Image processing-based cerebral vessel feature extraction method and system

Abstract

The application relates to a cerebral vascular feature extraction method and system based on image processing. The method comprises the following steps: filtering the image of the cerebral vascular to obtain an initial vascular probability, and mapping the image to a preset brain template space; updating the initial vascular probability according to the features of the anatomical region to which the voxel belongs to obtain a target vascular probability; extracting a plurality of skeleton lines, and obtaining the morphological regularity based on the boundary smoothness of the skeleton lines and the diameter difference between a plurality of skeleton points; calculating the occlusion rate based on the distance between each end point of the skeleton line and a plurality of voxels in the neighborhood, the target vascular probability and the vascular probability information; analyzing the difference in the boundary smoothness of the skeleton line, the position change of the suspected voxel in a plurality of continuous images and the enhancement coefficient to obtain the target enhancement coefficient of each suspected voxel; performing enhancement processing on the image according to the target enhancement coefficient, and extracting the cerebral vascular features based on the enhanced image. The application can improve the accuracy of cerebral vascular feature extraction.

Description

Technical Field

[0001] This application relates to the field of medical image processing technology, and in particular to a method and system for extracting cerebral vascular features based on image processing. Background Technology

[0002] Significant progress has been made in the automated reconstruction of large and medium-sized intracranial arteries using deep learning-based vascular segmentation and skeleton extraction techniques. This has provided an efficient tool for the quantitative assessment of diseases such as stroke and aneurysms, and has promoted the automation and standardization of cerebrovascular imaging analysis. However, accurate characterization of microvascular distal vessels (such as cortical branches and perforating arteries) still faces serious challenges. On the one hand, non-invasive imaging modalities such as MRA are affected by artifacts such as thermal noise, slow blood flow saturation effects, and cerebrospinal fluid flow, resulting in weak or even absent signals from distal branches. On the other hand, existing methods often employ a globally uniform segmentation strategy, making it difficult to account for the significant differences in anatomical structure, vascular density, and pathological susceptibility among different blood supply regions such as the temporal and frontal lobes. This leads to systematic breaks or misconnections in the microvascular network in key functional areas. Especially in temporal dynamic analysis, this severely limits the clinical reliability in early screening of small vessel disease and moyamoya disease, as well as in the assessment of collateral circulation. Summary of the Invention

[0003] To address the aforementioned technical problems, the purpose of this application is to provide a method and system for extracting cerebral blood vessel features based on image processing. The specific technical solution adopted is as follows: In a first aspect, a method for extracting cerebral vascular features based on image processing is provided, the method comprising: The acquired images of cerebral blood vessels are filtered to obtain the initial blood vessel probability of each voxel. The images are then mapped to a preset brain template space. The initial blood vessel probability is updated according to the characteristics of the anatomical region to which the voxel belongs, to obtain the target blood vessel probability. Multiple skeleton lines are extracted from the suspected blood vessel region in the target image. The morphological regularity of each skeleton line is obtained based on the smoothness of the skeleton line boundary and the difference in tube diameter between multiple skeleton points of the skeleton line. The target image is obtained by feature enhancement based on the probability of the target blood vessel. The suspected blood vessel region is obtained by threshold segmentation of the target image based on the gray value. Each skeleton line corresponds to a suspected blood vessel sub-region. The suspected blood vessel sub-region is obtained by dividing the suspected blood vessel region into connected components. Based on the distance between each endpoint of the skeleton line and multiple voxels in its neighborhood, the target blood vessel probability of the multiple voxels, and the blood vessel probability information corresponding to the multiple voxels, the occlusion rate of the skeleton line is calculated; the occlusion rate indicates the proportion of the occluded area in the skeleton line. The differences in boundary smoothness of each skeleton line in multiple consecutive images, the positional changes of suspected voxels in the suspected blood vessel sub-region corresponding to the skeleton line in multiple consecutive images, and the enhancement coefficient of the skeleton line are analyzed to obtain the target enhancement coefficient of each suspected voxel in the suspected blood vessel sub-region corresponding to the skeleton line; the enhancement coefficient of the skeleton line is calculated based on the morphological regularity and occlusion rate of the skeleton line. Based on the target enhancement coefficients of multiple suspected voxels in the image, the image is enhanced, and cerebral vascular features are extracted based on the enhanced image.

[0004] Optionally, the acquired images of brain blood vessels are filtered to obtain the initial blood vessel probability for each voxel, and the images are mapped to a preset brain template space. The initial blood vessel probabilities are then updated based on the characteristics of the anatomical region to which the voxel belongs, to obtain the target blood vessel probability, including: A multi-scale Hessian matrix filter was used to filter images of cerebral blood vessels to obtain the tubular response value of each voxel at each scale. Based on the largest tubular response value among all voxels, the largest tubular response value of each voxel at multiple scales is normalized to obtain the initial blood vessel probability of that voxel. The image is mapped to a preset brain template space, the anatomical region category of each voxel is determined based on the preset brain region atlas, and the vascular probability information of the corresponding voxel is obtained based on the preset cerebral vascular probability atlas. The initial blood vessel probability is updated based on the anatomical region category and blood vessel probability information of each voxel to obtain the target blood vessel probability.

[0005] Optionally, multiple skeleton lines are extracted from the suspected blood vessel region in the target image. Based on the smoothness of the skeleton line boundaries and the diameter difference between multiple skeleton points of the skeleton line, the morphological regularity of each skeleton line is obtained, including: The target image is obtained by performing feature enhancement on the image based on the probability of the target blood vessel. For each voxel, calculate the difference between the mean and standard deviation of the gray values ​​in the neighborhood of that voxel to obtain the segmentation threshold for that voxel. Based on the segmentation threshold of each voxel, threshold segmentation is performed on the target image to obtain the suspected blood vessel region; The suspected blood vessel region is divided into connected components to obtain multiple suspected blood vessel sub-regions. A morphological thinning algorithm is applied to each suspected blood vessel sub-region to iteratively erode its boundary until a center line with a single pixel width is obtained, which serves as the skeleton line corresponding to the suspected blood vessel sub-region. Multiple skeleton lines are extracted from the target image. Based on the smoothness of the skeleton line boundary and the difference in vessel diameter between multiple skeleton points of the skeleton line, the morphological regularity of each skeleton line is obtained. Morphological regularity characterizes the regularity of the blood vessel morphology.

[0006] Optionally, multiple skeleton lines are extracted from the target image. Based on the smoothness of the skeleton lines' boundaries and the pipe diameter differences between multiple skeleton points, the morphological regularity of each skeleton line is obtained, including: Extract multiple skeleton lines from the target image, calculate the Euclidean distance between the two sides of the blood vessel walls of each skeleton line, and obtain the diameter of the skeleton line. Obtain the left and right pipe wall point sequences for each skeleton line, and calculate the first curvature sequence of multiple adjacent points in the left pipe wall point sequence and the second curvature sequence of multiple adjacent points in the right pipe wall point sequence respectively. The boundary smoothness of the skeleton line is calculated based on the average of the curvature in the first curvature sequence and the curvature in the second curvature sequence. Calculate the pipe diameter difference sequence between adjacent skeleton points along the direction of each skeleton line, and calculate the attenuation coefficient based on the consistency of the difference direction in the pipe diameter difference sequence; The morphological regularity of each skeleton line is determined based on its boundary smoothness and attenuation coefficient.

[0007] Optionally, the morphological regularity of the skeleton line is determined based on the boundary smoothness and attenuation coefficient of each skeleton line, including: The boundary smoothness and attenuation coefficient of each skeleton line are weighted and fused to obtain the morphological regularity of the skeleton line.

[0008] Optionally, based on the distance between each endpoint of the skeleton line and multiple voxels in its neighborhood, the target blood vessel probability of the multiple voxels, and the preset probability corresponding to the multiple voxels, the occlusion rate of the skeleton line is calculated, including: For each endpoint of the skeleton line, the neighborhood of that endpoint is obtained, and the target blood vessel probability of each voxel in the neighborhood, the reciprocal of the Euclidean distance between the voxel and the endpoint, and the blood vessel probability information corresponding to the voxel are weighted and fused to obtain the continuity coefficient of the voxel. Calculate the mean and standard deviation of the continuity coefficients of multiple voxels in the neighborhood of each endpoint, and determine the difference between the mean and standard deviation of the continuity coefficients as the occlusion threshold. If the continuity coefficient of any voxel in the neighborhood of an endpoint is less than the occlusion threshold, then the endpoint is determined as an occlusion point. For any skeleton line, calculate the ratio of the number of multiple occlusion points to the number of multiple endpoints in the skeleton line to obtain the occlusion rate of the skeleton line.

[0009] Optionally, the differences in boundary smoothness of each skeleton line in multiple consecutive images, the positional changes of suspected voxels in the suspected blood vessel sub-region corresponding to the skeleton line in multiple consecutive images, and the enhancement coefficient of the skeleton line are analyzed to obtain the target enhancement coefficient of each suspected voxel in the suspected blood vessel sub-region corresponding to the skeleton line, including: The enhancement coefficient of each skeleton line is obtained by weighted fusion of its morphological regularity and occlusion rate. Calculate the boundary smoothness sequence and attenuation coefficient sequence for each skeleton line in multiple consecutive images; The first variation coefficient of the boundary smoothness sequence is obtained by calculating the difference between adjacent frames of the boundary smoothness sequence and the average difference of multiple adjacent frames. The second variation coefficient of the attenuation coefficient sequence is obtained by calculating the difference between adjacent frames of the attenuation coefficient sequence and the average difference of multiple adjacent frames. The morphological coefficient of the skeleton line is obtained by multiplying the first variation coefficient and the second variation coefficient. The morphological coefficients of each skeleton line, the positional changes of suspected voxels in the suspected blood vessel sub-region corresponding to the skeleton line in multiple consecutive images, and the enhancement coefficient of the skeleton line are analyzed to obtain the target enhancement coefficient of each suspected voxel in the suspected blood vessel sub-region corresponding to the skeleton line.

[0010] Optionally, the morphological coefficients of each skeleton line, the positional changes of suspected voxels in the suspected blood vessel sub-region corresponding to the skeleton line in multiple consecutive images, and the enhancement coefficient of the skeleton line are analyzed to obtain the target enhancement coefficient of each suspected voxel in the suspected blood vessel sub-region corresponding to the skeleton line, including: Using optical flow, the position coordinates of each suspected voxel in the suspected blood vessel sub-region corresponding to each skeleton line are sorted in chronological order in multiple consecutive images to obtain the position sequence of the suspected voxel. Calculate the difference in the position sequence of each suspected voxel between adjacent frames to obtain the movement sequence of the suspected voxel; Obtain the neighborhood of each suspected voxel and calculate the movement sequence of each voxel in the neighborhood; Calculate the cosine similarity between the movement sequence of each suspected voxel and the movement sequence of each suspected voxel in its neighborhood in each adjacent frame, and calculate the mean of the cosine similarity over multiple frames to obtain the similarity between the suspected voxel and the suspected voxel in its neighborhood. The consistency coefficient of a suspected voxel is determined based on the average similarity between the suspected voxel and multiple suspected voxels in its neighborhood. The target enhancement coefficient of a suspected voxel is obtained by taking the mean of the morphological coefficient of each suspected voxel and the enhancement coefficient of its corresponding skeleton line, and the consistency coefficient of the suspected voxel.

[0011] Optionally, the image is enhanced based on the target enhancement coefficients of multiple suspected voxels in the image, and cerebral vascular features are extracted based on the enhanced image, including: Using the target enhancement coefficient as the sharpening intensity control parameter for each suspected voxel, Laplacian sharpening is performed frame by frame on the image sequence to generate the enhanced image sequence. From the enhanced image sequence, cerebral vascular features are determined based on the morphological topology parameters and hemodynamic parameters of the blood vessels.

[0012] Secondly, a brain vascular feature extraction system based on image processing is provided, the system comprising: The filtering module is used to filter the acquired images of cerebral blood vessels, obtain the initial blood vessel probability of each voxel, map the image to a preset brain template space, and update the initial blood vessel probability according to the characteristics of the anatomical region to which the voxel belongs, so as to obtain the target blood vessel probability. The extraction module is used to extract multiple skeleton lines of suspected blood vessel regions in the target image. Based on the smoothness of the skeleton line boundary and the difference in tube diameter between multiple skeleton points of the skeleton line, the morphological regularity of each skeleton line is obtained. The target image is obtained by performing feature enhancement on the image according to the probability of the target blood vessel. The suspected blood vessel region is obtained by thresholding the target image according to the gray value. Each skeleton line corresponds to a suspected blood vessel sub-region. The suspected blood vessel sub-region is obtained by dividing the suspected blood vessel region into connected components. The calculation module is used to calculate the occlusion rate of the skeleton line based on the distance between each endpoint of the skeleton line and multiple voxels in its neighborhood, the target blood vessel probability of the multiple voxels, and the blood vessel probability information corresponding to the multiple voxels; the occlusion rate indicates the proportion of the occluded area in the skeleton line. The analysis module is used to analyze the differences in the boundary smoothness of each skeleton line in multiple consecutive images, the positional changes of the suspected voxels in the suspected blood vessel sub-region corresponding to the skeleton line in multiple consecutive images, and the enhancement coefficient of the skeleton line, so as to obtain the target enhancement coefficient of each suspected voxel in the suspected blood vessel sub-region corresponding to the skeleton line; the enhancement coefficient of the skeleton line is calculated based on the morphological regularity and occlusion rate of the skeleton line. The enhancement module is used to enhance the image based on the target enhancement coefficients of multiple suspected voxels in the image, and extract cerebral vascular features based on the enhanced image.

[0013] Based on common knowledge in the field, the above-mentioned preferred conditions can be combined arbitrarily to obtain various preferred embodiments of this application.

[0014] This application offers the following advantages: By updating the initial vessel probability based on anatomical regions, the imaging quality of micro-terminal vessels is effectively enhanced. Precise quantitative localization of vessel morphological defects and signal interruption regions is achieved by calculating the morphological regularity and occlusion rate of skeleton lines. Multiple skeleton lines are extracted from suspected vessel regions, and the smoothness of their boundaries and the diameter differences between skeleton points are analyzed to calculate the morphological regularity of each skeleton line. Suspected vessel sub-regions are determined by combining grayscale thresholding and connected component partitioning techniques. Based on the distance between multiple voxels in the neighborhood of the skeleton line endpoints and their target vessel probability information, the skeleton line occlusion rate is calculated, indicating the proportion of occluded areas. Furthermore, through comprehensive analysis of changes in the smoothness of skeleton line boundaries, changes in the position of suspected voxels, and enhancement coefficients in consecutive frames, the target enhancement coefficient for each suspected voxel is obtained. Finally, the images were enhanced based on the target enhancement coefficients of multiple suspected voxels, and cerebral vascular features were accurately extracted from the enhanced images. This improved the accuracy and efficiency of cerebral vascular feature extraction, providing a high-quality imaging basis for subsequent high-precision cerebral vascular feature extraction. This provides a more reliable imaging basis for the early diagnosis and assessment of diseases such as stroke and aneurysm. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of a brain blood vessel feature extraction method based on image processing in one embodiment; Figure 2 This is a raw cerebral vascular MRA image from an image processing-based cerebral vascular feature extraction method in one embodiment; Figure 3 This is a grayscale image obtained from a brain blood vessel feature extraction method based on image processing in one embodiment; Figure 4 This is a multi-scale tubular response result diagram of a brain blood vessel feature extraction method based on image processing in one embodiment; Figure 5 This is a multi-scale enhancement result image of a brain blood vessel feature extraction method based on image processing in one embodiment; Figure 6 An adaptive threshold distribution map of a brain blood vessel feature extraction method based on image processing in one embodiment; Figure 7This is a map of a suspected vascular region using an image processing-based brain vascular feature extraction method in one embodiment. Figure 8 This is a gradient magnitude response map of a brain blood vessel feature extraction method based on image processing in one embodiment; Figure 9 This is an edge detection result image of a brain blood vessel feature extraction method based on image processing in one embodiment; Figure 10 This is a skeleton line diagram of a brain blood vessel feature extraction method based on image processing in one embodiment; Figure 11 This is a schematic diagram of the structure of a brain blood vessel feature extraction system based on image processing in one embodiment. Detailed Implementation

[0017] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0018] The specific scheme of the image processing-based cerebral vascular feature extraction method provided in this application is described below with reference to the accompanying drawings. For example... Figure 1 As shown, the method includes: S11. Filter the acquired images of cerebral blood vessels to obtain the initial blood vessel probability of each voxel, and map the images to a preset brain template space. Update the initial blood vessel probability according to the characteristics of the anatomical region to which the voxel belongs to obtain the target blood vessel probability.

[0019] This application's embodiments are based on cerebral vascular imaging images, constructing a skeletal outline from the main vessels to the terminal vessels and their hemodynamic performance over time. This eliminates artifacts and interference caused by operational and external errors during the imaging process, and compares the results with standard vascular distribution maps for reference, thus obtaining accurate identification results. The input images described in this application are not limited to static images from a single scan, but are more preferably dynamic cerebral vascular imaging sequences containing a time dimension (such as dynamic contrast-enhanced MRA (Magnetic Resonance Angiography) or 3D TOF sequences with multiple repeated scans (Three-Dimensional Time-of-Flight)). This time-series sequence consists of T-frame three-dimensional volumetric data. The structure, with each frame having a spatial resolution that is isotropic or approximately isotropic, provides a basis for the temporal stability of optical flow rivers.

[0020] In one embodiment, the acquired images of cerebral blood vessels are filtered to obtain the initial blood vessel probability for each voxel, and the images are mapped to a preset brain template space. The initial blood vessel probabilities are then updated based on the characteristics of the anatomical region to which the voxel belongs to obtain the target blood vessel probability, including: A multi-scale Hessian matrix filter was used to filter images of cerebral blood vessels to obtain the tubular response value of each voxel at each scale. Based on the largest tubular response value among all voxels, the largest tubular response value of each voxel at multiple scales is normalized to obtain the initial blood vessel probability of that voxel. The image is mapped to a preset brain template space, the anatomical region category of each voxel is determined based on the preset brain region atlas, and the vascular probability information of the corresponding voxel is obtained based on the preset cerebral vascular probability atlas. The initial blood vessel probability is updated based on the anatomical region category and blood vessel probability information of each voxel to obtain the target blood vessel probability.

[0021] First, obtain 3D TOF MRA images of cerebral blood vessels, such as... Figure 2 As shown, the main blood vessels are clear in the original image, but the signals of the distal branches are weak. Grayscale processing of the image yields the following results: Figure 3 As shown, the grayscale values ​​are uniformly distributed. Applying a multi-scale Hessian matrix filter to the image, due to its eigenvalues, can effectively distinguish tubular, patchy, and layered structures at different scales ( Calculate the tubular response value for each voxel (a data point within a 3D image) using the following method. To match blood vessel structures with diameters of 1-4 mm. For The setting is mainly used to enhance major blood vessels such as the basilar artery and internal carotid artery, and avoid the loss of central signal due to flow void effect.

[0022] The multi-scale maximum response value is normalized to [0,1] to obtain the initial vessel probability for each voxel vessel. The calculation formula is: , The maximum tubular response value for each voxel at multiple scales. The largest tubular response value among all voxels. Global maximum response value. That is, the maximum value of the tubular response of all voxels at all scales. The higher the value, the greater the likelihood that the voxel belongs to a blood vessel. Figure 4 The results are for multi-scale tubular response. Figure 5 For enhanced vascular images, Figure 6 As shown in the adaptive threshold distribution map, the vascular region and the background region are clearly distinguishable in terms of threshold distribution.

[0023] The preprocessed images are registered to a standard brain template space (such as MNI152, a well-known acquisition method) using affine transformation, mapping the original image voxel coordinates to spatial coordinates. Within this unified anatomical space, pre-constructed standard brain region atlases and standard cerebral vascular probability atlases can be directly accessed. This process obtains the vascular probability information for the corresponding voxels, ensuring spatial consistency for all subsequent feature fusions based on the same anatomical location. The standard cerebral vascular probability atlas includes a probability vector field representing the main vascular orientations. Gradient constraints for microvascular regions are based on the principal direction of the local region, rather than the absolute direction of the standard atlas. Different regions, such as the frontal cortex and basal ganglia white matter, are also considered. By making restrictive adjustments, a more refined probability of the target blood vessel can be obtained. , specifically: In the cortex, for Medium probability voxels, according to Increase confidence while preserving the fine vascular network. For deep perforator regions, i.e., voxels... In areas where blood vessels have relatively fixed course but small diameters, local directional constraints are introduced. If the gradient direction of a voxel makes an angle with the gradient direction of a voxel at the same location in the standard atlas... ,but ,otherwise This suppresses directional artifacts. For the main trunk region of large blood vessels, i.e., voxels... The area, for A 3×3×3 local smoothing process is performed (using a smoothing algorithm, a well-known technique) to enhance continuity. (The above parameters are set as preferred examples in this embodiment; actual values ​​should be determined based on the imaging device's field strength or the true signal-to-noise ratio).

[0024] Thus, the target vessel probability of each voxel in the 3D TOF MRA imaging of cerebral blood vessels in each frame is obtained. Feature enhancement is performed on the image based on the target blood vessel probability to obtain the target image, where the target blood vessel probabilities of all voxels are... After min-max normalization processing.

[0025] S12. Extract multiple skeleton lines from the suspected blood vessel region in the target image. Based on the smoothness of the skeleton line boundary and the difference in tube diameter between multiple skeleton points of the skeleton line, obtain the morphological regularity of each skeleton line.

[0026] The target image is obtained by feature enhancement based on the probability of the target blood vessel. The suspected blood vessel region is obtained by threshold segmentation of the target image based on the gray value. Each skeleton line corresponds to a suspected blood vessel sub-region, which is obtained by dividing the suspected blood vessel region into connected components.

[0027] The brain's blood supply system mainly consists of two parts: the anterior circulation of the internal carotid artery system and the posterior circulation of the vertebrobasilar artery system. While the main blood vessels supplying different systems vary in different regions, most of the main large vessels and the overall vascular network exhibit a generally symmetrical distribution, with a few individual differences, primarily in the distal microvessels. Furthermore, in cerebral vascular imaging, some microvessels and even secondary blood supply branches are obscured due to occlusion by gyri, skull, and other tissues, as well as differences in blood supply areas. Additionally, artifacts caused by thermal noise from the equipment and respiration also exist, significantly hindering feature analysis. Therefore, this application first constructs a framework line based on clearly defined main blood vessels, tracks vascular course and the probability of distal vessel presence, and analyzes their temporal consistency. Areas with greater differences require more attention, necessitating higher requirements for noise reduction and enhancement in certain regions.

[0028] Feature enhancement is performed on the image based on the probability of the target blood vessel to obtain the target image. Adaptive threshold segmentation is used. For each voxel, a 5×5×5 neighborhood is constructed with its center, and the average gray value of the neighborhood is calculated. with standard deviation Set the segmentation threshold T, where, Voxels exceeding a certain size are marked as suspected blood vessel pixels. A suspected blood vessel region is constructed using all these pixels, and the Zhang-Suen morphological thinning algorithm (a known technique) is applied to iteratively erode the region to a single pixel width, resulting in multiple initial skeleton lines. Segments with a continuous length of 3 pixels or less are filtered out to remove false skeletons caused by noise. Within the suspected blood vessel region, adjacent suspected blood vessel pixels form a suspected blood vessel sub-region, with each sub-region corresponding to one skeleton line. Figure 7 This is a segmented area suspected to be a blood vessel. Figure 8 The gradient magnitude response reflects the gradient intensity distribution at the vessel boundary. Figure 9 This is the edge detection result.

[0029] Take the lower threshold To ensure that all real vascular voxels, especially those from small vessels with weak signals, are included in the suspected set, some non-vascular voxels will also be introduced. Therefore, subsequent analysis of their spatial and temporal performance is needed to gradually eliminate noise artifacts and obtain the true cerebral vascular features.

[0030] In one embodiment, multiple skeleton lines are extracted from the target image. Based on the boundary smoothness of the skeleton lines and the diameter difference between multiple skeleton points of the skeleton lines, the morphological regularity of each skeleton line is obtained, including: Extract multiple skeleton lines from the target image, calculate the Euclidean distance between the two sides of the blood vessel walls of each skeleton line, and obtain the diameter of the skeleton line. Obtain the left and right pipe wall point sequences for each skeleton line, and calculate the first curvature sequence of multiple adjacent points in the left pipe wall point sequence and the second curvature sequence of multiple adjacent points in the right pipe wall point sequence respectively. The boundary smoothness of the skeleton line is calculated based on the average of the curvature in the first curvature sequence and the curvature in the second curvature sequence. Calculate the pipe diameter difference sequence between adjacent skeleton points along the direction of each skeleton line, and calculate the attenuation coefficient based on the consistency of the difference direction in the pipe diameter difference sequence; The morphological regularity of each skeleton line is determined based on its boundary smoothness and attenuation coefficient.

[0031] Specifically, the morphological regularity of each skeleton line is determined based on its boundary smoothness and attenuation coefficient, including: The boundary smoothness and attenuation coefficient of each skeleton line are weighted and fused to obtain the morphological regularity of the skeleton line.

[0032] Figure 10 The extracted skeletal lines (red lines) clearly reflect the direction of the blood vessel's centerline, with skeletal points as the reference. Centered on the skeleton point, a one-dimensional grayscale profile is sampled along the normal plane of each skeleton point (perpendicular to the skeleton tangent direction, i.e., the direction of the blood vessel cross-section). Specifically, the gradient threshold is set to 100, starting from the skeleton point... Starting from the normal direction, traverse point by point to both sides, and obtain the first pixel with a grayscale gradient value greater than 100 as the boundary point between the blood vessel wall and the background, thus obtaining candidate points on both sides of the blood vessel wall. At this point, the local pipe diameter can be calculated. The calculation formula is: . for Euclidean distance.

[0033] The obtained pipe wall points are respectively denoted as the left pipe wall point sequence. and right-side tube wall point sequence Following the direction of the skeleton lines, the left side of the pipe wall... Point on the right side of the pipe wall Correspondingly; for the left-side pipe wall point sequence, calculate the curvature of three adjacent points, and take... Calculate the curvature corresponding to the i-th skeleton point from these three skeleton points, where i is the i-th point. Use least squares to fit the arc to obtain the normalized curvature of the i-th skeleton point. The smaller the curvature, the smoother the boundary. Calculate the first curvature sequence of multiple adjacent points in the left pipe wall point sequence, and calculate the mean of multiple curvatures in the first curvature sequence. At this point, the overall smoothness of the left pipe wall can be obtained. Similarly, the normalized curvature of the i-th point on the right-side pipe wall is obtained. And the overall smoothness of the right-side pipe wall.

[0034] Smoothness of skeleton lines boundary The calculation formula is: ;in, The overall smoothness of the left pipe wall. The overall smoothness of the right-side pipe wall. The larger the value, the smoother the overall boundary; the smaller the value, the more likely it is to be an artifact.

[0035] If each skeleton line has only two endpoints, starting from one endpoint and moving to the other, calculate the local pipe diameter of the previous skeleton point sequentially. Subtract the local pipe diameter of the last skeleton point The difference in pipe diameter between two adjacent skeleton points is obtained. The calculation formula is: The diameter difference sequence of the skeleton points is formed by the diameter difference of all adjacent skeleton points.

[0036] In the pipe diameter difference sequence, obtain the number of all positive numbers G1, the number of negative numbers G2, the number of zeros G3, and the attenuation coefficient. The calculation formula is: ,in It is the maximum value among G1 and G2.

[0037] In healthy cerebral blood vessels, from the aorta to secondary branches in different blood supply areas and distal microvessels, the diameter of the vessels gradually decreases from the main trunk to each level of branch due to differences in the function of the corresponding brain region and the blood supply area itself. That is, from one endpoint to another along the skeletal line, the diameter difference between adjacent skeletal points is either all positive or all negative (zero), hence the attenuation coefficient... The larger the size, the more it corresponds to the characteristics of healthy blood vessels.

[0038] If each skeleton line has more than two endpoints, perform pairwise combinations of all endpoints without repetition. Similarly, follow the steps above to obtain the attenuation coefficient for each combination. The average of the attenuation coefficients of all combinations is recorded as the attenuation coefficient of each skeleton line. .

[0039] The regularity of the shape of each skeleton line The calculation formula is: ,in, The preferred value is 0.5. ,and The actual value is determined based on the imaging resolution. Normalize to [0,1]. The smaller the value, the rougher the blood vessel edge and the more blurred the diameter attenuation, the more necessary subsequent enhancement processing is required.

[0040] S13. Based on the distance between each endpoint of the skeleton line and multiple voxels in its neighborhood, the target blood vessel probability of the multiple voxels, and the blood vessel probability information corresponding to the multiple voxels, calculate the occlusion rate of the skeleton line.

[0041] The occlusion rate indicates the percentage of the skeleton line that is occluded.

[0042] In one embodiment, the occlusion rate of the skeleton line is calculated based on the distance between each endpoint of the skeleton line and multiple voxels in its neighborhood, the target blood vessel probability of the multiple voxels, and the preset probability corresponding to the multiple voxels, including: For each endpoint of the skeleton line, the neighborhood of that endpoint is obtained, and the target blood vessel probability of each voxel in the neighborhood, the reciprocal of the Euclidean distance between the voxel and the endpoint, and the blood vessel probability information corresponding to the voxel are weighted and fused to obtain the continuity coefficient of the voxel. Calculate the mean and standard deviation of the continuity coefficients of multiple voxels in the neighborhood of each endpoint, and determine the difference between the mean and standard deviation of the continuity coefficients as the occlusion threshold. If the continuity coefficient of any voxel in the neighborhood of an endpoint is less than the occlusion threshold, then the endpoint is determined as an occlusion point. For any skeleton line, calculate the ratio of the number of multiple occlusion points to the number of multiple endpoints in the skeleton line to obtain the occlusion rate of the skeleton line.

[0043] It is necessary to analyze whether the endpoints of each skeletal line are breaks in blood vessels. In the brain region, the presence of brain tissue and skull affects the signal intensity and image quality to some extent during imaging. Furthermore, some blood vessels are embedded deep within the tissue and may be obscured by superior blood vessels or tissue during imaging, potentially causing breaks in the skeletal lines. This means that some areas are obscured or show weak signals, resulting in discontinuities in the image of continuous blood vessels. For these breaks, it is necessary to accurately identify these obstruction points to avoid missing micro-vessels. For any endpoint of each skeletal line... Get endpoint Place spherical neighborhood with radius At the endpoint The search is performed within the spherical neighborhood to obtain each voxel within the spherical neighborhood. The continuous possibilities, each voxel continuity coefficient The calculation formula is: ; in, voxels Target blood vessel probability. voxels and endpoints The smaller the Euclidean distance (the closer to the endpoint), the greater the contribution of this term, and the higher the likelihood of it representing a connection. voxels The corresponding blood vessel probability information.

[0044] Preferred values: blood flow signal intensity coefficient is 0.4, consistency coefficient with inflection point direction is 0.3, and matching coefficient with standard atlas is 0.3.

[0045] Calculate endpoints spherical neighborhood Mean of continuous coefficients within With the standard deviation of the continuity coefficient Occlusion threshold The calculation formula is: If a certain endpoint continuity coefficient If the value is positive, it indicates that the point is a suspected vascular morphology; otherwise, the point is an artifact or an obstruction.

[0046] For any skeleton line, count the total number of endpoints. and the number of multiple occlusion points The occlusion rate is obtained by calculating the proportion of occluded points. , ,like A value of 0 indicates that the skeleton line is unobstructed and the signal continuity is good; The larger the value, the smaller the occlusion area on the current skeleton line; The smaller the value, the more occluded parts there are on the current skeleton line.

[0047] S14. Analyze the differences in the boundary smoothness of each skeleton line in multiple consecutive images, the position changes of the suspected voxels in the suspected blood vessel sub-region corresponding to the skeleton line in multiple consecutive images, and the enhancement coefficient of the skeleton line to obtain the target enhancement coefficient of each suspected voxel in the suspected blood vessel sub-region corresponding to the skeleton line.

[0048] The enhancement coefficient of the skeleton line is calculated based on the regularity of its shape and the occlusion rate.

[0049] In one embodiment, the differences in boundary smoothness of each skeleton line in multiple consecutive images, the positional changes of suspected voxels in the suspected blood vessel sub-region corresponding to the skeleton line in multiple consecutive images, and the enhancement coefficient of the skeleton line are analyzed to obtain the target enhancement coefficient of each suspected voxel in the suspected blood vessel sub-region corresponding to the skeleton line, including: The enhancement coefficient of each skeleton line is obtained by weighted fusion of its morphological regularity and occlusion rate. Calculate the boundary smoothness sequence and attenuation coefficient sequence for each skeleton line in multiple consecutive images; The first variation coefficient of the boundary smoothness sequence is obtained by calculating the difference between adjacent frames of the boundary smoothness sequence and the average difference of multiple adjacent frames. The second variation coefficient of the attenuation coefficient sequence is obtained by calculating the difference between adjacent frames of the attenuation coefficient sequence and the average difference of multiple adjacent frames. The morphological coefficient of the skeleton line is obtained by multiplying the first variation coefficient and the second variation coefficient. The morphological coefficients of each skeleton line, the positional changes of suspected voxels in the suspected blood vessel sub-region corresponding to the skeleton line in multiple consecutive images, and the enhancement coefficient of the skeleton line are analyzed to obtain the target enhancement coefficient of each suspected voxel in the suspected blood vessel sub-region corresponding to the skeleton line.

[0050] For each skeleton line: based on the morphological regularity of the skeleton line and occlusion rate This yields an enhancement coefficient for enhancement grading. The calculation formula is: In one embodiment, The preferred value is 0.6, indicating that the morphological characteristics directly reflect the smoothness and natural attenuation of the blood vessels themselves, requiring a higher weight to determine whether enhancement is needed. The higher the value, the more likely the area is to be a "high-quality vessel" or a "vessel that is obscured but has a believable morphology," thus warranting strong enhancement to ensure... The physical meaning of this is positively correlated with the sharpening intensity.

[0051] After obtaining the above set of features, it is necessary to further analyze the blood flow in each blood vessel. In cerebral MRA images, the contrast agent injected into the blood vessels will flow with the blood in the cerebral blood vessels, gradually diffusing and filling from the proximal end of the blood vessel to the distal end. Based on this, optical flow method is used to analyze the blood flow in cerebral blood vessels, and at the same time, combined with the temporal performance of geometric features, suspected blood vessels are enhanced.

[0052] For each suspected voxel within the suspected vascular subregion corresponding to each skeleton line Taking frame t as an example, we analyze the mobility of the suspected voxels from frame t-5 to frame t+5.

[0053] For each skeleton line, extract its boundary smoothness sequence in the time sequence from frame t-5 to t+5. Sequence), attenuation coefficient sequence ( (Sequence). The fluctuations in boundary smoothness and attenuation coefficient of each skeleton line in the time series from frame t-5 to frame t+5 are analyzed. To this end, the difference in boundary smoothness between each adjacent frame is calculated. for: , Let be the boundary smoothness of frame t; the cumulative change from the initial value to the current time t. , ; and the average difference from the initial value to the current t-th frame. , At this point, the smoothness variation coefficient from the initial time to the current t-th frame is calculated, and used to calculate the first variation coefficient. : Similarly, we can obtain The second coefficient of change , , This represents the cumulative change in the attenuation coefficient from its initial value to the current time t. This represents the average difference in the decay coefficient from its initial value to the current time t.

[0054] The morphology coefficient of each skeleton line is obtained based on the changes in smoothness and the effectiveness of pipe diameter attenuation. for: .

[0055] In one embodiment, the morphological coefficients of each skeleton line, the positional changes of suspected voxels in the suspected blood vessel sub-region corresponding to the skeleton line in multiple consecutive images, and the enhancement coefficient of the skeleton line are analyzed to obtain the target enhancement coefficient of each suspected voxel in the suspected blood vessel sub-region corresponding to the skeleton line, including: Using optical flow, the position coordinates of each suspected voxel in the suspected blood vessel sub-region corresponding to each skeleton line are sorted in chronological order in multiple consecutive images to obtain the position sequence of the suspected voxel. Calculate the difference in the position sequence of each suspected voxel between adjacent frames to obtain the movement sequence of the suspected voxel; Obtain the neighborhood of each suspected voxel and calculate the movement sequence of each voxel in the neighborhood; Calculate the cosine similarity between the movement sequence of each suspected voxel and the movement sequence of each suspected voxel in its neighborhood in each adjacent frame, and calculate the mean of the cosine similarity over multiple frames to obtain the similarity between the suspected voxel and the suspected voxel in its neighborhood. The consistency coefficient of a suspected voxel is determined based on the average similarity between the suspected voxel and multiple suspected voxels in its neighborhood. The target enhancement coefficient of a suspected voxel is obtained by taking the mean of the morphological coefficient of each suspected voxel and the enhancement coefficient of its corresponding skeleton line, and the consistency coefficient of the suspected voxel.

[0056] Using optical flow, we obtain the suspected voxels in the suspected vascular connected regions corresponding to any skeleton line. Arrange the position coordinate sequence from frame t-5 to t+5 in order to construct a possible voxel. The position sequence of the suspected voxel was obtained by analyzing the optical flow field features in multiple frames of images. Analyze the temporal consistency of motion; calculate the motion vectors between adjacent frames. The migration sequence of the suspected voxel was obtained. For sequences with fewer than 5 frames at the beginning and end, mirror padding, zero padding, or only features within the valid frame range are calculated.

[0057] At the same time, select suspected voxels The set of all possible voxels within a 3×3×3 neighborhood is denoted as the neighboring voxel set. For any voxel With each adjacent voxel Similarly, the movement sequence of the suspected voxel was calculated. .

[0058] For each element and Analyze the consistency of their motion direction and amplitude, that is, calculate the cosine similarity of the motion vectors of the corresponding frames. The calculation formula is: The closer the value is to 1, the stronger the consistency, meaning that adjacent suspected angiocytes have more similar movement patterns; if there is a value of 0 in the denominator, then... A value of 1 indicates that the default is consistent when there is no movement.

[0059] Calculate the mean of the cosine similarity within 11 frames. As and The global consistency index is then used to calculate the voxels. The suspected voxels are obtained by averaging the global consistency index of all adjacent suspected vascular voxels. Consistency coefficient Normalize it to obtain Combining the aforementioned enhancement coefficients, the target enhancement coefficient for each suspected angiovulin is obtained. : ; in, suspected voxel The enhancement factor of the skeleton line, suspected voxel morphological coefficients, suspected voxel Consistency coefficient.

[0060] , The higher the value, the higher the probability that the suspected blood vessel is a real blood vessel, and the higher the enhancement intensity, especially for the micro-terminal small blood vessel part; A smaller value indicates a high probability of artifacts or non-vascular tissue, and a lower likelihood of it being a blood vessel, which is then directly removed to reduce interference from artifacts and background. This yields the voxel value for each suspected blood vessel within the suspected vascular region. The value is used to enhance blood vessels in different blood supply areas to varying degrees, depending on the magnitude of the value.

[0061] S15. Based on the target enhancement coefficients of multiple suspected voxels in the image, enhance the image and extract cerebral vascular features based on the enhanced image.

[0062] In one embodiment, the image is enhanced based on the target enhancement coefficients of multiple suspected voxels in the image, and cerebral vascular features are extracted based on the enhanced image, including: Using the target enhancement coefficient as the sharpening intensity control parameter for each suspected voxel, Laplacian sharpening is performed frame by frame on the image sequence to generate the enhanced image sequence. From the enhanced image sequence, cerebral vascular features are determined based on the morphological topology parameters and hemodynamic parameters of the blood vessels.

[0063] The Laplacian sharpening algorithm (a known technique) is used to perform targeted enhancement on each frame of suspected vascular images, with the sharpening intensity based on... To make a decision: For non-suspected blood vessel voxels, the sharpening intensity is set to 0 to avoid background or artifact enhancement that would suppress the real blood vessel portion.

[0064] For suspected angiotensin, the sharpening intensity is directly based on the final enhancement coefficient. The value of determines the sharpening intensity; the larger the value, the higher the sharpening intensity, emphasizing the signal edges and details in high-confidence vascular edge areas.

[0065] For each frame of the time-series image, Laplacian sharpening is performed as described above, and a frame-by-frame sharpened and enhanced vascular image sequence is output to ensure improved clarity and recognizability of vascular edges, while suppressing background noise and artifact interference.

[0066] Each sharpened and enhanced image frame is input into a vascular morphology and hemodynamics modeling algorithm. This algorithm first calculates the length and radius of vascular branches using a vascular morphology analysis algorithm, constructing an adjacency matrix of the vascular branches to clarify the connections between vessels. Then, based on these morphological parameters, it establishes a set of hemodynamic control equations including flow rate and pressure. After setting boundary conditions such as inlet and outlet blood pressure, it solves the equations to obtain the flow rate, velocity, and pressure characteristics of each vascular branch. This algorithm is a well-known technique and is suitable for the quantitative extraction of cerebral vascular hemodynamic features from CT and MR images.

[0067] This application also provides a brain blood vessel feature extraction system based on image processing, such as Figure 11 As shown, the system includes: The filtering module 111 is used to filter the acquired images of cerebral blood vessels, obtain the initial blood vessel probability of each voxel, map the image to a preset brain template space, update the initial blood vessel probability according to the characteristics of the anatomical region to which the voxel belongs, and obtain the target blood vessel probability. Extraction module 112 is used to extract multiple skeleton lines of suspected blood vessel regions in the target image. Based on the smoothness of the skeleton line boundary and the difference in tube diameter between multiple skeleton points of the skeleton line, the morphological regularity of each skeleton line is obtained. The target image is obtained by feature enhancement of the image according to the probability of the target blood vessel. The suspected blood vessel region is obtained by threshold segmentation of the target image according to the gray value. Each skeleton line corresponds to a suspected blood vessel sub-region. The suspected blood vessel sub-region is obtained by dividing the suspected blood vessel region into connected components. The calculation module 113 is used to calculate the occlusion rate of the skeleton line based on the distance between each endpoint of the skeleton line and multiple voxels in its neighborhood, the target blood vessel probability of the multiple voxels, and the blood vessel probability information corresponding to the multiple voxels; the occlusion rate indicates the proportion of the occluded area in the skeleton line. Analysis module 114 is used to analyze the differences in the boundary smoothness of each skeleton line in multiple consecutive images, the positional changes of the suspected voxels in the suspected blood vessel sub-region corresponding to the skeleton line in multiple consecutive images, and the enhancement coefficient of the skeleton line, so as to obtain the target enhancement coefficient of each suspected voxel in the suspected blood vessel sub-region corresponding to the skeleton line; the enhancement coefficient of the skeleton line is calculated based on the morphological regularity and occlusion rate of the skeleton line. The enhancement module 115 is used to enhance the image based on the target enhancement coefficients of multiple suspected voxels in the image, and extract cerebral vascular features based on the enhanced image.

[0068] For the system implementation, since it basically corresponds to the method implementation, the relevant parts can be referred to in the description of the method implementation.

[0069] This invention also provides an electronic device, which may include a processor, a memory, and a program stored in the memory and executable on the processor. When the program is executed by the processor, it can perform any of the steps in the above method embodiments and achieve the same beneficial effects, which will not be elaborated further here.

[0070] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0071] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for extracting cerebral blood vessel features based on image processing, characterized in that, The method includes: The acquired images of cerebral blood vessels are filtered to obtain the initial blood vessel probability of each voxel. The images are then mapped to a preset brain template space. The initial blood vessel probability is updated according to the characteristics of the anatomical region to which the voxel belongs, to obtain the target blood vessel probability. Multiple skeleton lines are extracted from the suspected blood vessel region in the target image. The morphological regularity of each skeleton line is obtained based on the smoothness of the skeleton line boundary and the difference in tube diameter between multiple skeleton points of the skeleton line. The target image is obtained by feature enhancement based on the probability of the target blood vessel. The suspected blood vessel region is obtained by threshold segmentation of the target image based on the gray value. Each skeleton line corresponds to a suspected blood vessel sub-region. The suspected blood vessel sub-region is obtained by dividing the suspected blood vessel region into connected components. Based on the distance between each endpoint of the skeleton line and multiple voxels in its neighborhood, the target blood vessel probability of the multiple voxels, and the blood vessel probability information corresponding to the multiple voxels, the occlusion rate of the skeleton line is calculated; the occlusion rate indicates the proportion of the occluded area in the skeleton line. The differences in boundary smoothness of each skeleton line in multiple consecutive images, the positional changes of suspected voxels in the suspected blood vessel sub-region corresponding to the skeleton line in multiple consecutive images, and the enhancement coefficient of the skeleton line are analyzed to obtain the target enhancement coefficient of each suspected voxel in the suspected blood vessel sub-region corresponding to the skeleton line; the enhancement coefficient of the skeleton line is calculated based on the morphological regularity and occlusion rate of the skeleton line. Based on the target enhancement coefficients of multiple suspected voxels in the image, the image is enhanced, and cerebral vascular features are extracted based on the enhanced image.

2. The method for extracting cerebral blood vessel features based on image processing as described in claim 1, characterized in that, The process involves filtering the acquired images of cerebral blood vessels to obtain the initial blood vessel probability for each voxel, mapping the images to a preset brain template space, and updating the initial blood vessel probabilities based on the characteristics of the anatomical region to which the voxel belongs to obtain the target blood vessel probability, including: A multi-scale Hessian matrix filter was used to filter images of cerebral blood vessels to obtain the tubular response value of each voxel at each scale. Based on the largest tubular response value among all voxels, the largest tubular response value of each voxel at multiple scales is normalized to obtain the initial blood vessel probability of that voxel. The image is mapped to a preset brain template space, the anatomical region category of each voxel is determined based on the preset brain region atlas, and the vascular probability information of the corresponding voxel is obtained based on the preset cerebral vascular probability atlas. The initial blood vessel probability is updated based on the anatomical region category and blood vessel probability information of each voxel to obtain the target blood vessel probability.

3. The method for extracting cerebral blood vessel features based on image processing as described in claim 1, characterized in that, The extraction of multiple skeleton lines from suspected blood vessel regions in the target image, based on the smoothness of the skeleton line boundaries and the diameter differences between multiple skeleton points, yields the morphological regularity of each skeleton line, including: The target image is obtained by performing feature enhancement on the image based on the probability of the target blood vessel. For each voxel, calculate the difference between the mean and standard deviation of the gray values ​​in the neighborhood of that voxel to obtain the segmentation threshold for that voxel. Based on the segmentation threshold of each voxel, threshold segmentation is performed on the target image to obtain the suspected blood vessel region; The suspected blood vessel region is divided into connected components to obtain multiple suspected blood vessel sub-regions. A morphological thinning algorithm is applied to each suspected blood vessel sub-region to iteratively erode its boundary until a center line with a single pixel width is obtained, which serves as the skeleton line corresponding to the suspected blood vessel sub-region. Multiple skeleton lines are extracted from the target image. Based on the smoothness of the skeleton line boundary and the difference in vessel diameter between multiple skeleton points of the skeleton line, the morphological regularity of each skeleton line is obtained. Morphological regularity characterizes the regularity of the blood vessel morphology.

4. The method for extracting cerebral blood vessel features based on image processing as described in claim 3, characterized in that, The extraction of multiple skeleton lines from the target image, based on the smoothness of the skeleton line boundaries and the diameter differences between multiple skeleton points, yields the morphological regularity of each skeleton line, including: Extract multiple skeleton lines from the target image, calculate the Euclidean distance between the two sides of the blood vessel walls of each skeleton line, and obtain the diameter of the skeleton line. Obtain the left and right pipe wall point sequences for each skeleton line, and calculate the first curvature sequence of multiple adjacent points in the left pipe wall point sequence and the second curvature sequence of multiple adjacent points in the right pipe wall point sequence respectively. The boundary smoothness of the skeleton line is calculated based on the average of the curvature in the first curvature sequence and the curvature in the second curvature sequence. Calculate the pipe diameter difference sequence between adjacent skeleton points along the direction of each skeleton line, and calculate the attenuation coefficient based on the consistency of the difference direction in the pipe diameter difference sequence; The morphological regularity of each skeleton line is determined based on its boundary smoothness and attenuation coefficient.

5. The method for extracting cerebral blood vessel features based on image processing as described in claim 4, characterized in that, The determination of the morphological regularity of the skeleton line based on the boundary smoothness and attenuation coefficient of each skeleton line includes: The boundary smoothness and attenuation coefficient of each skeleton line are weighted and fused to obtain the morphological regularity of the skeleton line.

6. The method for extracting cerebral vascular features based on image processing as described in claim 1, characterized in that, The occlusion rate of the skeleton line is calculated based on the distance between each endpoint of the skeleton line and multiple voxels in its neighborhood, the target blood vessel probability of the multiple voxels, and the preset probability corresponding to the multiple voxels, including: For each endpoint of the skeleton line, the neighborhood of that endpoint is obtained, and the target blood vessel probability of each voxel in the neighborhood, the reciprocal of the Euclidean distance between the voxel and the endpoint, and the blood vessel probability information corresponding to the voxel are weighted and fused to obtain the continuity coefficient of the voxel. Calculate the mean and standard deviation of the continuity coefficients of multiple voxels in the neighborhood of each endpoint, and determine the difference between the mean and standard deviation of the continuity coefficients as the occlusion threshold. If the continuity coefficient of any voxel in the neighborhood of an endpoint is less than the occlusion threshold, then the endpoint is determined as an occlusion point. For any skeleton line, calculate the ratio of the number of multiple occlusion points to the number of multiple endpoints in the skeleton line to obtain the occlusion rate of the skeleton line.

7. The method for extracting cerebral vascular features based on image processing as described in claim 4, characterized in that, The analysis of the differences in boundary smoothness of each skeleton line in multiple consecutive images, the positional changes of suspected voxels in the suspected blood vessel sub-region corresponding to the skeleton line in multiple consecutive images, and the enhancement coefficient of the skeleton line yields the target enhancement coefficient of each suspected voxel in the suspected blood vessel sub-region corresponding to the skeleton line, including: The enhancement coefficient of each skeleton line is obtained by weighted fusion of its morphological regularity and occlusion rate. Calculate the boundary smoothness sequence and attenuation coefficient sequence for each skeleton line in multiple consecutive images; The first variation coefficient of the boundary smoothness sequence is obtained by calculating the difference between adjacent frames of the boundary smoothness sequence and the average difference of multiple adjacent frames. The second variation coefficient of the attenuation coefficient sequence is obtained by calculating the difference between adjacent frames of the attenuation coefficient sequence and the average difference of multiple adjacent frames. The morphological coefficient of the skeleton line is obtained by multiplying the first variation coefficient and the second variation coefficient. The morphological coefficients of each skeleton line, the positional changes of suspected voxels in the suspected blood vessel sub-region corresponding to the skeleton line in multiple consecutive images, and the enhancement coefficient of the skeleton line are analyzed to obtain the target enhancement coefficient of each suspected voxel in the suspected blood vessel sub-region corresponding to the skeleton line.

8. The method for extracting cerebral vascular features based on image processing as described in claim 7, characterized in that, The analysis of the morphological coefficients of each skeleton line, the positional changes of suspected voxels in the suspected blood vessel sub-region corresponding to the skeleton line in multiple consecutive images, and the enhancement coefficient of the skeleton line yields the target enhancement coefficient of each suspected voxel in the suspected blood vessel sub-region corresponding to the skeleton line, including: Using optical flow, the position coordinates of each suspected voxel in the suspected blood vessel sub-region corresponding to each skeleton line are sorted in chronological order in multiple consecutive images to obtain the position sequence of the suspected voxel. Calculate the difference in the position sequence of each suspected voxel between adjacent frames to obtain the movement sequence of the suspected voxel; Obtain the neighborhood of each suspected voxel and calculate the movement sequence of each voxel in the neighborhood; Calculate the cosine similarity between the movement sequence of each suspected voxel and the movement sequence of each suspected voxel in its neighborhood in each adjacent frame, and calculate the mean of the cosine similarity over multiple frames to obtain the similarity between the suspected voxel and the suspected voxel in its neighborhood. The consistency coefficient of a suspected voxel is determined based on the average similarity between the suspected voxel and multiple suspected voxels in its neighborhood. The target enhancement coefficient of a suspected voxel is obtained by taking the mean of the morphological coefficient of each suspected voxel and the enhancement coefficient of its corresponding skeleton line, and the consistency coefficient of the suspected voxel.

9. The method for extracting cerebral blood vessel features based on image processing as described in claim 1, characterized in that, The step of enhancing the image based on the target enhancement coefficients of multiple suspected voxels in the image, and extracting cerebral vascular features based on the enhanced image, includes: Using the target enhancement coefficient as the sharpening intensity control parameter for each suspected voxel, Laplacian sharpening is performed frame by frame on the image sequence to generate the enhanced image sequence. From the enhanced image sequence, cerebral vascular features are determined based on the morphological topology parameters and hemodynamic parameters of the blood vessels.

10. A brain blood vessel feature extraction system based on image processing, characterized in that, The system includes: The filtering module is used to filter the acquired images of cerebral blood vessels, obtain the initial blood vessel probability of each voxel, map the image to a preset brain template space, and update the initial blood vessel probability according to the characteristics of the anatomical region to which the voxel belongs, so as to obtain the target blood vessel probability. The extraction module is used to extract multiple skeleton lines of suspected blood vessel regions in the target image. Based on the smoothness of the skeleton line boundary and the difference in tube diameter between multiple skeleton points of the skeleton line, the morphological regularity of each skeleton line is obtained. The target image is obtained by performing feature enhancement on the image according to the probability of the target blood vessel. The suspected blood vessel region is obtained by thresholding the target image according to the gray value. Each skeleton line corresponds to a suspected blood vessel sub-region. The suspected blood vessel sub-region is obtained by dividing the suspected blood vessel region into connected components. The calculation module is used to calculate the occlusion rate of the skeleton line based on the distance between each endpoint of the skeleton line and multiple voxels in its neighborhood, the target blood vessel probability of the multiple voxels, and the blood vessel probability information corresponding to the multiple voxels; the occlusion rate indicates the proportion of the occluded area in the skeleton line. The analysis module is used to analyze the differences in the boundary smoothness of each skeleton line in multiple consecutive images, the positional changes of the suspected voxels in the suspected blood vessel sub-region corresponding to the skeleton line in multiple consecutive images, and the enhancement coefficient of the skeleton line, so as to obtain the target enhancement coefficient of each suspected voxel in the suspected blood vessel sub-region corresponding to the skeleton line; the enhancement coefficient of the skeleton line is calculated based on the morphological regularity and occlusion rate of the skeleton line. The enhancement module is used to enhance the image based on the target enhancement coefficients of multiple suspected voxels in the image, and extract cerebral vascular features based on the enhanced image.

Images