An aneurysm segmentation landing zone positioning identification method and system based on an optimal blood flow guide model

Abstract

The application discloses an aneurysm segmentation landing zone positioning and identifying method and system based on an optimal blood flow guiding model, and the method comprises the following steps: collecting image data of a target blood vessel region, and pre-processing the image data to obtain a blood vessel enhanced image; performing binaryzation on the blood vessel enhanced image to obtain a blood vessel cavity binary mask; extracting a tree-shaped blood vessel center line from the blood vessel enhanced image by using a blood vessel center line extraction algorithm; analyzing the tree-shaped blood vessel center line to obtain a multi-modal fusion feature vector; predicting a neck aneurysm probability value of the multi-modal fusion feature vector by using a pre-constructed neck aneurysm detection deep learning model, and screening a candidate neck aneurysm region and a parent artery region; and performing membrane membrane on the candidate neck aneurysm region and the parent artery region by using a blood vessel cavity three-dimensional binary mask, and extracting a device anchoring landing zone. According to the scheme, the neck aneurysm precisely positioned by the membrane membrane is taken as a starting point, and an anchoring landing zone section can be accurately screened out by searching upstream and downstream along the parent artery center line.

Description

Technical Field

[0001] This invention relates to the field of imaging, and more particularly to a method and system for aneurysm segmentation and landing area localization and recognition based on an optimal blood flow guidance model. Background Technology

[0002] Intracranial aneurysms are abnormal bulges in the walls of intracranial arteries. Their rupture is a major cause of subarachnoid hemorrhage, with high mortality and disability rates. Flow diversion devices are innovative endovascular interventional treatments that reconstruct blood flow pathways at the aneurysm neck using a dense network stent, isolating the aneurysm from circulation, promoting thrombus formation, and ultimately healing. The success of this technology heavily relies on precise preoperative individualized planning. Its core challenge lies in how to automatically and accurately identify the aneurysm and its neck from the complex three-dimensional vascular anatomy, and then scientifically select a stable and suitable carrier artery segment as the device's "landing zone" to ensure stable anchoring, effective coverage of the aneurysm neck, and maintenance of branch vessel patency.

[0003] Currently, during preoperative assessment, physicians typically manually delineate the boundaries of the aneurysm and its neck on two-dimensional image slices or three-dimensional reconstructed models. This process is time-consuming and highly dependent on individual experience, leading to significant differences in assessment results among different physicians, thus affecting the consistency of the team's evaluation. Furthermore, the accuracy of aneurysm neck localization is insufficient. As a crucial interface for device anchoring, the precise three-dimensional spatial position and geometry (such as width and angle) of the aneurysm neck directly impact the selection of device model and release effectiveness. Existing methods often struggle to achieve automated, high-precision three-dimensional localization of the aneurysm neck.

[0004] In recent years, deep learning-based vascular segmentation technology has made progress, enabling the automatic segmentation of vascular trees from medical images. However, existing technologies mostly focus on the overall segmentation of vascular networks or the simple detection of aneurysms, lacking the ability to accurately identify the anatomical structure of aneurysms to determine the implantation location of the treatment device. Specifically, there is still no method that can systematically address the aforementioned technical problems. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for locating and identifying the landing zone of aneurysm segmentation based on an optimal blood flow guidance model, which solves the above-mentioned technical problems pointed out in the prior art.

[0006] This invention provides a method for aneurysm segmentation and landing zone localization and identification based on an optimal blood flow guidance model, comprising the following steps:

[0007] Image data is acquired from the target vascular region, and the image data is preprocessed to obtain an enhanced vascular image;

[0008] The enhanced blood vessel image is binarized to obtain a binary mask of the blood vessel cavity; the binary mask of the blood vessel cavity is projected into three-dimensional space to obtain a three-dimensional binary mask of the blood vessel cavity; the three-dimensional binary mask of the blood vessel cavity includes the blood vessel cavity, the blood vessel wall, and the complete blood vessel network;

[0009] A tree-like vascular centerline is extracted from the enhanced vascular image using a vascular centerline extraction algorithm; the centerline path points of the tree-like vascular centerline are collected and analyzed to obtain a multimodal fusion feature vector; a pre-built aneurysm neck detection deep learning model is used to predict the aneurysm neck probability value from the multimodal fusion feature vector, and candidate aneurysm neck regions and aneurysm-bearing artery regions are screened; the candidate aneurysm neck regions and aneurysm-bearing artery regions are masked using the three-dimensional binary mask of the vascular lumen to extract the device anchoring and landing area.

[0010] Accordingly, this invention also proposes an aneurysm segmentation landing area localization and identification system based on an optimal blood flow guidance model, comprising: an acquisition module; and an identification module;

[0011] The acquisition module is used to acquire image data of the target blood vessel region, preprocess the image data to obtain an enhanced blood vessel image, binarize the enhanced blood vessel image to obtain a binary mask of the blood vessel lumen, and project the binary mask of the blood vessel into three-dimensional space to obtain a three-dimensional binary mask of the blood vessel lumen.

[0012] The identification module is used to extract the tree-like vessel centerline from the enhanced vessel image using a vessel centerline extraction algorithm; analyze the centerline path points collected from the tree-like vessel centerline to obtain a multimodal fusion feature vector; use a pre-built aneurysm neck detection deep learning model to predict the aneurysm neck probability value from the multimodal fusion feature vector, and screen candidate aneurysm neck regions and aneurysm-bearing artery regions; use the three-dimensional binary mask of the vessel lumen to mask the candidate aneurysm neck regions and aneurysm-bearing artery regions, and extract the device anchoring landing area.

[0013] Compared with the prior art, the embodiments of the present invention have at least the following technical advantages:

[0014] Analysis of the above-mentioned aneurysm segmentation landing area localization and identification method and system based on the optimal blood flow guidance model provided by the present invention shows that, in specific applications, the medical image data of the target vascular region is first preprocessed, including denoising and vascular structure enhancement, to generate vascular enhancement images, and then further binarized to obtain a three-dimensional binary mask of the vascular cavity, so as to clearly characterize the vascular network topology.

[0015] Furthermore, the centerline of the dendritic vessels is extracted from the enhanced image, and local geometric features (curvature, radius) are calculated point by point along the centerline. Combined with the depth image features extracted from the local two-dimensional neighborhood of the image, a multimodal fusion feature vector is constructed. The vector sequence is input into a pre-trained deep learning model for aneurysm neck detection (such as a sequence model) to predict the probability that each point belongs to the aneurysm neck, and then continuous candidate aneurysm neck regions and corresponding aneurysm-bearing artery regions that meet the length threshold are selected.

[0016] Furthermore, starting from the precisely located aneurysm neck, the search proceeds upstream and downstream along the centerline of the aneurysm-bearing artery, and the anchoring and landing sections of the candidate device are determined based on geometric stability conditions (such as length, curvature, rate of change of diameter, and branch distance). Attached Figure Description

[0017] Figure 1 This is a flowchart of the main steps of an aneurysm segmentation landing area localization and identification method based on an optimal blood flow guidance model, as described in Example 1.

[0018] Figure 2 This is a flowchart of the candidate aneurysm neck region in an aneurysm segmentation and landing area localization and identification method based on an optimal blood flow guidance model, as described in Example 1.

[0019] Figure 3 This is a schematic diagram of the tree-like vessel centerline of an aneurysm segmentation landing area localization and identification method based on an optimal blood flow guidance model, as shown in Example 1.

[0020] Figure 4 This is a partial schematic diagram of the tree-like vessel centerline of an aneurysm segmentation landing area localization and identification method based on an optimal blood flow guidance model, as described in Example 1.

[0021] Figure 5 This is a flowchart of a candidate aneurysm region in an aneurysm segmentation and landing area localization and identification method based on an optimal blood flow guidance model, as described in Example 1.

[0022] Figure 6 This is a schematic diagram of the narrowest cross-section of the aneurysm neck in an aneurysm segmentation and landing area localization and identification method based on an optimal blood flow guidance model, as described in Example 1.

[0023] Figure 7 This is a flowchart of the final landing area of ​​an aneurysm segmentation landing area localization and identification method based on an optimal blood flow guidance model, as described in Example 1.

[0024] Figure 8 This is a flowchart of an aneurysm segmentation landing area localization and identification system based on an optimal blood flow guidance model, as described in Example 2.

[0025] Labels: Acquisition module 10; Recognition module 20. Detailed Implementation

[0026] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings.

[0028] Example 1

[0029] like Figure 1 As shown, this embodiment of the invention provides a method for aneurysm segmentation and landing area localization and identification based on an optimal blood flow guidance model, including the following steps:

[0030] S10: Acquire image data of the target blood vessel region, preprocess the image data to obtain an enhanced blood vessel image;

[0031] The enhanced blood vessel image is binarized to obtain a binary mask of the blood vessel cavity; the binary mask of the blood vessel cavity is projected into three-dimensional space (three-dimensional space of x-axis, y-axis, and z-axis) to obtain a three-dimensional binary mask of the blood vessel cavity; the three-dimensional binary mask of the blood vessel cavity includes the blood vessel cavity (i.e., the hollow part in which blood flows inside the blood vessel), the blood vessel wall (i.e., the surface wall of the blood vessel cavity, that is, the blood vessel wall surrounds the blood vessel cavity) and a complete blood vessel network (i.e., the blood vessels are interwoven, and a clear blood vessel network can be obtained after binarization, that is, the topological map of the blood vessel direction);

[0032] It should be noted that the preprocessing of the acquired image data includes denoising and enhancing the brightness of pixels in the vascular structure to obtain an enhanced vascular image.

[0033] S20: Extract tree-like vessel centerlines from the enhanced vessel image using a vessel centerline extraction algorithm; analyze the centerline path points collected from the tree-like vessel centerlines to obtain a multimodal fusion feature vector; predict the aneurysm probability value using a pre-built aneurysm neck detection deep learning model to screen candidate aneurysm neck regions and aneurysm-bearing artery regions; use the three-dimensional binary mask of the vessel lumen to mask the candidate aneurysm neck regions and aneurysm-bearing artery regions to extract the device anchoring landing area (i.e., locate the most suitable position for placing the treatment device, which is the treatment device and also the final landing area);

[0034] It should be noted that the vascular centerline extraction algorithm extracts the main paths of blood vessels in the enhanced vascular image, forming a tree structure, which helps to understand the overall layout and branching of blood vessels. Multiple centerline path points are selected along the extracted centerline, and for each centerline path point, its geometric features (such as curvature, diameter, etc.) and deep features based on local image patches are calculated. These two types of features are then combined to construct a multimodal fusion feature vector reflecting the comprehensive information of the point. A pre-trained deep learning model is used to analyze the above feature vector to predict the probability that each centerline path point will become a tumor neck. Based on the prediction results, regions that may contain tumor necks and the corresponding tumor-bearing artery are selected. Finally, a three-dimensional binary mask of the vascular lumen is used to further process the candidate tumor neck regions and tumor-bearing artery regions to accurately locate the most suitable position for placing the treatment device, ensuring the safety and effectiveness of subsequent surgery.

[0035] Specifically, such as Figure 2 As shown, in step S20, a tree-like vascular centerline is extracted from the enhanced vascular image using a vascular centerline extraction algorithm; the centerline path points of the tree-like vascular centerline are collected and analyzed to obtain a multimodal fusion feature vector; a pre-built aneurysm neck detection deep learning model is used to predict the aneurysm neck probability value from the multimodal fusion feature vector, and candidate aneurysm neck regions and aneurysm-bearing artery regions are selected; the candidate aneurysm neck regions and aneurysm-bearing artery regions are masked using the 3D binary mask of the vascular lumen to extract the device anchoring landing area. The specific operation steps are as follows:

[0036] S21: Extract a preset size image block from the enhanced blood vessel image, and obtain the blood vessel inlets and multiple blood vessel outlets in the extracted image block (i.e., the blood vessel inlets and outlets can be extracted from the edges of the extracted image block (or the blood vessel inlets and outlets can be extracted based on the complete blood vessel network in the three-dimensional binary mask of the blood vessel cavity), and the blood vessel inlets and outlets at the edges are determined based on the blood vessel vertices and end blood vessel points at the edges).

[0037] The blood vessel centerline extraction algorithm is used to track blood vessels from their inlets to their outlets to form initial paths (i.e., the paths of the blood vessels; although the blood vessel network has been described in step S10, the blood vessels are intertwined in the network, so it is necessary to extract the blood vessel paths through the inlets and outlets so that the main paths can be selected as the centerlines). Each initial path is discretized to form multiple centerline path points (i.e., the centerline path points are the path points collected from the blood vessel cavities in each initial path, which are also the center points), and the three-dimensional spatial coordinates of each centerline path point are determined. The blood vessel diameter of each centerline path point in the initial path is calculated (i.e., the radius of each centerline path point to the pixel point on the blood vessel wall is calculated to obtain the blood vessel diameter; the blood vessel diameter of the centerline path point represents the thickness of the blood vessel, thereby selecting the centerline blood vessels in the blood vessel network).

[0038] The initial paths are selected based on the largest vessel diameter. The largest vessel diameters in all initial paths are sorted, and the initial path with the largest vessel diameter is used as the centerline of the tree-like vascular network (i.e., the tree-like vascular network centerline includes branch nodes such as branches, parent branches, and child branches; the vessel diameter represents the thickest vessel among all vessels, and is used as the centerline of the vascular network). The remaining initial paths are used as branch vessels (i.e., the connections between branch vessels and the tree-like vascular network centerline are branch nodes, which are automatically obtained and will not be elaborated further). Figure 3 As shown, Figure 4 (as shown)

[0039] It should be noted that the process involves extracting the tree-like centerline, determining the entry (root) and exit (leaf), calculating the distance from the entry to the exit, and tracing backwards to form the initial path. The path is then discretized, and the vessel diameter at each centerline path point is calculated. The path with the largest diameter is selected as the "tree-like vessel centerline." Because vascular images (CTA / MRA) are three-dimensional volumetric data with a massive amount of data, extracting the centerline transforms it into a one-dimensional curve, greatly reducing computational complexity. Aneurysms typically occur at vessel bifurcation points or on the sidewall of the parent artery. By selecting the path with the largest diameter, the algorithm can automatically locate the "main trunk vessel" (i.e., the parent artery) in the vascular network, excluding interference from small capillary branches.

[0040] S22: Take several adjacent centerline path points from one of the centerline path points in the tree-like vascular centerline to form a local centerline segment;

[0041] Fit a local curve segment to the local centerline segment, and calculate the curvature of the centerline path point of the local curve segment (i.e., the centerline path point is "one of the above-mentioned centerline path points"; curvature represents the degree of bending of the centerline path point in the local curve segment (or the centerline of the dendritic vessel). Near the aneurysm neck, the curvature of the aneurysm-bearing artery usually undergoes abrupt changes).

[0042] The vessel radius is obtained by measuring the vessel diameter at the centerline path point; the curvature and vessel radius are combined to construct an initial geometric feature vector (that is, the initial geometric feature vector of the centerline path point, which is a two-dimensional vector; the initial geometric feature vector of each centerline path point is obtained through this method); the two are combined into a two-dimensional vector (such as curvature and radius) to describe morphological and dimensional information, which conforms to the basic logic of geometric feature expression.

[0043] Each centerline path point in the tree-like vascular centerline is mapped into a three-dimensional space (i.e., a three-dimensional space of x, y, z) to determine the three-dimensional coordinates of each centerline path point; a three-dimensional cubic region is uniformly divided around the three-dimensional coordinates of each centerline path point (i.e., the image block size is 32x32x32 voxels, and this neighborhood contains the original image information of the vessel wall, lumen, and any possible tumors around the centerline path point pi).

[0044] A convolutional neural network (CNN) is pre-constructed (i.e., this CNN is an encoder structure, such as the encoding part of 3D U-Net, VoxCNN, etc.). A 3D cube region is input into the CNN to obtain a depth feature vector (i.e., the 3D cube region is passed through the first convolutional layer of the CNN (extracting low-level features) and multiple (e.g., 32) 3D convolutional kernels are applied to slide across the image patch, generating 32 feature maps, each highlighting different types of low-level patterns (such as edges, blobs, gradients in specific directions) within the 3D cube region). A non-linear transformation is applied to the convolution results, enabling the network to learn more complex patterns. Pooling layers (usually 3D max pooling) are used to represent the maximum value within a small region (e.g., 2×2×2), reducing the amount of data and making the features robust to small positional changes, ensuring that the neck's precise location is detected regardless of whether it is at the center of the patch or slightly offset. This process is repeated multiple times, such as with 4-5 convolutions. As the pooling layers deepen, the receptive field (the area of ​​the original image that each neuron can "see") becomes larger and larger. Edges, corners, and curved segments and tubular structures of blood vessels are identified. After several layers of convolution and pooling, 512 feature maps are obtained. The average value of all values ​​within each feature map is calculated, and each feature map (representing the presence of a specific high-level pattern in the entire image) is compressed into a scalar (i.e., the scalar represents the average intensity or confidence of the high-level pattern appearing in the entire image patch). This results in a 512-dimensional vector. Each dimension of this vector corresponds to a high-level, semantic image pattern learned by the network during training. For example, the 101st dimension might strongly respond to the "discontinuous blood vessel wall" pattern; the 205th dimension might respond to the "tumor region" pattern (the operation of convolutional neural networks is common knowledge and will not be elaborated further).

[0045] The initial geometric feature vector and the deep feature vector (i.e., the deep feature vector output by the convolutional neural network) are concatenated by a nonlinear transformation to obtain a multimodal fusion feature vector;

[0046] It should be noted that the initial geometric feature vector (low-dimensional, such as 3D) and the depth feature vector (high-dimensional, such as 512D) are first aligned in dimensions. The above steps map the low-dimensional initial geometric feature vector to a dimension more compatible with the depth feature vector by performing a non-linear transformation on it (such as through a small fully connected layer). The two feature vectors are then concatenated in the dimensional direction; for example, concatenating a 2D vector and a 512D vector results in a 514-dimensional vector, which is the multimodal fusion feature vector. The multimodal fusion feature vector contains both accurate, physically based geometric information and rich, data-driven image information.

[0047] S23: Pre-construct a deep learning model for aneurysm detection; input the multimodal fusion feature vectors of all centerline path points into the deep learning model for aneurysm detection according to the order of each centerline path point in the tree-like vascular centerline, to obtain the aneurysm probability value (i.e., the deep learning model for aneurysm detection is a sequence model or a fully connected network that considers the context (i.e., the convolutional neural network in the above steps). The deep learning model for aneurysm detection outputs a probability value between 0 and 1 for each centerline path point Pi, representing the confidence that the centerline path point belongs to the aneurysm region. This is common knowledge and will not be elaborated further).

[0048] A preset probability threshold is set; it is then determined whether the probability value of the aneurysm neck is greater than the probability threshold.

[0049] If so, the centerline path point is determined as a high-probability centerline path point, and consecutive high-probability centerline path points are combined to form a high-probability segment.

[0050] A minimum length threshold is preset; the Euclidean distance between adjacent high-probability centerline path points in the high-probability segment is calculated to obtain the path length of the high-probability segment; it is then determined whether the path length is greater than the minimum length threshold.

[0051] If so, then the high-probability fragment is determined to be a candidate tumor neck region;

[0052] If not, then it is determined to be the region of the tumor-bearing artery;

[0053] It should be noted that the aneurysm neck mentioned above is not an isolated point, but a region; a sequence model is used to consider the preceding and following relationships between central path points; for example, if the curvature of a central path point suddenly increases and the diameters of its preceding and following neighbors also change, the model will be more certain that this is a bifurcation or aneurysm neck; a minimum length threshold is set because vascular noise or small protrusions may generate high-probability values, but are usually very short; the step uses a minimum length to filter out these false positives, ensuring that what is detected is a true vascular lesion (such as an aneurysm neck); an aneurysm neck is usually a narrow "neck," so the corresponding high-probability segment is shorter; while the parent artery is a segment of blood vessel, so the corresponding segment is longer;

[0054] S24: Using the three-dimensional binary mask of the vascular cavity, masks for the candidate aneurysm neck region and the carrier artery region are extracted to obtain a preliminary candidate aneurysm region mask; the preliminary candidate aneurysm region mask and the carrier artery region mask are judged to have similar gray-level distribution (that is, the voxel gray levels of the candidate aneurysm neck region mask and the carrier artery region mask are similar, hence the term "similar gray-level distribution"); the preliminary candidate aneurysm region mask and the carrier artery region mask with similar gray-level distribution are merged to obtain the intersection region; the intersection region is cut into cross-sections to calculate the area, and the narrowest cross-section of the aneurysm neck is selected; the narrowest cross-section of the aneurysm neck is identified by the anchoring and landing area of ​​the identification device;

[0055] It should be noted that the above steps cannot determine the width of the aneurysm neck solely based on the centerline. Combining this with the original vascular lumen mask (i.e., the boundary of the vessel wall) is used for three-dimensional reconstruction and cutting. In interventional therapy, the width of the aneurysm neck is a key indicator for determining the treatment plan (such as whether stent-assisted intervention is suitable). Finding the narrowest cross-section identifies the bottleneck of the aneurysm neck. The landing zone refers to the area where the stent or microcatheter needs to cross the aneurysm neck and remain stable against the vessel wall. By calculating the intersection and cross-section, the algorithm can calculate the diameter, length, and shape of the landing zone, determining the appropriate stent diameter and length, as well as the stent deployment location, thereby avoiding the stent covering important collateral vessels.

[0056] Specifically, such as Figure 5 As shown, in step S24, the candidate aneurysm neck region and the carrier artery region are extracted using the three-dimensional binary mask of the blood vessel lumen, and a preliminary candidate aneurysm region mask is extracted; the grayscale distribution of the preliminary candidate aneurysm region mask and the carrier artery region mask is judged to be similar; the preliminary candidate aneurysm region mask and the carrier artery region mask with similar grayscale distribution are merged to obtain the intersection region; the area of ​​the intersection region is calculated by cutting the cross-section, and the narrowest cross-section of the aneurysm neck is selected; the narrowest cross-section of the aneurysm neck is identified by the anchoring and landing area of ​​the identification device. The specific operation steps are as follows:

[0057] S241: Extract each candidate aneurysm neck region from the three-dimensional binary mask of the vascular cavity to obtain the candidate aneurysm neck region mask; extract the vascular cavity voxels from the candidate aneurysm neck region mask (i.e., the candidate aneurysm neck region is a two-dimensional image, while the three-dimensional binary mask of the vascular cavity is a three-dimensional image, and a voxel is a unit of three-dimensional volume, so first convert the candidate aneurysm neck region into a three-dimensional image, and then use the three-dimensional coordinates of the candidate aneurysm neck region to find the three-dimensional coordinates of the corresponding three-dimensional binary mask of the vascular cavity, that is, the candidate aneurysm neck region mask, and extract the corresponding voxels).

[0058] All vascular lumen elements of the candidate aneurysm neck region mask are used as the initial seed point set (i.e., the initial seed point is considered as the interface connecting the aneurysm and the parent artery).

[0059] Extract each tumor-bearing artery region corresponding to the three-dimensional binary mask of the vascular lumen to obtain the tumor-bearing artery region mask; extract the vascular lumen voxels in the tumor-bearing artery region mask;

[0060] Calculate the average intensity of the voxels in the masked vascular cavity of adjacent tumor-bearing artery regions (i.e., the average gray level of the voxels).

[0061] Determine whether the voxel intensity of the initial seed point is greater than the average intensity of the voxel intensity of the mask of the tumor-bearing artery region (i.e., starting from the seed point, check its 26 neighborhoods (the average intensity of the voxel intensity of the voxel intensity of the mask of the tumor-bearing artery region in three-dimensional space, including the top, bottom, left, right, front, back, and all diagonal areas). Here, the determination is made by comparing the average intensity (i.e., voxel intensity) with the gray value of the voxel intensity of the initial seed point.

[0062] If so, the vascular lumen voxels of the adjacent tumor-bearing artery region mask of the initial seed point are determined as new seed point voxels (i.e., the voxels after growth are new seed points and the above steps are continued). The above calculation of the average intensity of the vascular lumen voxels of the adjacent tumor-bearing artery region and the new seed point voxels are repeated until the new seed point voxels stop growing, and a preliminary candidate tumor region mask is obtained (i.e., all the grown voxels are collected to form a preliminary candidate tumor region mask).

[0063] It should be noted that the preceding steps (such as S23) only provide a high-probability path along the centerline, but do not provide the complete three-dimensional shape of the aneurysm. Interventional treatment requires knowledge of the actual volume of the aneurysm and its connection to the blood vessel. The blood flow within the aneurysm cavity is usually connected to the parent artery, so the signal intensity (grayscale) is similar in the image. Region growing utilizes this prior knowledge to extend from the seed point to the low impedance (high grayscale) region, outlining the possible aneurysm cavity. The above comparison with the average grayscale of the parent artery can prevent the algorithm from mistaking nearby high-density structures (such as calcifications and bones) for aneurysms.

[0064] S242: Gaussian kernel density estimation is performed on the vascular cavity elements of the preliminary candidate tumor region mask and the vascular cavity elements of the tumor-bearing artery region mask to obtain the gray value probability density of the vascular cavity elements of the preliminary candidate tumor region mask and the gray value probability density of the vascular cavity elements of the tumor-bearing artery region mask.

[0065] The Batachalia coefficient (i.e., the Batachalia coefficient is used to measure the similarity between two probability distributions) is calculated using the gray value probability density of the vascular lumen elements of the preliminary candidate tumor region mask and the gray value probability density of the vascular lumen elements of the tumor-bearing artery region mask. The formula for calculating the Batachalia coefficient is as follows: ,in, Represented as the grayscale values ​​of all vascular lumen pigments. and This represents the probability density of voxels in the vascular cavities of the preliminary candidate tumor region mask and the tumor-bearing artery region mask at this gray value (estimated by Gaussian kernel density). In the continuous case, a Batachalia coefficient (BC) close to 1 indicates that the voxel gray value distributions of the two regions are almost identical; close to 0 indicates almost no overlap.

[0066] Set a preset similarity threshold; determine whether the Batachalia coefficient is greater than the similarity threshold;

[0067] If so, the vascular lumen of the preliminary candidate aneurysm region mask is determined to have a similar grayscale distribution to the vascular lumen of the carrier artery region mask. Studies have shown that the blood flow in the actual aneurysm lumen and the blood flow in the carrier artery are homologous, and therefore should have similar signal intensity distributions in medical images (such as CTA and MRA). This step verifies the consistency between the candidate aneurysm region and the normal vascular region in terms of imaging characteristics, which helps to exclude some imaging artifacts or adjacent non-vascular structures (such as bones and calcifications).

[0068] It should be noted that the Gaussian kernel density estimation mentioned above is a non-parametric method used to estimate the probability density function (PDF) of a random variable. By placing a Gaussian kernel (bell curve) on the gray value of each voxel and then superimposing all kernel functions, the probability of the gray value of each voxel can better reflect the continuous characteristics of the gray distribution (i.e., continuity is the density formed after the probability of occurrence; for example, the higher the probability of voxels with the same gray value, the more voxels with the same gray value there are, and the higher the density).

[0069] S243: Using a 3D morphological dilation operation, the initial candidate tumor region mask with similar gray-level distribution is merged with the tumor-bearing artery region mask to obtain the intersection region (i.e., the dilated structural element is a small sphere with a radius of 1-2 voxels, bridging the tiny gaps at the junction of the two initial candidate tumor region masks and the tumor-bearing artery region mask that may be caused by the resolution limitations or segmentation errors of the vascular enhancement image; the two dilated region masks are merged; then 3D connected component analysis is applied. If, in this merged mask, the region originally belonging to A and the region originally belonging to P belong to the same connected component, then they are determined to be connected in 3D space, that is, the intersection region).

[0070] The intersection region is vertically cut into a cross-sectional plane (that is, the intersection region of the blood vessel is vertically cut to form many cross-sections, so as to find the narrowest bottleneck at the connection point (that is, it may be a tumor, because the presence of a tumor will cause blood vessel congestion, so the blood flow in the cross-section will become a small pore bottleneck)). The cross-sectional area is calculated by projecting a two-dimensional plane onto each cross-sectional plane.

[0071] All cross-sectional areas are sorted in descending order, and the narrowest cross-sectional area is selected. If the narrowest cross-sectional area is located within the candidate tumor neck region, then the narrowest cross-sectional area is determined to be the narrowest cross-section of the tumor neck. (That is, although the candidate tumor neck region was obtained in step S241 above, it only provides an approximate location of the tumor neck along the centerline of the dendritic vessels, indicating roughly which segment of the centerline the tumor neck is along (i.e., the range of the centerline path points). The candidate tumor neck region can only estimate a local radius at the centerline point, but the tumor neck is often not a perfect circle and is not necessarily perpendicular to the centerline. Therefore, based on the measurement of the narrowest cross-section, the minimum area of ​​the connecting part of the actual anatomical structure is directly measured, and the equivalent diameter, major axis, minor axis, etc., can be calculated, making the selection more accurate. At the same time, the candidate tumor region is used as a probe or bridge.) This process cross-validates and precisely locates the aneurysm neck, the most critical structure, while confirming the authenticity of the entire aneurysm structure. It verifies the anatomical connectivity between the two regions (the candidate aneurysm region and the parent artery region), identifying the narrowest point at this connection as the true and precise location of the aneurysm neck. Furthermore, research has shown that if one region is an abnormal bulge (the candidate aneurysm region) and the other a normal vessel (the parent artery region), the interface is the boundary between normal and abnormal. Anatomically, this boundary is the neck of the aneurysm. Therefore, the narrowest connection between the two connected three-dimensional regions is the information bottleneck, the only channel for the exchange of substances (blood flow) between the two regions, and the only path for blood flow into and out of the aneurysm—the aneurysm neck. Figure 6 (as shown)

[0072] It should be noted that vessel segmentation algorithms often exhibit "disconnections" at the aneurysm neck due to resolution limitations or insufficient contrast. The dilation operation can bridge the gap of 1–2 voxels, ensuring that the aneurysm and artery are considered connected. A true aneurysm must be physically connected to the parent artery, and connectivity component analysis can exclude structures that are similar in grayscale but are actually separate (such as adjacent cysts). Aneurysms are usually gourd-shaped, consisting of the parent artery, a stenotic aneurysm neck, and a dilated aneurysm body. Therefore, the minimum cross-sectional area at the connection point is the aneurysm neck, which is a key anatomical parameter for determining whether interventional treatment (such as stent-assisted embolization) is suitable.

[0073] S244: Using the geometric center point of the narrowest cross-section of the aneurysm neck, search starting points are selected on the central line of the dendritic vessels. The search starting points are used to move upstream and downstream to obtain the upstream endpoint and the downstream endpoint, forming candidate landing segments. The distance between the candidate landing segments is calculated based on the upstream endpoint and the downstream endpoint. The safe distance between the branch nodes in the central line of the dendritic vessels and the candidate landing segments is calculated. Boundary rules are set for the candidate landing segments with safe distances to simulate a complete cardiac cycle, calculate the non-uniformity index, and determine the device anchoring landing area.

[0074] Specifically, such as Figure 7 As shown, in step S244, the geometric center point of the narrowest cross-section of the aneurysm neck is used to screen the search starting point on the central line of the dendritic vessels. The search starting point is then used to move upstream and downstream to obtain the upstream and downstream endpoints, forming candidate landing segments. The distance between the candidate landing segments is calculated based on the upstream and downstream endpoints. The safe distance between the branch nodes in the central line of the dendritic vessels and the candidate landing segments is calculated. Boundary rules are set for the candidate landing segments with safe distances to simulate a complete cardiac cycle, calculate the non-uniformity index, and determine the device's anchoring landing area. The specific operation steps are as follows:

[0075] S2441: Determine the geometric center point (i.e., the center point of the cross-section) of the narrowest cross-section of the aneurysm neck.

[0076] The Euclidean distance to the geometric center point is calculated using each centerline path point on the centerline of the dendritic blood vessel, and the shortest Euclidean distance is selected as the search starting point (that is, the geometric center point represents the center point of the cross section, while the centerline path point is a point collected on the centerline of the dendritic blood vessel according to a fixed step size, and the two are not the same point).

[0077] The initial cumulative path length is preset; the search starting point is used to move the centerline path point upstream of the tree-like blood vessel centerline (that is, to move against the direction of blood flow), the distance of the cumulative path length is calculated, and the curvature value and blood vessel radius of the current centerline path point are extracted (that is, the curvature value and blood vessel radius of the centerline path point are all obtained in the above steps and will not be repeated).

[0078] Calculate the rate of change of the vessel radius between the search starting point and the current centerline path point (which reflects the degree of expansion or contraction of the vessel diameter relative to the diameter at the aneurysm neck).

[0079] Set a maximum search length; determine whether the distance of the accumulated path length is greater than the maximum search length;

[0080] If so, the current centerline path point is taken as the upstream endpoint (i.e., the blood flow guide device needs to have sufficient anchoring length (usually covering 10-20mm on both sides of the aneurysm neck), but the search range should not be expanded indefinitely and should be within a reasonable anatomical range).

[0081] A preset local curvature threshold is established; it is then determined whether the curvature value of the current centerline path point is greater than the local curvature threshold.

[0082] If so, the current centerline path point is taken as the upstream endpoint (i.e., in excessively tortuous vascular segments, the device is difficult to adhere to the vessel wall, which may lead to "bridging" or poor endothelialization, increasing the risk of thrombosis; at the same time, high curvature means complex local hemodynamics).

[0083] A preset radius change rate threshold is established; it is then determined whether the blood vessel radius change rate exceeds the radius change rate threshold.

[0084] If so, the current centerline path point is taken as the upstream endpoint (i.e., the area where the blood vessel diameter changes drastically (severe stenosis), the radial support force of the device is unevenly distributed, and it is easy to cause displacement or blood vessel damage. The ideal anchoring area should have a relatively uniform diameter).

[0085] If the current centerline path point satisfies any of the above conditions, then the current centerline path point is taken as the upstream endpoint; otherwise, the current centerline path point is taken as the new search starting point, and the above steps are repeated to move upstream to the centerline path point until the conditions are met and the movement stops.

[0086] Using the search starting point, move the centerline path point of the tree-like blood vessel to the downstream centerline (in the direction of blood flow), repeat the same steps above, and filter out the downstream endpoints (that is, the filtering of downstream endpoints is the same as the filtering of upstream endpoints in the above steps, except that the direction of movement is changed from moving against the upstream to moving downstream in the direction of blood flow).

[0087] Connect the upstream endpoint and the downstream endpoint to form a candidate landing section (that is, this also includes many centerline path points, and the candidate landing section is also a segment of a blood vessel).

[0088] S2442: Project the candidate landing segment into three-dimensional space and calculate the Euclidean distance between the upstream and downstream endpoints as the candidate landing segment distance (i.e., the candidate landing segment distance is obtained by summing the distances between adjacent centerline path points; the candidate landing segment distance is the three-dimensional arc length of the vessel centerline, accurately reflecting the actual length of the vessel segment, and is the most direct parameter for determining the required length of the blood flow guiding device; too short a length may lead to unstable anchoring and easy displacement; the projection into three-dimensional space is for identifying the landing area, or for making the installation of the blood flow guiding device more solid and three-dimensional in three-dimensional space).

[0089] Convert the vessel radius of each centerline path point in the candidate landing segment to a diameter and calculate the coefficient of variation; (i.e., first calculate the mean and standard deviation using the diameters of all centerline path points, then calculate the diameter coefficient of variation). The specific process involves N centerline path points for the candidate landing segment, with each centerline path point having a diameter of... (i.e., obtaining the diameter of the i-th centerline path point), average diameter Diameter standard deviation The coefficient of variation of diameter is calculated using the mean diameter and the standard deviation of diameter. The coefficient of variation (i.e., the diameter variation rate) is obtained by dividing the standard deviation by the mean. The mean diameter represents the nominal diameter of the segment and is the main basis for selecting the device model (such as the stent diameter). The standard deviation of the diameter quantifies the uniformity of the vessel diameter. The smaller the standard deviation of the diameter, the more regular the cylindrical shape of the vessel, the better the device adheres to the wall, and the less blood flow disturbance. An excessively large standard deviation of the diameter indicates the presence of severe taper, dilation, or stenosis, which is not conducive to device stability.

[0090] Calculate the average curvature of the centerline path points in the candidate landing segment (which reflects the overall curvature of the vessel segment). The curvature is represented by the curvature at each centerline path point. The mean curvature has already been obtained in S22. ); i represents the i-th centerline path point; N represents the number of centerline path points in the candidate landing segment;

[0091] The center point of the candidate landing segment is extracted (i.e., the position where the total length is divided by 2 is the center point), and the branch nodes in the center line of the tree-like blood vessel are extracted (i.e., as explained in step S21, there are many initial paths (blood vessels), from which the blood vessel with the largest diameter is selected as the center line of the tree-like blood vessel. However, since the blood vessels are all connected, the remaining initial paths are connected to the center line of the tree-like blood vessel, and thus become branched paths. Therefore, each branched path has a branch node).

[0092] Calculate the Euclidean distance between each branch node and the center point of the segment, and select the minimum branch distance as the safe distance (i.e., the actual distance along the blood vessel, not the straight-line distance, because it determines the safe margin of the device end from the branch blood vessel; if it is too close to the branch node, it is easy to affect blood flow).

[0093] S2443: Divide the candidate landing segments of the safety distance into grid nodes (i.e., each grid is a node), and set the boundary rules of the candidate landing segments according to the coefficient of variation and the average curvature.

[0094] The candidate landing segment is simulated with a complete cardiac cycle according to the boundary rules, and the instantaneous wall shear stress value of each grid node is collected (i.e., instantaneous wall shear stress refers to the tangential friction force generated by the flowing blood on a unit area of ​​the blood vessel wall at a specific moment).

[0095] The instantaneous wall shear stress value is used to calculate the average wall shear stress of the grid nodes (i.e., the average wall shear stress refers to the time average of the instantaneous wall shear stress over a complete cardiac cycle, which eliminates the details of pulsation and reflects the average shear force level acting on the blood vessel wall).

[0096] The mean and standard deviation of the time-averaged wall shear stress at all grid nodes are calculated, and the non-uniformity index is further calculated (i.e., the calculation formula is expressed as obtaining all...). ( The time-averaged wall shear stress of (number of) grid nodes Where k represents the k-th grid node; calculate its mean time-averaged wall shear stress. ; Calculate its standard deviation The wall shear stress non-uniformity index is defined as: WSSUI The smaller this value, the more uniform the distribution of the time-averaged wall shear stress on the vessel wall of the candidate segment.

[0097] A preset uniformity threshold is set; it is then determined whether the non-uniformity index is less than the uniformity threshold.

[0098] If so, the candidate landing segment is determined to be the device anchoring landing area (i.e., the non-uniformity index is low, indicating that the endothelial cell growth environment is stable after device implantation, which is conducive to rapid and uniform endothelialization to cover the device and reduce the long-term risk of thrombosis, so the candidate landing segment is determined to be the device anchoring landing area).

[0099] It should be noted that the boundary rules include the inlet (i.e., the upstream endpoint), the outlet (the downstream endpoint), the vessel wall, and blood properties. The inlet is located at the upstream end face of the segment, where a Poiseuille flow profile is defined, meaning that the velocity distribution on the cross-section is parabolic, with the highest velocity at the center and zero at the vessel wall. The inlet flow rate needs to be estimated based on the vessel level (e.g., internal carotid artery, middle cerebral artery) and individualized information (e.g., cardiac output). The outlet is located at the downstream end face of the segment, where a static pressure boundary condition is defined (usually set to 0). (Pa is used as a reference); the vessel wall is assumed to be rigid with no slip boundary conditions (this is a common assumption in most macroscopic blood flow simulations); blood properties are set as blood density and dynamic viscosity, which are usually regarded as Newtonian fluids (this is a reasonable approximation for blood flow in large vessels); the candidate landing section is simulated using the above boundary rules to analyze the stress of blood flow on the vessel wall and cavity by simulating cardiac cycles (2-3 cycles); in step S2443, the center velocity or flow rate value of the Poiseuille flow profile, the specific velocity waveform of the cardiac cycle, and the specific values ​​of blood density and dynamic viscosity are used as simulation parameters such as blood flow velocity based on typical data published in the literature that are applicable to large intracranial vessels;

[0100] Example 2

[0101] like Figure 8 As shown, the present invention also provides an aneurysm segmentation landing area localization and identification system based on an optimal blood flow guidance model, comprising: a data acquisition module 10; and an identification module 20.

[0102] The acquisition module 10 is used to acquire image data of the target blood vessel region, preprocess the image data to obtain a blood vessel enhanced image, binarize the blood vessel enhanced image to obtain a blood vessel lumen binary mask, and project the blood vessel binary mask into three-dimensional space to obtain a blood vessel lumen three-dimensional binary mask.

[0103] The recognition module 20 is used to extract the tree-like vessel centerline from the enhanced vessel image using a vessel centerline extraction algorithm; analyze the centerline path points collected from the tree-like vessel centerline to obtain a multimodal fusion feature vector; predict the aneurysm probability value using a pre-built aneurysm neck detection deep learning model, and screen candidate aneurysm neck regions and aneurysm-bearing artery regions; and use the three-dimensional binary mask of the vessel lumen to mask the candidate aneurysm neck regions and aneurysm-bearing artery regions to extract the device anchoring landing area (i.e., locate the most suitable position for placing the treatment device, which is the treatment device and also the final landing area).

[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; those skilled in the art can modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some or all of the technical features; and these modifications or substitutions 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 method for aneurysm segmentation and landing area localization and identification based on an optimal blood flow guidance model, characterized in that, The following steps are included: Image data is acquired from the target vascular region, and the image data is preprocessed to obtain an enhanced vascular image; The enhanced blood vessel image is binarized to obtain a binary mask of the blood vessel cavity; the binary mask of the blood vessel cavity is projected into three-dimensional space to obtain a three-dimensional binary mask of the blood vessel cavity; the three-dimensional binary mask of the blood vessel cavity includes the blood vessel cavity, the blood vessel wall, and the complete blood vessel network; A tree-like vascular centerline is extracted from the enhanced vascular image using a vascular centerline extraction algorithm; the centerline path points of the tree-like vascular centerline are collected and analyzed to obtain a multimodal fusion feature vector; a pre-built aneurysm neck detection deep learning model is used to predict the aneurysm neck probability value from the multimodal fusion feature vector, and candidate aneurysm neck regions and aneurysm-bearing artery regions are screened; the candidate aneurysm neck regions and aneurysm-bearing artery regions are masked using the three-dimensional binary mask of the vascular lumen to extract the device anchoring and landing area.

2. The aneurysm segmentation and landing area localization and identification method based on an optimal blood flow guidance model according to claim 1, characterized in that, The tree-like vessel centerline is extracted from the enhanced vessel image using a vessel centerline extraction algorithm. The specific steps are as follows: A preset-size image block is cropped from the enhanced blood vessel image, and blood vessel inlets and multiple blood vessel outlets are obtained from the cropped image block. A blood vessel centerline extraction algorithm is used to track the blood vessels from each blood vessel inlet to the blood vessel outlet to form an initial path. The largest blood vessel diameter is selected from the initial paths, and the largest blood vessel diameters in all initial paths are sorted. The initial path with the largest blood vessel diameter is used as the center line of the tree-like blood vessel structure.

3. The aneurysm segmentation and landing area localization and identification method based on an optimal blood flow guidance model according to claim 2, characterized in that, The centerline path points of the tree-like blood vessel centerline are collected and analyzed to obtain a multimodal fusion feature vector. A pre-built deep learning model for aneurysm neck detection is used to predict the aneurysm neck probability value from the multimodal fusion feature vector, and candidate aneurysm neck regions and tumor-bearing artery regions are screened. The specific operation steps are as follows: Take several adjacent centerline path points from one of the centerline path points in the tree-like vascular centerline to form a local centerline segment; fit a local curve segment to the local centerline segment, and calculate the curvature of the centerline path point of the local curve segment; The blood vessel radius is obtained by measuring the blood vessel diameter at the centerline path point. An initial geometric feature vector is constructed by combining the curvature with the blood vessel radius; Map each centerline path point in the tree-like blood vessel centerline to three-dimensional space to determine the three-dimensional coordinates of each centerline path point; and uniformly divide a three-dimensional cubic region with the three-dimensional coordinates of each centerline path point as the center. A convolutional neural network is pre-constructed; the three-dimensional cube region is input into the convolutional neural network to obtain a depth feature vector; The initial geometric feature vector and the depth feature vector are concatenated by performing a nonlinear transformation to obtain a multimodal fusion feature vector; A deep learning model for aneurysm detection is pre-constructed; the multimodal fusion feature vectors of all centerline path points are input into the deep learning model for aneurysm detection according to the order of each centerline path point in the tree-like vascular centerline to obtain the aneurysm probability value. A preset probability threshold is set; it is then determined whether the probability value of the aneurysm neck is greater than the probability threshold. If so, the centerline path point is determined as a high-probability centerline path point, and consecutive high-probability centerline path points are combined to form a high-probability segment. A minimum length threshold is preset; the Euclidean distance between adjacent high-probability centerline path points in the high-probability segment is calculated to obtain the path length of the high-probability segment; it is determined whether the path length is greater than the minimum length threshold; if so, the high-probability segment is determined to be a candidate aneurysm neck region. If not, it is determined to be the region of the tumor-bearing artery.

4. The aneurysm segmentation and landing area localization and identification method based on an optimal blood flow guidance model according to claim 3, characterized in that, The candidate aneurysm neck region and the aneurysm-bearing artery region are masked using the three-dimensional binary mask of the vascular lumen to extract the device anchoring and landing area. The specific operation steps are as follows: The candidate tumor neck region and the tumor-bearing artery region are extracted using the three-dimensional binary mask of the blood vessel lumen, and a preliminary candidate tumor region mask is extracted; the gray-level distribution of the preliminary candidate tumor region mask and the tumor-bearing artery region mask is judged to be similar; the preliminary candidate tumor region mask and the tumor-bearing artery region mask with similar gray-level distribution are merged to obtain the intersection region; The area of ​​the intersection region is calculated by cutting the cross-section, and the narrowest cross-section of the neck is selected; the narrowest cross-section of the neck is used to identify the anchoring landing area of ​​the device.

5. The aneurysm segmentation and landing area localization and identification method based on an optimal blood flow guidance model according to claim 4, characterized in that, The candidate aneurysm neck region and the carrier artery region are extracted using the three-dimensional binary mask of the vascular lumen, and a preliminary candidate aneurysm region mask is extracted. The grayscale distribution of the preliminary candidate aneurysm region mask and the carrier artery region mask is judged to be similar. The specific operation steps are as follows: Extract each candidate aneurysm neck region from the three-dimensional binary mask of the vascular lumen to obtain a candidate aneurysm neck region mask; extract the vascular lumen voxels from the candidate aneurysm neck region mask; Use the vascular cavity voxels of all candidate tumor neck region masks as the initial seed point set. Extract each tumor-bearing artery region corresponding to the three-dimensional binary mask of the vascular lumen to obtain the tumor-bearing artery region mask; extract the vascular lumen voxels in the tumor-bearing artery region mask; calculate the average intensity of the vascular lumen voxels of adjacent tumor-bearing artery region masks; determine whether the vascular lumen voxels of the initial seed point are greater than the average intensity of the vascular lumen voxels of the tumor-bearing artery region mask; If so, the vascular lumen voxel of the mask of the tumor-bearing artery region adjacent to the vascular lumen voxel of the initial seed point is determined as the new seed point voxel. The above calculation of the average intensity of the vascular lumen voxels of the adjacent tumor-bearing artery region and the new seed point voxel are repeated until the new seed point voxel stops growing, and a preliminary candidate tumor region mask is obtained. Gaussian kernel density estimation is performed on the vascular lumen elements of the preliminary candidate tumor region mask and the vascular lumen elements of the tumor-bearing artery region mask to obtain the gray value probability density of the vascular lumen elements of the preliminary candidate tumor region mask and the gray value probability density of the vascular lumen elements of the tumor-bearing artery region mask. The Batacharia coefficient is calculated using the gray value probability densities of the vascular lumen elements of the preliminary candidate tumor region mask and the gray value probability densities of the vascular lumen elements of the tumor-bearing artery region mask. A similarity threshold is preset. It is then determined whether the Batacharia coefficient is greater than the similarity threshold. If so, it is determined that the vascular lumen element of the preliminary candidate tumor region mask and the vascular lumen element of the tumor-bearing artery region mask have similar gray-scale distribution.

6. The aneurysm segmentation and landing area localization and identification method based on an optimal blood flow guidance model according to claim 5, characterized in that, The preliminary candidate tumor region mask with similar gray-level distribution is merged with the tumor-bearing artery region mask to obtain the intersection region; the area of ​​the intersection region is calculated by cutting the cross-section, and the narrowest cross-section of the tumor neck is selected; the narrowest cross-section of the tumor neck is used to identify the landing area of ​​the identification device. The specific operation steps are as follows: The initial candidate tumor region mask with similar gray-level distribution is merged with the tumor-bearing artery region mask using the dilation operation of three-dimensional morphology to obtain the intersection region; The intersection region is cut vertically with a cross-sectional plane, and each cross-sectional plane is projected onto a two-dimensional plane to calculate the cross-sectional area; Sort all cross-sectional areas in descending order and select the narrowest cross-sectional area; If the narrowest cross-sectional area is in the candidate aneurysm neck region, then the narrowest cross-sectional area is determined to be the narrowest cross-section of the aneurysm neck; The search starting point on the central line of the dendritic vessel is selected using the geometric center point of the narrowest cross-section of the aneurysm neck. The search starting point is then moved upstream and downstream to obtain the upstream and downstream endpoints, forming candidate landing segments. The distance between the candidate landing segments is calculated based on the upstream and downstream endpoints. The safe distance between the branch nodes in the central line of the dendritic vessel and the candidate landing segments is calculated. Boundary rules are set for the candidate landing segments with safe distances to simulate a complete cardiac cycle, calculate the non-uniformity index, and determine the device's anchoring landing area.

7. The aneurysm segmentation and landing area localization and identification method based on an optimal blood flow guidance model according to claim 6, characterized in that, The search starting point on the center line of the dendritic vessels is screened using the geometric center point of the narrowest cross section of the aneurysm neck. The search starting point is then used to move upstream and downstream to obtain the upstream and downstream endpoints, forming candidate landing segments. The specific operation steps are as follows: The geometric center point of the narrowest cross section of the aneurysm neck is determined; the Euclidean distance to the geometric center point is calculated using each centerline path point on the central line of the dendritic vessels, and the shortest Euclidean distance is selected as the search starting point. A preset initial cumulative path length is set; the search starting point is used to move the centerline path point upstream of the tree-like blood vessel centerline, the cumulative path length is calculated, and the curvature value and blood vessel radius of the current centerline path point are extracted; the rate of change of blood vessel radius between the search starting point and the current centerline path point is calculated. A maximum search length is preset; it is determined whether the distance of the accumulated path length is greater than the maximum search length; if so, the current centerline path point is taken as the upstream endpoint. Preset local curvature threshold; Determine whether the curvature value of the current centerline path point is greater than the local curvature threshold; if so, then take the current centerline path point as the upstream endpoint. A preset radius change rate threshold is set; it is determined whether the radius change rate of the blood vessel is greater than the radius change rate threshold; if so, the current centerline path point is taken as the upstream endpoint. If the current centerline path point satisfies any of the above conditions, then the current centerline path point is taken as the upstream endpoint; otherwise, the current centerline path point is taken as the new search starting point, and the above steps are repeated to move upstream to the centerline path point until the conditions are met and the movement stops. Using the search starting point, move the centerline path point of the tree-like blood vessel centerline downstream, repeat the same steps above, and filter out the downstream endpoints; connect the upstream endpoints and the downstream endpoints to form candidate landing segments.

8. The aneurysm segmentation and landing area localization and identification method based on an optimal blood flow guidance model according to claim 7, characterized in that, The distance to the candidate landing segment is calculated based on the upstream and downstream endpoints. The safe distance of the branch node is calculated using the distance between the branch node in the centerline of the tree-like vessel and the candidate landing segment. The specific operation steps are as follows: The candidate landing segment is projected into three-dimensional space, and the Euclidean distance between the upstream endpoint and the downstream endpoint is calculated as the distance of the candidate landing segment; Convert the vessel radius of each centerline path point in the candidate landing segment to a diameter and calculate the coefficient of variation; calculate the average curvature of the centerline path points in the candidate landing segment; The center point of the candidate landing segment is extracted, and the branch nodes in the center line of the tree-like blood vessel are extracted; Calculate the Euclidean distance between each branch node and the center point of the segment, and select the minimum branch distance as the safe distance.

9. The aneurysm segmentation and landing area localization and identification method based on an optimal blood flow guidance model according to claim 8, characterized in that, Boundary rules are set for candidate landing zones at safe distances to simulate a complete cardiac cycle, calculate non-uniformity indices, and determine the device's anchoring landing zone. The specific operation steps are as follows: The candidate landing segments of the safe distance are divided into grid nodes, and the boundary rules of the candidate landing segments are set according to the coefficient of variation and the average curvature. The candidate landing segments are simulated to complete the cardiac cycle according to the boundary rules, and the instantaneous wall shear stress value of each grid node is collected. The instantaneous wall shear stress value is used to calculate the average wall shear stress of the grid nodes; the mean and standard deviation of the average wall shear stress of all grid nodes are calculated, and the non-uniformity index is further calculated. A preset uniformity threshold is set; it is then determined whether the non-uniformity index is less than the uniformity threshold. If so, the candidate landing section is determined to be the device anchoring landing area.

10. A system for locating and identifying the landing zone of an aneurysm based on an optimal blood flow guidance model, characterized in that, include: Acquisition module; Recognition module; The acquisition module is used to acquire image data of the target blood vessel region, preprocess the image data to obtain an enhanced blood vessel image, binarize the enhanced blood vessel image to obtain a binary mask of the blood vessel lumen, and project the binary mask of the blood vessel into three-dimensional space to obtain a three-dimensional binary mask of the blood vessel lumen. The identification module is used to extract the tree-like vessel centerline from the enhanced vessel image using a vessel centerline extraction algorithm; analyze the centerline path points collected from the tree-like vessel centerline to obtain a multimodal fusion feature vector; use a pre-built aneurysm neck detection deep learning model to predict the aneurysm neck probability value from the multimodal fusion feature vector, and screen candidate aneurysm neck regions and aneurysm-bearing artery regions; use the three-dimensional binary mask of the vessel lumen to mask the candidate aneurysm neck regions and aneurysm-bearing artery regions, and extract the device anchoring landing area.

Images