Multi-level Vascular Network Reconstruction and Segmentation Method Based on Watershed Analysis

Abstract

The present invention relates to the field of medical image processing technology, and more specifically, to a multi-level vascular network reconstruction and segmentation method based on watershed analysis. The method comprises the following steps: obtaining three-dimensional computed tomography angiography (3D CTA) scan results; reconstructing a 3D vascular network based on the 3D CTA scan results; performing image segmentation and enhancement processing on the 3D CTA scan results to obtain original vascular centerlines; constructing a pressure field based on the original vascular centerlines; determining watershed locations in the vascular network based on the pressure field; segmenting the vascular network based on the watershed locations; and outputting the reconstructed 3D vascular network and vascular segmentation results. By cleverly combining watershed analysis theory with traditional image processing techniques, the vascular reconstruction and segmentation process simultaneously considers both vascular geometry and hemodynamic characteristics. This method not only improves the accuracy of reconstruction and segmentation but also provides a new perspective for vascular function assessment.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and more specifically, to a multi-level vascular network reconstruction and segmentation method based on watershed analysis. Background Art

[0002] With the continuous progress of medical imaging technology, angiographic imaging plays an increasingly important role in clinical diagnosis and treatment. Accurate vascular network reconstruction and segmentation can not only provide valuable diagnostic information for doctors, but also provide important support for surgical planning and navigation. However, due to the complexity of the vascular structure and various interference factors in the imaging process, precise vascular network reconstruction and segmentation remain a challenging problem.

[0003] Traditional vascular reconstruction methods mainly rely on basic image processing techniques such as threshold segmentation and morphological processing. These methods perform well when dealing with simple vascular structures, but are often inadequate when faced with complex multi-level vascular networks. Especially when dealing with small blood vessels, bifurcation points, and lesion areas, these methods often produce problems such as breaks, over-segmentation, or under-segmentation, resulting in inaccurate and incomplete reconstruction results.

[0004] In recent years, model-based methods and machine learning techniques have been widely used in the field of vascular reconstruction. These methods improve the accuracy of vascular reconstruction to a certain extent by introducing prior knowledge or using a large amount of labeled data for training. However, these methods also have some inherent limitations. Model-based methods often require carefully designed mathematical models with limited generalization ability and are difficult to adapt to the morphological changes of blood vessels under different cases and imaging conditions. Machine learning methods, on the other hand, highly rely on the quality and quantity of training data and may perform poorly when faced with rare cases or new imaging modalities.

[0005] In addition, most existing vascular reconstruction methods focus on the geometric morphology reconstruction of blood vessels while ignoring the hemodynamic characteristics inside the blood vessels. This purely morphology-based reconstruction method is difficult to reflect the functional characteristics of the vascular network and limits its value in some clinical applications. For example, when evaluating the risk of aneurysm rupture or planning vascular intervention surgery, relying solely on the geometric information of blood vessels is far from enough.

[0006] Another issue worthy of attention is the accuracy and consistency of vascular segmentation. Existing methods often have difficulty accurately identifying the boundaries between blood vessels at different levels when dealing with complex multi-level vascular networks, resulting in inaccurate segmentation results. This not only affects subsequent quantitative analysis but may also mislead clinical decision-making.

[0007] In summary, the existing methods for vascular network reconstruction and segmentation still have many deficiencies when dealing with complex multi-level vascular networks, making it difficult to meet the growing clinical needs. Therefore, there is an urgent need for a vascular network reconstruction and segmentation method that can comprehensively consider the geometric shape and hemodynamic characteristics of blood vessels, and at the same time has high precision, high reliability and good adaptability. Summary of the Invention

[0008] The present invention is precisely proposed in view of the above technical problems. The present invention proposes a method for reconstructing and segmenting multi-level vascular networks based on watershed analysis, aiming to overcome the limitations of the prior art and achieve more accurate, reliable and comprehensive vascular network reconstruction and segmentation.

[0009] To solve the above technical problems, the present invention adopts the following technical solutions:

[0010] A method for reconstructing and segmenting multi-level vascular networks based on watershed analysis, comprising:

[0011] An acquisition step, comprising:

[0012] Obtaining the three-dimensional computed tomography angiography (3DCTA) scan results;

[0013] A processing step, comprising:

[0014] Reconstructing a 3D vascular network based on the 3DCTA scan results;

[0015] Performing image segmentation and enhancement processing on the 3DCTA scan results to obtain the original vascular centerline;

[0016] Constructing a pressure field based on the original vascular centerline;

[0017] Determining the watershed positions in the vascular network based on the pressure field;

[0018] Segmenting the vascular network according to the watershed positions;

[0019] An output step, comprising:

[0020] Outputting the reconstructed 3D vascular network and the vascular segmentation results.

[0021] Preferably, the reconstruction of the 3D vascular network specifically includes:

[0022] Obtaining a series of vascular regions through a vascular extraction algorithm;

[0023] Reconstructing a 3D vascular network based on the skeleton of the vascular regions according to the vascular centerline extraction algorithm.

[0024] Preferably, the vascular regions are generated by calculating the cumulative amount of watershed water to form different watershed regions.

[0025] Preferably, the obtaining of the original blood vessel centerline specifically includes:

[0026] Performing non-linear enhancement on the 3DCTA scan result by using a blood vessel enhancement algorithm to highlight blood vessel structures and details;

[0027] Identifying the blood vessel centerline by using the maximum radius filtering technique to filter out non-blood-vessel regions of interest and small blood vessels;

[0028] Obtaining a rough original blood vessel centerline based on a three-dimensional reconstruction algorithm;

[0029] Obtaining a refined blood vessel centerline by using a neighborhood-based skeleton refinement technique.

[0030] Preferably, the constructing of the pressure field specifically includes:

[0031] Constructing a blood pressure field by combining the original blood vessel centerline and blood vessel imaging parameters;

[0032] Generating a gradient field of the blood pressure field and optimizing the gradient field;

[0033] Calculating the pressure gradient of the blood pressure field.

[0034] Preferably, the determining of the watershed position in the blood vessel network specifically includes:

[0035] Projecting the original blood vessel centerline into a two-dimensional space by using a pressure gradient calculation method to obtain the projected blood vessel centerline;

[0036] Projecting the projected blood vessel centerline onto the blood vessel image to mark the position of the watershed.

[0037] Preferably, it further includes:

[0038] Performing connectivity analysis on the blood vessel segmentation result;

[0039] Based on the result of the connectivity analysis and the blood vessel segmentation result, obtaining a reconstructed and complete multi-level blood vessel segmentation result.

[0040] Preferably, the connectivity analysis specifically includes:

[0041] Performing voxelization operation on the blood vessel segmentation result to label each voxel as an independent region;

[0042] Scanning all connected regions and testing the connection between each two connected regions;

[0043] Merging the connected regions based on the connection test result and the input image blood vessels.

[0044] Preferably, it further includes:

[0045] Based on the reconstructed 3D vascular network, calculate the vessel length and volume.

[0046] Preferably, it further includes:

[0047] When intraoperative detection is required, based on the vessel centerlines of the individual images, obtain the segmentation results;

[0048] According to the segmentation results, adjust the positions of the vessel segments;

[0049] When the vessel segmentation results and the vessel centerlines change, reconstruct the watershed again to obtain the updated segmentation results.

[0050] The method of the present invention has the following remarkable technical effects:

[0051] The method of the present invention cleverly combines the watershed analysis theory with traditional image processing techniques, and simultaneously considers the geometric morphology and hemodynamic characteristics of blood vessels during the process of blood vessel reconstruction and segmentation. This innovative method not only improves the accuracy of reconstruction and segmentation, but also provides a new perspective for blood vessel function evaluation. Especially when dealing with complex multi-level vascular networks, the method of the present invention shows significant advantages.

[0052] First of all, by introducing the watershed analysis theory, this method can better capture the topological structure of the vascular network and effectively solve the difficulties of traditional methods in dealing with bifurcation points and small blood vessels. Secondly, this method considers the hemodynamic characteristics during the reconstruction process. By constructing the pressure field and calculating the pressure gradient, it provides a more reliable basis for blood vessel segmentation. This not only improves the accuracy of segmentation, but also lays a foundation for subsequent hemodynamic analysis.

[0053] In addition, the method of the present invention also has good adaptability and robustness. By adopting a multi-step and multi-scale processing strategy, this method can effectively cope with different imaging conditions and changes in blood vessel morphology. Especially when dealing with noise and artifacts, this method shows excellent anti-interference ability. This characteristic makes this method widely applicable in clinical practice and can provide reliable support for the diagnosis and treatment of various blood vessel-related diseases.

[0054] Another remarkable advantage of the method of the present invention lies in its comprehensiveness and systematicness. From the initial image segmentation to the final blood vessel segmentation, this method provides a complete set of solutions. This end-to-end processing flow not only improves the overall efficiency, but also ensures the consistency and coherence between each processing step. Especially in key steps such as blood vessel centerline extraction, pressure field construction, and watershed position determination, this method adopts innovative algorithms and optimization strategies to achieve the organic combination and mutual promotion between these steps.

[0055] It is worth mentioning that the method of the present invention has also made remarkable progress in terms of computational efficiency. By optimizing the algorithm design and utilizing parallel computing technology, this method can complete the reconstruction and segmentation tasks of complex vascular networks within a reasonable time. This feature is of great significance for supporting real-time or near-real-time clinical applications, such as intraoperative navigation and interventional therapy.

[0056] In summary, the multi-level vascular network reconstruction and segmentation method based on watershed analysis proposed by the present invention realizes high-precision, high-reliability, and comprehensiveness in vascular network reconstruction and segmentation through an innovative combination of geometric shape analysis and hemodynamic simulation. Brief Description of the Drawings

[0057] Figure 1 It is the overall flowchart of the method of the present invention.

[0058] Figure 2 It is the flowchart of reconstructing the 3D vascular network of the present invention.

[0059] Figure 3 It is the flowchart of obtaining the original vascular centerline of the present invention.

[0060] Figure 4 It is the flowchart of constructing the pressure field of the present invention.

[0061] Figure 5 It is the flowchart of determining the watershed position and segmenting the vascular network of the present invention. Detailed Description of the Invention

[0062] As Figures 1-5 shown, the present invention discloses a multi-level vascular network reconstruction and segmentation method based on watershed analysis. This method first obtains the three-dimensional computed tomography angiography (3DCTA) scan results, and then reconstructs the 3D vascular network based on these scan results. Next, the 3DCTA scan results are subjected to image segmentation and enhancement processing to obtain the original vascular centerline. Subsequently, a pressure field is constructed based on the obtained original vascular centerline, and the watershed position in the vascular network is determined using this pressure field. Finally, the vascular network is segmented according to the determined watershed position, and the reconstructed 3D vascular network and the vascular segmentation results are output.

[0063] Preferably, in an embodiment of the present invention, the process of reconstructing the 3D vascular network specifically includes obtaining a series of vascular regions through a vascular extraction algorithm, and then based on the vascular centerline extraction algorithm, reconstructing the 3D vascular network according to the skeletons of the obtained vascular regions. This method can effectively capture the geometric features of blood vessels and improve the accuracy of reconstruction. For example, when dealing with the cerebrovascular network, a multi-scale vesselness filter can be used as the vascular extraction algorithm, and the scale range of this filter can be set from 0.5 mm to 2 mm with a step size of 0.1 mm. Such parameter settings can effectively process blood vessels of different calibers, thereby improving the comprehensiveness of the reconstruction results.

[0064] Furthermore, the method of the present invention also includes forming different watershed regions by calculating the accumulation of watershed water volume, thereby generating the above-mentioned vascular regions. This method based on watershed analysis can better reflect the branching structure of blood vessels, especially showing excellent performance when dealing with complex vascular networks. For example, when calculating the watershed water volume, the following formula can be used:

[0065] ,

[0066] where W(x) is the accumulated water volume at position x, U(x) is the upstream region of x, and f(y) is the local water volume at position y. In practical applications, the gray value or vesselness response value of the blood vessel can be used as a measure of the local water volume. By setting an appropriate threshold (for example, the accumulated water volume is greater than 1.5 times the average gray value of the image), the vascular region and the background can be effectively distinguished.

[0067] In the process of obtaining the original vascular centerline, the present invention adopts a multi-step processing method. First, a vascular enhancement algorithm is used to perform non-linear enhancement on the 3D CTA scan results to highlight the vascular structure and details. Here, the Frangi filter can be selected as the vascular enhancement algorithm, and its mathematical expression is:

[0068] ,

[0069] where V(s) is the vesselness measure, 、 、 are the eigenvalues of the Hessian matrix, 、 and represent the cross-sectional ratio, sphericity, and structural degree of the blood vessel respectively. The parameters 、 and control the sensitivity of the filter and can usually be set to = 0.5, = 0.5, c = half maximum intensity width / 2.

[0070] Next, the present invention uses the maximum radius filtering technique to identify the vascular centerline, filtering out the non-vascular regions of interest and small blood vessels. This step can effectively reduce the complexity of subsequent processing and improve the efficiency of the algorithm. In practice, the maximum radius threshold can be set to 1% of the image diagonal length, which can effectively remove noise and small branches while retaining the main blood vessels.

[0071] Subsequently, a rough original vascular centerline is obtained based on a three-dimensional reconstruction algorithm. Here, the minimum path method can be adopted, and the centerline is obtained by solving the Eikonal equation:

[0072] ,

[0073] where T(x) is the shortest path length to the starting point, P(x) is the cost function, which can be defined as , V(x) is the vesselness response, and k is the adjustment parameter, which can be set to a value between 2 and 5.

[0074] Finally, a refined vascular centerline is obtained using a neighborhood-based skeleton refinement technique. This step can further improve the accuracy of the centerline. A parallel refinement algorithm can be adopted, iteratively deleting boundary points until no more can be deleted. In each iteration, a 3x3x3 neighborhood can be used to determine whether to delete the current point, and the judgment conditions include connectivity preservation, endpoint preservation, etc.

[0075] Through the above steps, the method of the present invention can effectively reconstruct complex vascular networks and accurately extract the vascular centerline. This is of great significance for subsequent vascular analysis, disease diagnosis, and surgical planning. For example, in the diagnosis of cerebral aneurysms, accurate vascular reconstruction and centerline extraction can help doctors better observe the location, size, and shape of the aneurysm, thus formulating a more reasonable treatment plan.

[0076] The method of the present invention is not only applicable to the analysis of cerebral blood vessels, but can also be extended to the vascular analysis of other organs, such as coronary arteries, pulmonary blood vessels, etc. By adjusting relevant parameters, it can adapt to the characteristics of blood vessels in different organs and achieve wide applications.

[0077] In summary, the multi-level vascular network reconstruction and segmentation method based on watershed analysis proposed by the present invention realizes high-precision and high-efficiency vascular network reconstruction and segmentation by combining a variety of advanced image processing techniques and watershed analysis theory. This provides a powerful tool support for medical image analysis and clinical diagnosis.

[0078] In a preferred embodiment of the present invention, the process of reconstructing the 3D vascular network specifically includes obtaining a series of vascular regions through a vascular extraction algorithm, and then based on the vascular centerline extraction algorithm, reconstructing the 3D vascular network according to the skeleton of the obtained vascular regions. This method can effectively capture the geometric features of blood vessels and improve the accuracy of reconstruction.

[0079] Specifically, the vascular extraction algorithm can adopt a multi-scale Hessian filter. The basic principle of this algorithm is to utilize the tubular structure characteristics of blood vessels and enhance the vascular structure by analyzing the second-order derivative information of the image. The main steps of the algorithm are as follows:

[0080] 1. Perform Gaussian filtering on the input image to obtain smoothed images at different scales: , where is a 3D Gaussian kernel, is the standard deviation of the Gaussian kernel.

[0081] 2. Calculate the Hessian matrix at each scale: , where , etc. represent the second-order partial derivatives of the image.

[0082] 3. Perform eigenvalue decomposition on the Hessian matrix to obtain eigenvalues , , (assuming . , Half maximum intensity width / 2.

[0083] 4. Calculate the vesselness response at multiple scales and take the maximum value as the final result . By setting an appropriate threshold (for example, the vesselness response is greater than 0.3), a preliminary vascular region can be obtained.

[0084] Next, based on the vascular centerline extraction algorithm, the position and shape of the blood vessels can be further refined. A commonly used method is the centerline extraction algorithm based on distance transformation. Its main steps are as follows:

[0085] 1. Perform distance transformation on the vascular region to obtain a distance map .

[0086] 2. Calculate the gradient field of the distance map: ;

[0087] 3. Calculate the divergence of the gradient field: ;

[0088] 4. Find the local maximum points in the divergence field, and these points form the initial centerline.

[0089] 5. Use the minimum spanning tree algorithm to connect these points to form a continuous centerline.

[0090] In another embodiment of the present invention, different watershed regions are formed by calculating the accumulation of watershed water volume to generate the vascular region. This watershed analysis-based method can better reflect the branching structure of blood vessels. Specifically, the following formula can be used to calculate the accumulated water volume:

[0091] ,

[0092] where, is the accumulated water volume at position , is 's upstream region, is the position 's local water volume. In practical applications, the vesselness response value of blood vessels can be used as a measure of local water volume.

[0093] Preferably, in the process of obtaining the original blood vessel centerline, the present invention adopts a multi-step processing method. First, the 3DCTA scan results are nonlinearly enhanced using a blood vessel enhancement algorithm to highlight the blood vessel structure and details. Here, the Frangi filter can be selected as the blood vessel enhancement algorithm, and its mathematical expression has been described in detail above.

[0094] Next, the present invention uses the maximum radius filtering technique to identify the blood vessel centerline, filtering out non-blood vessel regions of interest and small blood vessels. The basic idea of maximum radius filtering is to calculate the maximum sphere radius that can be completely contained inside the blood vessel at each voxel position. The specific algorithm is as follows:

[0095] 1. Perform a distance transform on the binary blood vessel image to obtain the distance from each voxel to the nearest background point .

[0096] 2. At each voxel position , calculate the maximum inscribed sphere radius centered at this point

[0097] ,

[0098] where, is 's neighborhood.

[0099] 3. Set the radius threshold (for example, 1% of the image diagonal length), and retain the points with a radius greater than the threshold: .

[0100] 4. Morphologically thin the result to obtain the preliminary vascular centerline.

[0101] Subsequently, obtain the rough original vascular centerline based on the 3D reconstruction algorithm. Here, the minimum path method can be used to obtain the centerline by solving the Eikonal equation:

[0102] ,

[0103] where is the shortest path length to the starting point, is the cost function, which can be defined as , is the vesselness response, is the adjustment parameter, which can be set to a value between 2 and 5. The Eikonal equation can be solved using the Fast Marching Method.

[0104] Finally, obtain the refined vascular centerline using the neighborhood-based skeleton thinning technique. The commonly used parallel thinning algorithms are as follows:

[0105] 1. Initialize the boundary point set B, which contains all the foreground points adjacent to the background.

[0106] 2. For each point p in B, if it satisfies the following conditions, mark it as deletable: deleting p does not change the local connectivity; p is not an endpoint; deleting p does not cause excessive erosion;

[0107] 3. Delete all the marked points.

[0108] 4. Update the boundary point set B.

[0109] 5. Repeat steps 2 - 4 until no points can be deleted.

[0110] When judging the local connectivity, the Euler number can be used to ensure topological invariance. For a 3x3x3 neighborhood, the calculation formula for the Euler number E is:

[0111] ,

[0112] where V is the number of vertices, E is the number of edges, F is the number of faces, and C is the number of volumes. If the Euler number remains unchanged before and after deleting the point, it is considered that the local connectivity has not changed.

[0113] Through the above detailed algorithm description, the method of the present invention can effectively reconstruct complex vascular networks and accurately extract the centerlines of blood vessels. This is of great significance for subsequent vascular analysis, disease diagnosis, and surgical planning. For example, in the diagnosis of cerebral aneurysms, accurate vascular reconstruction and centerline extraction can help doctors better observe the location, size, and shape of aneurysms, thus formulating more reasonable treatment plans.

[0114] In addition, the method of the present invention can be further optimized and extended. For example, machine learning techniques can be introduced to automatically adjust algorithm parameters, improving the adaptability and robustness of the method. At the same time, considering the characteristics of blood vessels in different organs, specific preprocessing and postprocessing steps can be designed to adapt to different application scenarios.

[0115] In summary, the multi-level vascular network reconstruction and segmentation method based on watershed analysis proposed by the present invention realizes high-precision and high-efficiency vascular network reconstruction and segmentation by combining a variety of advanced image processing techniques and watershed analysis theory. This provides strong tool support for medical image analysis and clinical diagnosis and is expected to play an important role in the fields of cerebrovascular diseases, cardiovascular diseases, etc.

[0116] In another preferred embodiment of the present invention, the process of constructing the pressure field includes multiple steps, and the ingenious combination of these steps can effectively simulate the pressure distribution in blood vessels, providing an important basis for determining the watershed position subsequently.

[0117] First, the method of the present invention constructs a blood pressure field by combining the original blood vessel centerline and blood vessel imaging parameters. The core idea of this step is to estimate the pressure distribution in blood vessels using the geometric characteristics of blood vessels and hemodynamic principles. Specifically, the following model can be adopted:

[0118] ,

[0119] where represents the blood pressure at position , is the blood pressure at the inlet (usually can be set to 120 mmHg, which is the average value of normal human arterial blood pressure), is the distance along the blood vessel centerline from the inlet to position , is the blood vessel radius at position , is the attenuation coefficient (which can be set to a value between 0.1 - 0.5 according to experience).

[0120] It should be noted that the blood vessel radius The vessel radius can be estimated by analyzing the grayscale distribution around the original vessel centerline. For example, a Gaussian fit can be performed on a plane perpendicular to the centerline, and the standard deviation of the fitted curve can be used as an estimate of the vessel radius.

[0121] Next, the method of the present invention generates the gradient field of the blood pressure field and optimizes it. The calculation of the gradient field can use the central difference method:

[0122] ,

[0123] in,

[0124] ,

[0125] ,

[0126] ,

[0127] Here, h is the step size, which can usually be set to the voxel size.

[0128] In order to optimize the gradient field, anisotropic diffusion filtering can be used. This method can smooth the noise while preserving the edge information. Its mathematical expression is:

[0129] ,

[0130] Where I is the image intensity and c(x,y,z,t) is the diffusion coefficient, which can be defined as:

[0131] ,

[0132] K is a control parameter that can be set according to image characteristics and is usually 10% to 20% of the image gradient amplitude.

[0133] Finally, the method of the present invention calculates the pressure gradient of the blood pressure field. The pressure gradient reflects the rate of change of blood pressure in space and is crucial for understanding hemodynamic characteristics. The pressure gradient can be calculated using the following formula:

[0134] ,

[0135] in, is the blood density (approximately 1060 kg / m³), is the blood flow velocity vector. Blood flow velocity can be obtained using techniques such as phase contrast magnetic resonance imaging (PC-MRI), or estimated based on vascular geometry and fluid mechanics.

[0136] The pressure field constructed through the above steps provides an important basis for the subsequent determination of the watershed position. This method of constructing the pressure field based on the principles of fluid mechanics can better reflect the physiological characteristics of the vascular network, which helps to improve the accuracy and physiological relevance of vascular segmentation.

[0137] In another embodiment of the present invention, the process of determining the watershed position in the vascular network includes the following steps: First, project the original vascular centerline onto a two-dimensional space through a pressure gradient calculation method to obtain the projected vascular centerline; then, project the projected vascular centerline onto the vascular image and mark the position of the watershed.

[0138] Specifically, the pressure gradient calculation method can adopt the following formula:

[0139] ,

[0140] where, is the pressure gradient, is the blood density, is the acceleration due to gravity, is the height difference. In practical applications, the points on the vascular centerline can be projected onto a two-dimensional plane along the direction of the pressure gradient. The projection process can use the Runge-Kutta method for numerical integration:

[0141] ,

[0142] where, [

[0143] ,

[0144] ,

[0145] ,

[0146] ,

[0147] is the pressure gradient field, is the step size.

[0148] When projecting the projected vascular centerline onto the vascular image, the nearest neighbor interpolation or bilinear interpolation method can be adopted. The watershed position usually corresponds to the saddle point or local minimum point in the pressure gradient field. These characteristic points can be identified by analyzing the topological structure of the pressure gradient field.

[0149] In another embodiment of the present invention, perform a connectivity analysis on the vascular segmentation results, and based on the results of the connectivity analysis and the vascular segmentation results, obtain a reconstructed and complete multi-level vascular segmentation result. The specific steps of the connectivity analysis are as follows:

[0150] 1. Voxelize the vascular segmentation results and label each voxel as an independent region.

[0151] 2. Scan all connected regions and test the connections between each pair of connected regions. The connection test can adopt the following criteria:

[0152] ,

[0153] where and are two regions, represents the set of boundary voxels of region , and represents the cardinality of the set. If is greater than the preset threshold .

[0154] 3. Based on the connection test results and the input image vessels, merge the connected regions. The merging process can be efficiently implemented using the Union-Find data structure. Through this connectivity analysis, the problem of blood vessel breaks caused by noise or segmentation errors can be effectively handled, improving the integrity and continuity of blood vessel segmentation.

[0155] Finally, the method of the present invention further includes calculating the blood vessel length and volume based on the reconstructed 3D blood vessel network. The blood vessel length can be calculated by accumulating the Euclidean distances between adjacent points on the blood vessel centerline:

[0156] ,

[0157] where is the coordinate of the th point on the blood vessel centerline. The blood vessel volume can be calculated by the integration method:

[0158] ,

[0159] where is the blood vessel length, is the radius function along the blood vessel centerline. In the discrete case, it can be approximated as:

[0160] ,

[0161] where is the average radius of the th segment, is the length of the th segment.

[0162] By calculating these geometric parameters, doctors can gain a more comprehensive understanding of the morphological characteristics of blood vessels, providing important references for disease diagnosis and treatment plan formulation.

[0163] In summary, the multi-level blood vessel network reconstruction and segmentation method based on watershed analysis proposed in the present invention realizes high-precision and high-reliability blood vessel network reconstruction and segmentation through a series of innovative steps such as constructing an accurate pressure field, determining the watershed position, and performing connectivity analysis. This method not only considers the geometric characteristics of blood vessels but also incorporates hemodynamic principles, thus being able to better reflect the physiological characteristics of the blood vessel network.

[0164] In another preferred embodiment of the present invention, based on the reconstructed 3D blood vessel network, the blood vessel length and volume are further calculated. The accurate calculation of these geometric parameters is of great significance for comprehensively evaluating the morphological characteristics of blood vessels, assisting in disease diagnosis, and formulating treatment plans.

[0165] Specifically, the cumulative method is used to calculate the blood vessel length, that is, the Euclidean distance between adjacent points is accumulated along the blood vessel centerline. Its mathematical expression is as follows:

[0166] ,

[0167] where, is the total blood vessel length, represents the three-dimensional coordinates of the th point on the blood vessel centerline, is the total number of points on the centerline. To improve the calculation accuracy, the B-spline interpolation method can be used to smooth the centerline to reduce the error caused by discrete sampling.

[0168] The integral method is used to calculate the blood vessel volume, regarding the blood vessel as a series of continuous cylinders. Its mathematical expression is:

[0169] ,

[0170] where, is the total blood vessel volume, is the blood vessel length, is the radius function along the blood vessel centerline. In practical applications, since the blood vessel radius is discretely sampled, the numerical integration method can be used for approximate calculation:

[0171] ,

[0172] where, is the average radius of the th segment, is the The length of each segment. To improve the calculation accuracy, high-order numerical integration methods such as Simpson's method or Gauss-Legendre integration method can be adopted.

[0173] It should be noted that the estimation of the blood vessel radius is crucial for the accuracy of volume calculation. The present invention adopts a multi-planar reconstruction technique to improve the accuracy of radius estimation. Specifically, at each point on the blood vessel centerline, a plane perpendicular to the centerline is constructed, and the blood vessel contour is fitted on this plane. The fitting can adopt a circular or elliptical model, and the elliptical model can better adapt to blood vessels with non-circular cross-sections. The mathematical model of ellipse fitting is as follows:

[0174] ,

[0175] where, are the major and minor axes of the ellipse, is the tilt angle of the ellipse. The least squares method or RANSAC algorithm can be used in the fitting process to improve the robustness.

[0176] In addition, the present invention also considers the special treatment of blood vessel bifurcation points. Near the bifurcation points, a simple cylinder model may lead to deviations in volume calculation. For this reason, the present invention introduces a method for modeling bifurcation points based on Voronoi diagrams. The specific steps are as follows:

[0177] 1. Identify the bifurcation points in the blood vessel network.

[0178] 2. Construct a three-dimensional Voronoi diagram around the bifurcation points.

[0179] 3. Use the boundary of the Voronoi diagram to define the geometry of the bifurcation region.

[0180] 4. Calculate the volume of the Voronoi polyhedron as the volume of the bifurcation region.

[0181] This method can more accurately describe complex bifurcation structures, thereby improving the accuracy of the overall blood vessel volume calculation.

[0182] In another embodiment of the present invention, when intraoperative detection is required, the method further includes the following steps: First, based on the blood vessel centerlines of each independent image, a segmentation result is obtained; then, according to the obtained segmentation result, the positions of each blood vessel segment are adjusted; finally, when the blood vessel segmentation result and the blood vessel centerline change, the watershed is reconstructed to obtain an updated segmentation result.

[0183] The introduction of this dynamic update mechanism greatly improves the flexibility and adaptability of the present method in actual clinical applications. Specifically, the intraoperative detection process can be described as follows:

[0184] 1. Obtain segmentation results based on the vessel centerlines of each independent image. This step uses a fast centerline tracking algorithm, the core idea of which is to use known centerline information to perform local search and matching on the new image. Mathematically, this can be expressed as minimizing the following energy function:

[0185] ,

[0186] in, represents the parameterized centerline curve, 、 and Represent the image term, smoothing term and prior term respectively, 、 and is the weight coefficient.

[0187] 2. Adjust the position of each vessel segment. This step uses deformation registration technology to deform the previous segmentation results into the new image space. The deformation field can be obtained by solving the following partial differential equation: ,

[0188] in, is the deformation field, As driving force, 、 and is the control parameter.

[0189] 3. Reconstruct the watershed. When the vessel segmentation results and the vessel centerline change significantly, the watershed needs to be reconstructed to ensure the accuracy of the segmentation. Here, the fast watershed algorithm is used. Its core idea is to use the previous watershed result as an initialization and only perform local updates in the changed areas. The main steps of the algorithm are as follows:

[0190] a. Construct a minimum spanning tree (MST) of image gradients. The local watershed algorithm can be formulated as the following optimization problem: b. Extract the watershed line from the MST. c. Compare the old and new watershed lines to identify areas requiring update. d. Apply the local watershed algorithm to the areas requiring update. ,

[0191] in, Indicates segmentation, For areas that need to be updated, For the watershed basins, Pixels The gray value of for The average gray value.

[0192] Through this dynamic update mechanism, the method of the present invention can adapt to the changes in blood vessel morphology during the operation in real time, provide continuously updated blood vessel segmentation information for doctors, and thus support more accurate surgical navigation and decision-making.

[0193] In addition, in order to further improve the robustness and adaptability of the method, the present invention also introduces an adaptive parameter adjustment mechanism. Specifically, the system continuously monitors the quality of blood vessel segmentation and centerline extraction, and automatically adjusts the algorithm parameters according to the quality assessment results. The quality assessment can be based on the following indicators:

[0194] 1. Segmentation consistency: Use the Dice coefficient or Jaccard index to measure the consistency of segmentation results between consecutive frames.

[0195] 2. Centerline stability: Calculate the average displacement of the centerline between consecutive frames.

[0196] 3. Topological consistency: Check whether the topological structure of the blood vessel network has changed significantly.

[0197] Based on these indicators, the system uses a reinforcement learning algorithm to optimize the parameter settings. The state space of the reinforcement learning model includes the current image features and algorithm parameters, the action space is the adjustment direction and amplitude of the parameters, and the reward function is defined based on the above quality assessment indicators. In this way, the system can automatically select the optimal parameter settings under different imaging conditions and blood vessel morphologies, thus ensuring the stability and reliability of the method.

[0198] In summary, the multi-level blood vessel network reconstruction and segmentation method based on watershed analysis proposed by the present invention can not only accurately calculate the geometric parameters of blood vessels, but also has the ability of dynamic update and adaptive adjustment. These characteristics enable the method to better adapt to complex clinical scenarios.

[0199] The above; only the preferred specific embodiments of the present invention; but the protection scope of the present invention is not limited thereto; any person familiar with the art within the scope disclosed by the present invention; according to the solution of the present invention and its improved conceptions for equivalent replacement or change; should be covered within the protection scope of the present invention.

Claims

1. A method for reconstructing and segmenting a multi-level vascular network based on watershed analysis, characterized in that, Comprising: An obtaining step, comprising: Obtaining a three-dimensional computed tomography angiography (3DCTA) scan result; A processing step, comprising: Reconstructing a 3D vascular network based on the 3DCTA scan result; Performing image segmentation and enhancement processing on the 3DCTA scan result to obtain an original vascular centerline; Constructing a pressure field based on the original vascular centerline; Determining the watershed position in the vascular network based on the pressure field; Segmenting the vascular network according to the watershed position; An output step, comprising: Outputting the reconstructed 3D vascular network and the vascular segmentation result; The constructing of the pressure field specifically comprises: Constructing a blood pressure field by combining the original vascular centerline and vascular imaging parameters; Generating a gradient field of the blood pressure field and optimizing the gradient field; Calculating the pressure gradient of the blood pressure field; The determining of the watershed position in the vascular network specifically comprises: Projecting the original vascular centerline into a two-dimensional space by a pressure gradient calculation method to obtain a projected vascular centerline; Projecting the projected vascular centerline onto the vascular image to mark the position of the watershed.

2. The method according to claim 1, wherein The reconstructing of the 3D vascular network specifically comprises: Obtaining a series of vascular regions by a vascular extraction algorithm; Reconstructing a 3D vascular network based on the skeleton of the vascular regions according to a vascular centerline extraction algorithm.

3. The method according to claim 2, characterized in that, Forming different watershed regions by calculating the cumulative water volume of the watersheds to generate the vascular regions.

4. The method according to claim 1, wherein The obtaining of the original vascular centerline specifically comprises: Performing non-linear enhancement on the 3DCTA scan result by a vascular enhancement algorithm to highlight the vascular structure and details; Identifying the vascular centerline by using a maximum radius filtering technique to filter out non-vascular regions of interest and small blood vessels; Obtaining a rough original vascular centerline based on a three-dimensional reconstruction algorithm; Obtaining a refined vascular centerline by using a neighborhood-based skeleton refinement technique.

5. The method according to claim 1, characterized in that, Further comprising: Performing connectivity analysis on the vascular segmentation result; Based on the result of the connectivity analysis and the vascular segmentation result, obtaining a reconstructed and complete multi-level vascular segmentation result.

6. The method according to claim 5, characterized in that The connectivity analysis specifically comprises: Performing a voxelization operation on the vascular segmentation result to label each voxel as an independent region; Scanning all connected regions and testing the connection between each two connected regions; Merging the connected regions based on the connection test result and the input image blood vessels.

7. The method according to claim 1, characterized in that Further comprising: Calculating the length and volume of the blood vessels based on the reconstructed 3D vascular network.

8. The method according to claim 1, characterized in that, Further comprising: When intraoperative detection is required, obtaining a segmentation result based on the vascular centerlines of each independent image; Adjusting the positions of each vascular segment according to the segmentation result; When the vascular segmentation result and the vascular centerline change, reconstructing the watershed to obtain an updated segmentation result.

Images