 1868 pixels / mm3. Tetrahedral cells were generated from meshes with different mesh densities for surface meshing. In all these models, the minimum mesh resolution was approximately 0.1 mm for the walls, inlets, and outlets, and 0.05 mm for the aneurysm sac.

[0077] A polygonal vessel mesh is reconstructed from the image data. The aneurysm ostium and the centerlines of the adjacent vessels are determined, separating the aneurysm sac from the adjacent vessels. These geometric description values ​​and the polygonal vessel mesh form the input for an automated vessel analysis based on an approximation method. The result of the analysis is a transition region that separates the proximal vessel region from the distal vessel region. From this decomposition, the axial plane is derived, which is another geometric description value. Two applications of this algorithm are proposed for this versatile platform: automatic definition of vessel organization and support of 3D interaction through the placement of virtual cameras.

[0078] Since exploring blood flow within aneurysms for risk analysis and developing new treatment options is not part of common clinical procedures, a decomposition method that facilitates this process would be a tool for medical research. However, it could be used clinically as long as flow data are available in clinical routine. Based on the present algorithm, the vessels are automatic for two reasons: to produce reproducible results and to avoid tedious interactions, as medical researchers have to deal with the interaction of multiple degrees of freedom on multidimensional data when using established exploration software. The geometric descriptors derived from the decomposition should not only facilitate the analysis process but also support the interaction with the data.

[0079] The algorithm integrates regional and boundary information based on the geodesic active contour model and the Chan-Vese model to generate a velocity function to segment the vessels. The initial threshold is calculated based on the Chan-Vese model and then iteratively updated throughout the segmentation process. When the aneurysm boundary is reached, the change in the threshold will decrease due to the contrast between the aneurysmal intensity and the non-aneurysmal intensity, and the iteration will stop to achieve the segmentation of the tumor-bearing and non-tumor-bearing vessels. The algorithm can then be implemented in an automatic or semi-automatic manner depending on the complexity of the aneurysm shape.

[0080] A polygonal vessel mesh is reconstructed from the image data. The aneurysm ostium and the center lines of the adjacent vessels are determined, which separate the aneurysm sac from the adjacent vessels. These geometric description values ​​and the polygonal vessel mesh form the input for an automated vessel analysis based on an approximation method. The result of the analysis is a transition region that separates the proximal vessel region from the distal vessel region. From this decomposition, an axial plane is derived, which is another geometric description value.

[0081] The dilated (33 kernel) binary segmentation mask is applied to the original dataset. The masked dataset forms the input to the Marching Cubes algorithm. Depending on the application (visualization, etc.) and the quality of the image data, post-processing steps are optional. These include removing surface noise, cutting in and out of orthogonal openings or removing container mixing due to partial volume effects during blending. Once the shape quality of the surface is passed, the quality of the surface pixels can be improved by performing a remeshing beforehand.

[0082] In one embodiment, it further includes:

[0083] Whether the blood vessel segmentation is completed is determined based on the following formula:

[0084] I(X)∈[X-T1,X+T2]

[0085] Where T1 and T2 represent the low and high thresholds for region intensity, I(X) represents the image, and X represents the location of a specific neighboring pixel to be considered for inclusion in the region;

[0086] When there are no more neighboring pixels that meet the criteria of the above formula, the vessel segmentation is considered to be completed.

[0087] To decompose the vessel surface, two additional geometric description values ​​need to be derived from the surface: the centerline of the adjacent vessels and the plane of the aneurysm ostium. To calculate the centerline, the method uses the Vascular Modeling Toolkit (VMTK) and interacts at the start and end points of the centerline. A 3D Voronoi diagram is calculated from the surface mesh and used to define the centerline.

[0088] Since the generated centerline comes directly from the polygonal surface, it is suitable as a geometric path for subsequent surface sampling. To generate the aneurysm ostium plane, this method uses an iterative method that takes the surface on the aneurysm sac, the centerline and a user-defined point as input, and the output is four control points located between adjacent vessels and the aneurysm. A specific interpolation scheme is used to generate a curved, smooth surface from the control points.

[0089] In one embodiment, Figure 3 As shown, it also includes:

[0090] Get three consecutive centerline vertices p1, p2 and p3, where p2 is the origin of the plane; calculate the normal vector n orthogonal to the cutting plane by the following formula

[0091]

[0092] The set closest to the centerline represents the cross section of the current centerline container, and the cross section is calculated based on the ellipse.

[0093] This algorithm is between the Gauss-Newton algorithm and the gradient descent method. It uses the contour value and the Gaussian function as input and iteratively changes the three function parameters a, b and c until the squared distance between the contour and the function value is minimized. Once the model function is parameterized, the transition position p of p = n-√(π / b) = 39 can be obtained, which is derived from the variance of the generalized Gaussian curve.

[0094] Once the approximate ellipses for all centerline locations are computed, the algorithm takes the area A = π ab, where a and b are the lengths of the major and minor axes of each ellipse, and samples the surface along the centerline, the approximate ellipse at a selected set of cut vertices in the 2D domain, and the minor / major axes of the approximate ellipse mapped back to the 3D domain.

[0095] For adjacent vessels, the two main mechanisms of the decomposition method are based on approximations: an approximation of the cross-sectional shape of the vessel and an approximation of the distribution of the vessel area along the centerline. Both are based on sampling the vessels along their centerlines. For each adjacent vessel, sampling starts at the inlet / outlet and proceeds towards the aneurysm. A suitable initial viewpoint can be defined as the one that provides the lowest degree of visual obstruction. The axial plane should be oriented so that its distance to the centerpoint of the orifice contour and the transition zone of the adjacent vessels is minimized. Therefore, its normal vector automatically provides a view “sideways” to a set of centerpoints, minimizing the possibility of occlusion. The choice of viewpoint should also reflect that the aneurysmal sac is the primary feature of interest. This can be achieved by using a standard reverse plane as the viewing direction if most of the aneurysmal sac vertices are on the opposite side of the axial plane.

[0096] The second step is to optimize the vascular segmentation image thus obtained. Binary image microvessels can be achieved by the OTSU image binarization method. In the initial vascular network, some noise and artifacts may be mistakenly marked as vessels because the distance or grayscale value is too close to the actual blood vessels. Therefore, this algorithm applies the noise and artifact reduction method (combining morphological and segment-by-segment analysis methods) to CTA. Here, the skeleton of the microvessels in the OCTA image is extracted to obtain a thinner vascular map that only shows the vascular skeleton. The segment-by-segment analysis method is used to evaluate the correctness of the extracted vascular skeleton to further reduce noise and artifacts. The branch and breakpoint information of each vascular curve is used to obtain the branch length and total length of each vessel. If the branch length is too short, or the ratio of the total length to the number of bifurcation points plus the number of breakpoints is lower than the preset threshold, it will be regarded as noise or an artifact to be eliminated. Therefore, noise and artifact pixels can be easily distinguished from vascular pixels. After performing all the above steps, a fully optimized vascular network is finally obtained.

[0097] Finally, the blood vessel segmentation result of the present invention is as follows Figure 4 As shown. The center point (pi) is calculated from the transition area (red) and the orifice contour (blue) and projected into the planar space. There, the plane center point (pM) is used to define the order by comparing the angles (a–c). The results obtained from the surface distance metrics (maximum surface distance (HD) and mean absolute surface distance (MASD)) show the reliability of the segmentation method. The HD values ​​of this method range from 0.51 to 1.89 pixels, and the maximum MASD is 0.08. In this way, the segmentation of blood vessels by this algorithm is achieved.

[0098] The present invention has the following advantages, including:

[0099] 1. The commonly used method for auxiliary vascular analysis provided by the present invention is to decompose the anatomical structure into aneurysm-bearing vessels and adjacent vessels. This decomposition is used for geometry-based analysis and calculation of the aneurysm sac shape index. Other methods focus on extracting the aneurysm neck or aneurysm opening. The morphology of the blood vessels has a significant impact on flow characteristics such as flow velocity and turbulence. Therefore, including the geometric information of the extracted anatomical features (vascular neck, aneurysm opening, etc.) in the analysis process is helpful for studying specific parts of the flow field or for extracting certain flow features. The automatic method of this algorithm to decompose adjacent blood vessels into proximal and distal vessel regions introduces other geometric description values. These can be used to better distinguish functionally related flow field segments.

[0100] 2. The algorithm was applied to 9 datasets in the database during the testing phase. The selected datasets represent the anatomical variability of cerebral aneurysms, including small, medium, and large lateral aneurysms as well as small to giant bifurcation aneurysms with up to 6 adjacent vessels. Decomposition can be achieved automatically on all datasets. Since the noise level randomly changes the local surface diameter up to 15%, the method can be successfully applied to all datasets. Six of the 9 datasets can be successfully decomposed with a noise level of up to 30%.

[0101] 3. Compared with the traditional vascular segmentation results, this algorithm can effectively eliminate a large amount of surrounding scattered noise and obtain a clean vascular network using CTA images. In addition, the improved algorithm can effectively reduce near-point noise and completely eliminate any artificial noise, so that the length and width of the blood vessels can be measured more accurately.

[0102] 4. Qualitatively compare the average results of the dataset with the results of the algorithm using the automatic method. The manual and automatic results show that they have sufficient consistency for more than 85% of the transition areas (center point position deviation <10% container diameter).

[0103] 5. According to the experimental results, the algorithm scheme has very good sensitivity for aneurysms. In fact, even using a data set without manual input, the sensitivity is much higher than previous reports (reported sensitivity ranges from 69% to 99%). In addition, and perhaps more importantly, for small aneurysms, the sensitivity of the tumor-bearing vessel segmentation scheme measured by this method can be as high as 91%, which is much higher than the 35% to 56% sensitivity reported in current clinical detections.

[0104] 6. This algorithm can reduce and ignore the noise level after vessel segmentation, and successfully decompose adjacent vessels into near-vessel and far-vessel regions. Compared with existing algorithms, the proposed method shows better performance in accuracy, so it can be used for vascular aneurysm treatment monitoring. After conducting a study on vascular aneurysm treatment monitoring, data showed that from the first day to the seventh day of treatment, the average (arithmetic mean) length of neurovascular conflict (NVC) had been greatly shortened by 36.8% (P<0.01), indicating a significant computational effect. And the original misdiagnosis rate of doctors for tumor-bearing vessels was reduced from 16.2% to 5%. At the same time, this algorithm is easy to operate, and the detection results of the CTA scheme can be integrated into the image viewing tool and displayed in a variety of ways, including 2D slice-based highlighting and 3D vessel-based annotations. The radiologist can then take into account other factors such as patient history, machine variability, and other coexisting pathologies. By limiting the number of detections, the radiologist can adjust the confidence level of the CTA output. Since the probability of being an aneurysm is assigned to each aneurysm candidate, the radiologist can also adjust the confidence level by thresholding this probability. The number of false positives is very manageable (less than ten false positives are acceptable per case), and an efficient user interface can make reviewing this number of candidates a quick task.

[0105] The detection of intracranial aneurysms is an important task. Unexplained aneurysm rupture with subsequent intracranial hemorrhage is a significant cause of morbidity and mortality in young people. The role of screening has been controversial but is generally accepted for high-risk populations, such as those with a strong family history of intracranial aneurysms or those who are prone to polycystic kidney disease. The value of screening is directly related to accuracy. This value is low if many positive cases are missed. This value is also low when the cost of confirmatory testing is high in terms of chance and potential morbidity if there are many false positives. Improving the accuracy of the test makes it more suitable for screening a wider population. From this perspective, improving the accuracy of intracranial aneurysm parent vessel screening and segmentation using the present algorithm is an important advance. In addition, the number of intracranial CTA examinations for all suspected pathologies (including stroke, neurological dysfunction, and others) is very large, and vessel segmentation should be evaluated for each examination.

[0106] After measuring 9 data sets, the differences in vessel segmentation calculated by using this algorithm remain at a low level. Compared with the conventional diagnosis misdiagnosis rate using only CTA visual inspection, the misdiagnosis rate of this algorithm is only about 5%, and the noise and artifacts are eliminated by combining morphological features. Since the automatic decomposition method used by this algorithm is stable even under severe surface noise, it can provide adequate results in all tested cases. This robustness is mainly attributed to the approximation of the vessel area and area contour with the help of the model function. Therefore, this method is also suitable for data of lower quality, which may be caused by suboptimal scanning conditions or simplified reconstruction processes.

[0107] This algorithm can handle vasculature with intensity variations and aneurysm-bearing vessel regions with low contrast that are affected by neighboring vessels. It is based on the use of multi-range filters and local variance to extract intensity-based image features to identify vasculature with contrast variations. At the same time, in order to solve the local profile changes in the transition zone that are not similar to the existing vessel segmentation model function, this algorithm optimizes the found transition position by local comparison of the model function and the profile. If the contour value of the transition position exceeds the value of the model function, the backward iterated vessel (the vessel "away from the aneurysm") can be moved until the model function value is higher than the contour value. If the contour value exceeds the model function value, a similar forward transition is performed. It is a technique that can dissect aneurysms and parent vessels by setting seed points. This method is also suitable for the segmentation of complex cerebrovascular anatomical shapes.

[0108] A key feature of the present vessel segmentation algorithm is that it automates the viewpoint setting. With current clinical medical technology, exploring multidimensional data while navigating in 3D space is a tedious task. The algorithm's carefully selected automatic viewpoints can help to obtain a spatial overview of the given data. It can also be used as a starting point for the exploration process to which the user can return once a specific exploration step is completed. In the field of 3D interaction, automatic viewpoint selection is a proactive technology that supports the exploration process. This means that the results of the operator will not vary and can be applied to screening centers where advanced operator expertise may not be available. At the same time, the algorithm is computationally reasonable. The median execution time is 8 minimum central processing units (CPU min ), while desktop computers have a maximum of 27 (CPU min ). This means it can be inserted into a typical radiology department’s workflow with little to no disruption.

[0109] In addition, the algorithm focuses on the relationship between complex time-dependent blood flow patterns and the location of aneurysm rupture. The algorithm uses other geometric description values ​​besides the orifice contour and centerline, and points out the role of transition zones for other tasks such as seeding streamlines and masking flow field parts. And the medical community generally recognizes the benefits of automatically selecting viewpoints for vascular segmentation.

[0110] The automatic calculation of geometric description values ​​is a valuable extension to the existing vessel segmentation workflow and research database applications. Future work will focus on the integration process and will focus on other application-dependent requirements that may arise. The results and analysis have confirmed that the optimization process proposed by the improved vessel segmentation algorithm is both effective and feasible for the segmentation and quantification of vessels with lower noise and fewer artifacts. Therefore, it can potentially be used to monitor intracranial aneurysms during treatment. Finally, the study of the statistical analysis based on the results also shows that the method can be potentially applied in practice to monitor changes in neovascularization after treatment in the future with further improvements.

[0111] The embodiment of the present invention also provides a blood vessel segmentation device, such as Figure 5 As shown, including:

[0112] A processing module, used for obtaining aneurysm image data through 3D-CTA scanning, extracting a vascular network in the aneurysm image, and optimizing the vascular network;

[0113] a determination module for reconstructing a polygonal blood vessel mesh based on the aneurysm image data, and determining the aneurysm ostium and the centerline of the adjacent blood vessel separating the aneurysm sac from the adjacent blood vessel;

[0114] An acquisition module is used to analyze and process adjacent blood vessels, obtain a transition area separating a proximal blood vessel area from a distal blood vessel area, and obtain an axial plane;

[0115] a calculation module, used for calculating the axial plane so as to minimize the distance between the direction of the axial plane and the center point of the orifice contour and the transition zone of the adjacent blood vessel;

[0116] The segmentation module is used to obtain a blood vessel segmentation image after blood vessel segmentation based on the calculation result of the axial plane.

[0117] In one embodiment, the processing module is further configured to perform the following steps, including:

[0118] Determine the starting point of the vascular network;

[0119] Search all vascular networks connected to the starting point;

[0120] The searched vascular network is optimized.

[0121] In one embodiment, a cutting module is further included, wherein the cutting unit performs the following steps, including:

[0122] The geodesic active contour model and Chan-Vese model are used to integrate the region and boundary information, and the velocity function is generated to segment the blood vessels by using the overall threshold and gradient amplitude;

[0123] The initial threshold is calculated based on the Chan-Vese model;

[0124] The Chan-Vese model is continuously trained during the process of segmenting blood vessels.

[0125] Among them, the readable storage medium can be a computer storage medium or a communication medium. The communication medium includes any medium that facilitates the transmission of a computer program from one place to another. The computer storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer. For example, a readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application-specific integrated circuit (Application Specific Integrated Circuits, abbreviated as: ASIC). In addition, the ASIC can be located in a user device. Of course, the processor and the readable storage medium can also exist in a communication device as discrete components. The readable storage medium can be a read-only memory (ROM), a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0126] The present invention also provides a program product, which includes an execution instruction, which is stored in a readable storage medium. At least one processor of a device can read the execution instruction from the readable storage medium, and at least one processor executes the execution instruction so that the device implements the methods provided in the above various embodiments.

[0127] In the above-mentioned terminal or server embodiments, it should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the present invention may be directly implemented as being executed by a hardware processor, or may be executed by a combination of hardware and software modules in the processor.

[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A blood vessel segmentation method, characterized in that: include: Acquire aneurysm image data through 3D-CTA scanning, extract the vascular network in the aneurysm image, and optimize the vascular network; reconstructing a polygonal blood vessel mesh based on the aneurysm image data, and determining the aneurysm ostium and the centerline of the adjacent blood vessels separating the aneurysm sac from the adjacent blood vessels; Analyze and process the adjacent blood vessels to obtain a transition area separating the proximal blood vessel area from the distal blood vessel area, and obtain an axial plane; The axial plane is calculated so that the distance between the direction of the axial plane and the center point of the orifice contour and the transition zone of the adjacent blood vessel is minimized; Obtaining a blood vessel segmentation image after blood vessel segmentation based on the calculation result of the axial plane; The step of obtaining aneurysm image data by 3D-CTA scanning, extracting a vascular network in the aneurysm image, and optimizing the vascular network comprises: Determine the starting point of the vascular network; The matrix of the CTA image is detected by local line transection, and the number of lines without valid points is recorded; if the ratio of the detection value is lower than the preset threshold, the point is regarded as the starting point; then Z-shaped transection detection is performed, and other starting points of the qualified vascular network are continuously searched until all transection detections are completed; Search for all vascular networks connected to the starting point; first, mark all starting points as part of the initial vascular network and restore them in the network point set; then, move the detection coordinates from the starting point to the next position in the optical coherence tomography angiography image, and calculate the minimum distance dmin between the detection point and all points in the set; if dmin is small enough, that is, the detected point is close enough or connected to some blood vessels restored in the network point set, it will be marked as part of the initial vascular network and stored in the set; through global image cross-section detection, all point sets that meet the requirements are stored to obtain the initial vascular network; the original image will be rotated 90°, 180° and 270° respectively to search for different starting points and initial blood vessels; finally, the vascular networks obtained from different starting points are merged to obtain the complete vascular network; The searched vascular network is optimized; the skeleton of the microvessels in the OCTA image is extracted to obtain a thinner vascular map showing only the vascular skeleton; a segment-by-segment analysis method is used to evaluate the correctness of the extracted vascular skeleton to further reduce noise and artifacts; the branch and breakpoint information of each vascular curve is used to obtain the branch length and total length of each blood vessel; if the branch length is too short, or the ratio of the total length to the number of bifurcation points plus the number of breakpoints is lower than a preset threshold, it will be regarded as noise or an artifact to be eliminated.

2. The blood vessel segmentation method according to claim 1, characterized in that: Also includes: The geodesic active contour model and Chan-Vese model are used to integrate the region and boundary information, and the velocity function is generated to segment the blood vessels by using the overall threshold and gradient amplitude; The initial threshold is calculated based on the Chan-Vese model; The Chan-Vese model is continuously trained during the process of segmenting blood vessels.

3. The blood vessel segmentation method according to claim 1, characterized in that: Also includes: Whether the blood vessel segmentation is completed is determined based on the following formula: I(X)∈[X-T1,X-T2] Where T1 and T2 represent the low and high thresholds for region intensity, I(X) represents the image, and X represents the location of a specific neighboring pixel to be considered for inclusion in the region; When there are no more neighboring pixels that meet the criteria of the above formula, the vessel segmentation is considered to be completed.

4. The blood vessel segmentation method according to claim 1, characterized in that: Also includes: Get three consecutive centerline vertices p1, p2 and p3, where p2 is the origin of the plane; The normal vector n orthogonal to the cutting plane is calculated by the following formula The set closest to the centerline represents the cross section of the current centerline container, and the cross section is calculated based on the ellipse.

5. A blood vessel segmentation device, characterized in that: include: A processing module, used for obtaining aneurysm image data through 3D-CTA scanning, extracting a vascular network in the aneurysm image, and optimizing the vascular network; a determination module for reconstructing a polygonal blood vessel mesh based on the aneurysm image data, and determining the aneurysm ostium and the centerline of the adjacent blood vessel separating the aneurysm sac from the adjacent blood vessel; An acquisition module is used to analyze and process adjacent blood vessels, obtain a transition area separating a proximal blood vessel area from a distal blood vessel area, and obtain an axial plane; a calculation module, used for calculating the axial plane so as to minimize the distance between the direction of the axial plane and the center point of the orifice contour and the transition zone of the adjacent blood vessel; A segmentation module, used for obtaining a blood vessel segmentation image after blood vessel segmentation based on the calculation result of the axial plane; The processing module is further configured to perform the following steps, including: Determine the starting point of the vascular network; The matrix of the CTA image is detected by local line transection, and the number of lines without valid points is recorded; if the ratio of the detection value is lower than the preset threshold, the point is regarded as the starting point; then Z-shaped transection detection is performed, and other starting points of the qualified vascular network are continuously searched until all transection detections are completed; Search for all vascular networks connected to the starting point; first, mark all starting points as part of the initial vascular network and restore them in the network point set; then, move the detection coordinates from the starting point to the next position in the optical coherence tomography angiography image, and calculate the minimum distance dmin between the detection point and all points in the set; if dmin is small enough, that is, the detected point is close enough or connected to some blood vessels restored in the network point set, it will be marked as part of the initial vascular network and stored in the set; through global image cross-section detection, all point sets that meet the requirements are stored to obtain the initial vascular network; the original image will be rotated 90°, 180° and 270° respectively to search for different starting points and initial blood vessels; finally, the vascular networks obtained from different starting points are merged to obtain the complete vascular network; The searched vascular network is optimized; the skeleton of the microvessels in the OCTA image is extracted to obtain a thinner vascular map showing only the vascular skeleton; a segment-by-segment analysis method is used to evaluate the correctness of the extracted vascular skeleton to further reduce noise and artifacts; the branch and breakpoint information of each vascular curve is used to obtain the branch length and total length of each blood vessel; if the branch length is too short, or the ratio of the total length to the number of bifurcation points plus the number of breakpoints is lower than a preset threshold, it will be regarded as noise or an artifact to be eliminated.

6. The blood vessel segmentation device according to claim 5, characterized in that: Also included is a cutting module, which performs the following steps, including: The geodesic active contour model and Chan-Vese model are used to integrate the region and boundary information, and the velocity function is generated to segment the blood vessels by using the overall threshold and gradient amplitude; The initial threshold is calculated based on the Chan-Vese model; The Chan-Vese model is continuously trained during the process of segmenting blood vessels.

7. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and the computer program is used to implement any one of the methods of claims 1 to 4 when executed by a processor.

Images