A blood vessel image segmentation method, device and electronic equipment

Abstract

The present disclosure provides a blood vessel image segmentation method, device and electronic equipment, which comprises: performing blood vessel edge enhancement processing on three-dimensional blood vessel image data to obtain blood vessel edge enhancement data; performing multi-scale fusion on the blood vessel edge enhancement data to obtain multi-scale blood vessel edge response data; performing distance weighted suppression processing on the multi-scale blood vessel edge response data to obtain weighted blood vessel edge response data; performing adaptive binaryzation processing on the weighted blood vessel edge response data to obtain a blood vessel binary mask; performing multi-modal fusion processing on the blood vessel binary mask and a contrast agent specific signal map to obtain a blood vessel fusion mask; and performing geometric morphological constraint on a connected domain in the blood vessel fusion mask to obtain a target blood vessel segmentation mask. The present disclosure can accurately segment blood vessels from images.

Description

Technical Field

[0001] This disclosure relates to the field of medical imaging technology, and in particular to a method, apparatus and electronic device for segmenting blood vessel images. Background Technology

[0002] In the clinical application of 3D medical imaging, accurate segmentation of vascular structures is a crucial foundation for early diagnosis, preoperative planning, interventional treatment, and postoperative quantitative assessment of vascular diseases, playing a key role in improving diagnostic and treatment accuracy and the reliability of clinical decisions. Existing vascular segmentation methods are easily affected by bone and surrounding tissues, making it difficult to meet the high-precision segmentation requirements of clinical practice. Therefore, a more precise vascular segmentation method is urgently needed. Summary of the Invention

[0003] This disclosure provides a method, apparatus, and electronic device for segmenting blood vessel images, in order to at least solve the above-mentioned technical problems existing in the prior art.

[0004] According to a first aspect of this disclosure, a method for segmenting blood vessel images is provided, the method comprising: Three-dimensional vascular image data is processed to enhance the vascular edge, resulting in enhanced vascular edge data. Multi-scale fusion of the enhanced blood vessel edge data is performed to obtain multi-scale blood vessel edge response data; Distance-weighted suppression processing is performed on the multi-scale blood vessel edge response data to obtain weighted blood vessel edge response data; The weighted blood vessel edge response data is subjected to adaptive binarization to obtain a blood vessel binary mask; The binary vascular mask and the contrast agent-specific signal map are fused using multimodal processing to obtain a vascular fusion mask; Geometric constraints are applied to the connected regions in the vascular fusion mask to obtain the target vascular segmentation mask.

[0005] In one possible implementation, the process of performing vascular edge enhancement processing on the three-dimensional vascular image data to obtain vascular edge enhancement data includes: Gaussian filtering is applied to voxels within the region of interest in the three-dimensional vascular image data to obtain filtered data. Edge response enhancement calculations are performed on the filtered data to obtain enhanced blood vessel edge data.

[0006] In one possible implementation, the multi-scale fusion of the enhanced blood vessel edge data to obtain multi-scale blood vessel edge response data includes: Based on the diameter range of the target blood vessel and the voxel size of the three-dimensional vascular image data, determine the Gaussian kernel scale set; For each scale in the Gaussian kernel scale set, edge response calculation is performed on the enhanced blood vessel edge data to obtain the blood vessel edge response data corresponding to the scale. For each voxel, the maximum value among the blood vessel edge response data corresponding to all scales is taken to obtain multi-scale blood vessel edge response data.

[0007] In one possible implementation, the step of performing distance-weighted suppression processing on the multi-scale vessel edge response data to obtain weighted vessel edge response data includes: For each voxel in the multi-scale blood vessel edge response data, determine the straight-line distance from the voxel to the boundary of the region of interest; A distance field is constructed based on the straight-line distances from all voxels to the boundary of the region of interest; Based on the distance field, spatial adaptive weights are generated; The multi-scale blood vessel edge response data is multiplied by the spatial adaptive weights voxel-by-voxel to obtain weighted blood vessel edge response data.

[0008] In one possible implementation, the adaptive binarization processing of the weighted vessel edge response data to obtain a vessel binary mask includes: Determine the statistical distribution characteristics of weighted vascular edge response data within the core region of the region of interest; Based on the aforementioned statistical distribution characteristics, an adaptive segmentation threshold is determined; The weighted blood vessel edge response data is segmented based on the adaptive segmentation threshold to obtain a binary blood vessel mask.

[0009] In one possible implementation, the step of performing multimodal fusion processing on the binary vascular mask and the contrast agent-specific signal map to obtain a vascular fusion mask includes: The contrast agent-specific signal map was normalized to obtain the normalized contrast agent-specific signal map. Based on the aforementioned binary vascular mask and the normalized contrast agent-specific signal map, an initial fusion mask is obtained; Connectivity analysis was performed on the initial fusion mask to obtain the blood vessel fusion mask.

[0010] In one possible implementation, the step of geometrically constraining the connected regions in the vascular fusion mask to obtain the target vascular segmentation mask includes: For each connected region in the vascular fusion mask, extract its voxel space coordinates and construct a coordinate set; Principal component analysis is performed on the coordinate sets of each connected domain to obtain the principal component eigenvalues; Candidate connected components are obtained by filtering each connected component based on the principal component eigenvalues. The candidate connected components are filtered based on the skeleton length threshold to obtain the target blood vessel segmentation mask.

[0011] In one possible implementation, the method further includes: A three-dimensional sliding window is used to perform block-based reasoning on the segmentation method described in the aforementioned implementation, resulting in multiple block segmentation results; The segmented results are spliced ​​and fused to obtain the target blood vessel segmentation mask.

[0012] According to a second aspect of this disclosure, a blood vessel image segmentation apparatus is provided, the apparatus comprising: The edge enhancement module is used to perform edge enhancement processing on three-dimensional vascular image data to obtain vascular edge enhancement data. A multi-scale fusion module is used to perform multi-scale fusion on the enhanced blood vessel edge data to obtain multi-scale blood vessel edge response data. The suppression module is used to perform distance-weighted suppression processing on the multi-scale blood vessel edge response data to obtain weighted blood vessel edge response data; An adaptive binarization module is used to perform adaptive binarization processing on the weighted blood vessel edge response data to obtain a blood vessel binary mask; The multimodal discrimination module is used to perform multimodal fusion processing on the binary vascular mask and the contrast agent-specific signal map to obtain a vascular fusion mask; The geometric constraint module is used to impose geometric constraints on the connected regions in the vascular fusion mask to obtain the target vascular segmentation mask.

[0013] According to a third aspect of this disclosure, an electronic device is provided, comprising: At least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the methods described in this disclosure.

[0014] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions for causing the computer to perform the methods described in this disclosure.

[0015] The disclosed scheme first performs edge enhancement processing on 3D vascular image data, which effectively highlights vascular edge features and suppresses background interference. Then, through multi-scale fusion, considering the responses of vessels with different diameters, comprehensive edge response data covering the entire vessel diameter is obtained. Next, distance-weighted suppression processing is performed to specifically reduce the influence of boundary artifacts and improve the purity of the edge response data. Then, adaptive binarization processing is used to generate a stable initial vascular binary mask based on the data's own features. Following this, multimodal fusion processing is performed, and contrast agent signal verification is used to further eliminate false positive regions, resulting in a vascular fusion mask with higher purity. Finally, geometric morphological constraints are applied to the connected components in the vascular fusion mask to filter out invalid connected components that do not conform to vascular features, ultimately obtaining a highly accurate and robust target vessel segmentation mask, providing a reliable foundation for subsequent analysis of 3D vascular images.

[0016] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0017] The above and other objects, features, and advantages of this disclosure will become readily apparent from the following detailed description of exemplary embodiments, taken in conjunction with the accompanying drawings. Several embodiments of this disclosure are illustrated in the drawings by way of example and not limitation, in which: In the accompanying drawings, the same or corresponding reference numerals indicate the same or corresponding parts.

[0018] Figure 1 A schematic diagram illustrating the implementation flow of a blood vessel image segmentation method according to an embodiment of the present disclosure is shown; Figure 2 A schematic diagram of a blood vessel image according to an embodiment of the present disclosure is shown; Figure 3 A schematic diagram of the composition structure of a blood vessel image segmentation device according to an embodiment of the present disclosure is shown; Figure 4 A schematic diagram of the composition structure of an electronic device according to an embodiment of the present disclosure is shown. Detailed Implementation

[0019] To make the objectives, features, and advantages of this disclosure more apparent and understandable, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0020] In related technologies, blood vessel segmentation methods are mainly divided into two categories: traditional image processing methods and deep learning-based methods. Traditional image processing methods include blood vessel enhancement methods based on the Hessian matrix, with the Frangi filter being a typical example. This type of method constructs a blood vessel likelihood response function by calculating the three eigenvalues ​​of the image's Hessian matrix, analyzing its distribution characteristics, and then distinguishing between tubular, plate-like, and cluster-like structures based on the differences in grayscale changes along the axial and cross-sectional directions. However, the Frangi filter faces the following problems in CT cerebral angiography: both blood vessels and skulls exhibit high-density signals after iodine contrast enhancement, and both may show tubular or plate-like morphological features at a local scale. The Frangi filter, relying only on the second derivative features of single-channel grayscale information, cannot physically distinguish morphologically similar blood vessels from bones, easily generating a large number of false positive responses in the skull boundary region.

[0021] Deep learning-based blood vessel segmentation methods can automatically learn feature representations through end-to-end training and possess a certain degree of adaptability. However, these methods heavily rely on a large amount of high-quality labeled data. In specific fields such as small animal CT brain imaging, publicly available labeled data is extremely scarce. Researchers can only rely on manually labeling 3D blood vessel masks, a time-consuming and laborious process. Labeling a single whole-brain sample often requires tens to hundreds of hours, severely restricting the application and development of deep learning methods in this field.

[0022] To address the aforementioned technical problems, a first aspect of this disclosure proposes a method for segmenting blood vessel images, such as... Figure 1 As shown, the method includes the following steps: Step 101: Perform vascular edge enhancement processing on the three-dimensional vascular image data to obtain vascular edge enhancement data.

[0023] The three-dimensional vascular image data is three-dimensional volumetric data containing vascular structures acquired using medical imaging equipment (such as photon counting CT), denoted as V(x,y,z). x, y, and z correspond to three coordinates in three-dimensional space, and the voxel value V(x,y,z) represents the tissue density information at the corresponding location. It includes various structures such as blood vessels, skull, brain tissue, and background regions. Because the edges of blood vessels in the original three-dimensional vascular image data are easily blurred by background noise and interference from adjacent tissues, it can easily lead to insufficient edge enhancement and boundary distortion. Therefore, this step first performs targeted edge enhancement processing on the three-dimensional vascular image data to highlight the edges and tubular structural features of the blood vessels, resulting in enhanced vascular edge data.

[0024] Step 102: Perform multi-scale fusion on the enhanced blood vessel edge data to obtain multi-scale blood vessel edge response data.

[0025] Because target blood vessels vary in diameter, such as the significant difference between the diameter of small branch vessels and the main vessels, edge response calculation at a single scale is insufficient to detect vessels of all diameters, easily leading to missed detections of small branch vessels and blurred edges of large vessels. Therefore, this step employs a multi-scale fusion strategy, using edge response calculations and fusion at multiple different scales to effectively capture the edges of vessels of all diameters, ensuring accurate identification of vessels of varying thicknesses.

[0026] Step 103: Perform distance-weighted suppression processing on the multi-scale vessel edge response data to obtain weighted vessel edge response data.

[0027] In multi-scale fused vascular edge response data, the boundaries of regions of interest (ROIs) are susceptible to artifacts due to factors such as partial volume effects and residual background signals. These artifacts can interfere with the accuracy of subsequent vascular segmentation, leading to false positives. Therefore, this step employs a distance-weighted suppression strategy to specifically suppress the response signals in the ROI boundary regions of the multi-scale vascular edge response data, while preserving the effective responses of vessels within the ROI. This results in weighted vascular edge response data that can effectively distinguish between artifacts and vascular signals.

[0028] Step 104: Perform adaptive binarization on the weighted blood vessel edge response data to obtain a blood vessel binary mask.

[0029] After distance-weighted suppression, the vascular edge response has been effectively purified, and ROI boundary artifacts and background interference have been significantly suppressed. However, due to inherent differences in vascular distribution density and grayscale response intensity among different 3D vascular images and individuals, using a fixed threshold for binarization segmentation makes it difficult to balance segmentation accuracy across various application scenarios. This can easily lead to undersegmentation (e.g., missing small vessels) or oversegmentation (e.g., noise being misidentified as vessels), affecting the accuracy of subsequent segmentation processes. Therefore, this step first analyzes the statistical characteristics of the weighted vascular edge response data to determine an adaptive segmentation threshold that fits the actual data characteristics of the current image. Then, based on this adaptive segmentation threshold, the weighted vascular edge response data is binarized, thereby achieving accurate adaptive binarization processing.

[0030] Step 105: Perform multimodal fusion processing on the binary vascular mask and the contrast agent-specific signal map to obtain the vascular fusion mask.

[0031] The vascular binary mask obtained after the above adaptive binarization process, although having initially extracted the vascular foreground region and removed most of the background interference, may still retain a small number of false positive regions. These false positive regions mostly originate from tissues with similar morphology to blood vessels but not enriched with contrast agent (such as the skull). Relying solely on edge response features makes them difficult to effectively distinguish from real blood vessels, and they are easily misidentified as vascular structures. Conversely, vascular regions, due to their contrast agent enrichment, will exhibit significantly high signals on the corresponding substance-specific signal map, while non-vascular tissues (such as the skull and normal brain tissue) will show virtually no corresponding signals. Therefore, the contrast agent-specific signal can be used to achieve accurate identification of vascular regions. In practical applications, the appropriate specific signal map can be flexibly selected according to the type of contrast agent: for example, a contrast agent-specific signal map is used when using iodine-based contrast agents, and a barium density map is used when using barium-based contrast agents; in multi-contrast agent imaging scenarios, multiple verifications can also be performed based on the substance decomposition maps corresponding to each contrast agent. Therefore, this step identifies real vascular regions with contrast agent enrichment characteristics by using contrast agent-specific signal maps, and locks candidate regions with tubular edge characteristics of blood vessels by combining a binary vascular mask. The two are then fused in a multimodal manner to achieve joint discrimination of morphological features and contrast agent signal features, thereby accurately eliminating false positive regions without effective contrast agent signals, preserving the real vascular structure, and obtaining a vascular fusion mask with higher purity and stronger reliability.

[0032] Step 106: Perform geometric morphological constraints on the connected regions in the blood vessel fusion mask to obtain the target blood vessel segmentation mask.

[0033] The vascular fusion mask has eliminated most non-vascular false positive areas through the specific identification of contrast agent-specific signal maps, significantly improving the purity of the vascular mask. However, a small number of small, irregularly shaped noise areas, as well as localized blocky or near-circular non-tubular artifacts, may still remain. These residual areas lack the inherent slender tubular geometric features of blood vessels and differ significantly from the morphology of real blood vessels. If not filtered out, they will interfere with the accuracy and completeness of the final vascular segmentation results, affecting subsequent applications such as vascular morphology analysis and quantitative calculations. Therefore, this step, based on the inherent geometric morphological features of blood vessels, such as slender extension, tubular distribution, and a certain length and aspect ratio, performs targeted constraint screening on all connected regions in the vascular fusion mask. Invalid areas such as tiny noise and non-tubular artifacts that do not conform to the geometric features of blood vessels are filtered out, thereby obtaining a precise, regular target vascular segmentation mask that closely matches the morphology of real blood vessels.

[0034] The vascular image segmentation method in this embodiment first performs vascular edge enhancement processing on the 3D vascular image data, which can effectively highlight vascular edge features and suppress background interference. Then, through multi-scale fusion, taking into account the responses of vessels of different diameters, a comprehensive edge response data covering the entire diameter is obtained. Next, distance-weighted suppression processing is performed to specifically reduce the influence of boundary artifacts and improve the purity of edge response data. Then, adaptive binarization processing is performed to generate a stable initial vascular binary mask based on the data's own features. After that, multimodal fusion processing is performed, and contrast agent signal verification is used to further eliminate false positive regions, resulting in a vascular fusion mask with higher purity. Finally, geometric morphology constraint processing is applied to the connected components in the vascular fusion mask to filter out invalid connected components that do not conform to vascular features, ultimately obtaining a highly accurate and robust target vascular segmentation mask, providing a reliable foundation for subsequent analysis of 3D vascular images.

[0035] In another embodiment of this disclosure, the three-dimensional vascular image data is subjected to vascular edge enhancement processing to obtain vascular edge enhancement data. Specifically, this can be achieved by the following technical means: performing Gaussian filtering on voxels in the region of interest of the three-dimensional vascular image data to obtain filtered data; and performing edge response enhancement calculation on the filtered data to obtain vascular edge enhancement data.

[0036] Since the three-dimensional vascular image data in this application is three-dimensional volume data, and conventional edge enhancement methods (such as XDoG) are usually designed for two-dimensional images, they are difficult to directly apply to three-dimensional volume data processing. Therefore, the processing method is adapted to three dimensions before enhancement. Specifically, two modes can be adopted: (a) a layer-by-layer two-dimensional processing mode, in which edge enhancement is performed independently on each z-layer slice; (b) a true three-dimensional processing mode, which uses a three-directional separable Gaussian convolution kernel to achieve efficient three-dimensional Gaussian filtering. This method preferentially adopts mode (a), which makes full use of the edge structure information within each slice while ensuring computational efficiency.

[0037] Based on this, according to the anatomical structural features of the three-dimensional vascular image data, the region of interest (ROI) containing the target blood vessel is determined by manual delineation or automatic segmentation to eliminate interference from background areas such as external air and irrelevant tissues. Subsequently, Gaussian convolution is performed on the voxels within the ROI, and mask-normalized convolution is used to include only voxels within the ROI in the weighted calculation, avoiding the introduction of background signals from outside the region that could affect the filtering results. This process can be achieved through formula (1): (1) in, The output value of voxel x in the region of interest after mask-normalized Gaussian filtering; A Gaussian kernel function with a standard deviation of σ is used to characterize voxels. x Weighting is applied to surrounding neighborhood voxels; The original voxel value of voxel y in the region of interest is represented; the denominator is a normalization factor, which is used to eliminate the bias caused by the difference in the number of effective neighbors at different locations, and to avoid the problem of abnormal decay of edge values ​​at the ROI boundary due to the participation of background voxels in convolution.

[0038] Furthermore, edge enhancement response calculation is performed on the obtained filtered data to highlight the response signals of blood vessel edges and tubular structures, and suppress interference from background and irrelevant tissues. This can be achieved through the following formula (2): (2) in, Characteristic voxels x After edge enhancement, the response value of the region with a response value greater than 0 corresponds to the structure with local brightness higher than the surrounding average, i.e., the potential blood vessel edge or tubular structure. , The standard deviations are σ and k, respectively. The Gaussian filtering result is σ (where k is the scale factor and k>1); p is the weight parameter used to adjust the weight of the difference between the two Gaussian filtering results, thereby controlling the intensity of edge enhancement.

[0039] Finally, through the above mask-normalized Gaussian filtering and edge enhancement response calculation, the enhanced blood vessel edge data is obtained. In this data, the blood vessel edge and tubular structure are significantly enhanced, and background noise and irrelevant tissue interference are effectively suppressed, providing high-quality input data for subsequent multi-scale fusion, artifact suppression and other steps.

[0040] like Figure 2 As shown in the attached figure, this illustrates the vascular edge response characteristics of the method in this embodiment at different Gaussian kernel scales (σ), and the overall effect after multi-scale fusion: First, the four regional comparison images in the top row present the specific vascular response characteristics at a single scale. As the Gaussian kernel scale σ gradually increases, the response results show an evolutionary pattern from enrichment of small branches to prominence of large vascular trunks. The response image corresponding to the small scale σ focuses on enhancing the edge details of small branches in the brain's vascular network, clearly outlining the texture of small-diameter vessels; while the response image corresponding to the large scale σ is more suitable for the spatial extension characteristics of large vessels, effectively highlighting the edge contour of the vascular trunk, and solving the technical problem that a single scale cannot cover the entire diameter of vessels.

[0041] Secondly, the three magnified images in the bottom row further demonstrate the comprehensive effect of multi-scale fusion. The response image after multi-scale fusion, taking the maximum value, retains both the edge details of small blood vessels at the small scale and the contour features of large blood vessels at the large scale, achieving a synergistic enhancement of the edges of blood vessels of all diameters. Compared to the limitations of a single-scale response, the fused image presents a more complete vascular network, with both details and main trunks preserved, intuitively verifying the effectiveness of this embodiment and ensuring that blood vessels of different diameters can be accurately identified.

[0042] This embodiment effectively suppresses background noise and interference from irrelevant regional signals by performing Gaussian filtering and edge response enhancement calculations only within the region of interest. While highlighting the tubular structure and edge features of blood vessels, it avoids response distortion caused by boundary signal leakage, thereby improving the accuracy and anti-interference ability of blood vessel edge enhancement and providing stable and reliable basic data for subsequent blood vessel segmentation.

[0043] In 3D medical images, the boundaries of regions of interest (such as intracranial regions) are highly irregular. Standard Gaussian convolution at the boundaries includes background voxels (such as air and extracranial tissue) outside the region of interest in the weighted average, resulting in an abnormally low filter response value inside the boundary, which in turn causes blood vessels near the boundary to be missed. Our proposed solution uses masked normalized convolution, which dynamically adjusts the normalization factor to utilize only voxels within the region of interest in the convolution calculation, thus eliminating this boundary effect problem. Thanks to this processing method, blood vessels running close to the inner wall of the skull (such as large veins like the dural sinuses) are correctly enhanced and segmented, no longer incorrectly suppressed due to boundary signal leakage.

[0044] In another embodiment of this disclosure, multi-scale fusion of vascular edge enhancement data is performed to obtain multi-scale vascular edge response data. Specifically, this can be achieved through the following technical means: determining a Gaussian kernel scale set based on the diameter range of the target vascular vessel and the voxel size of the three-dimensional vascular image data; calculating the edge response of the vascular edge enhancement data for each scale in the Gaussian kernel scale set to obtain the vascular edge response data corresponding to the scale; and taking the maximum value among the vascular edge response data corresponding to all scales for each voxel to obtain the multi-scale vascular edge response data.

[0045] First, by combining the actual diameter range of the target blood vessel (e.g., the diameter of blood vessels in the brain of small animals is typically 23~120μm) and the voxel size of the three-dimensional vascular image data, multiple Gaussian kernel sizes of different sizes are determined, forming a Gaussian kernel scale set, denoted as . (N is the number of scales, which can be adjusted according to the actual application scenario). For example, if the voxel size of the 3D vascular image data is 23μm, 7 Gaussian kernel scales can be set, corresponding to σ1 to σ7 respectively, covering the target blood vessel diameter range of 23~120μm. This ensures that each scale can specifically match the blood vessel edge response calculation for the corresponding diameter. The small-scale Gaussian kernel is used to capture the edges of small blood vessels, and the large-scale Gaussian kernel is used to capture the edges of large blood vessels.

[0046] For each scale in the Gaussian kernel scale set Σ (i=1,2,…,N), perform edge response calculations on the enhanced vessel edge data to obtain the vessel edge response data for each scale. Subsequently, for each voxel x, the maximum value among the edge response data corresponding to all scales is selected as the final response value of that voxel, that is, multi-scale fusion is achieved through the following formula (3): (3) in, The final edge response value of voxel x after multi-scale fusion is represented by Σ, where Σ is the Gaussian kernel scale set. Let i be the Gaussian kernel scale in the set; Characterizing voxel x at scale The formula extracts the edge response data at each scale. By taking the maximum value of the response at each scale, it can retain the strongest edge response of each voxel at different scales, ensuring that the edges of small and large blood vessels are fully preserved, and avoiding the problems of missed detection or insufficient response caused by a single scale.

[0047] The solution in this embodiment determines a reasonable set of Gaussian kernel scales by combining the target blood vessel diameter and voxel size, and adopts a multi-scale response calculation and maximum value fusion method. This effectively solves the technical problem that a single scale is difficult to adapt to blood vessels of all diameters, and achieves comprehensive capture of the edges of blood vessels of different thicknesses, thereby improving the integrity and accuracy of blood vessel edge response data.

[0048] In another embodiment of this disclosure, distance-weighted suppression processing is performed on the multi-scale vascular edge response data to obtain weighted vascular edge response data. Specifically, this can be achieved through the following technical means: for each voxel in the multi-scale vascular edge response data, the straight-line distance from the voxel to the boundary of the region of interest is determined; a distance field is constructed based on the straight-line distances from all voxels to the boundary of the region of interest; a spatial adaptive weight is generated based on the distance field; and the multi-scale vascular edge response data is multiplied by the spatial adaptive weight on a voxel-by-voxel basis to obtain the weighted vascular edge response data.

[0049] Specifically, for each voxel x in the multi-scale vascular edge response data, combined with the region of interest (ROI) determined in the previous steps, the shortest straight-line distance from the voxel to the ROI boundary is calculated. A global distance field, denoted as d(x), is constructed based on the straight-line distances of all voxels. The larger the value of d(x), the closer the voxel is to the interior of the ROI, and the more likely it is to be a real vascular region; the smaller the value of d(x), the closer the voxel is to the ROI boundary, and the more likely it is to be an artifact region. Then, a spatial adaptive weight w(x) is generated based on the distance field d(x). The weight decay of the boundary region and the preservation of the weight of the interior region are achieved through nonlinear mapping. Specifically, this can be achieved through the following formula (4): (4) in, The spatial adaptive weights corresponding to voxel x are represented by values ​​in the range (0,1); α is the weight adjustment coefficient, which is used to control the slope of the sigmoid function and thus adjust the rate of weight decay; d(x) is the straight-line distance from voxel x to the boundary of the ROI. The distance threshold is used to divide the ROI boundary region from the interior region. When d(x) > When d(x) approaches 1, the response signal of the voxel is preserved; when d(x) < When w(x) approaches 0, the spurious effect of the voxel is suppressed.

[0050] After the weights are generated, the multi-scale vascular edge response data are multiplied on a voxel-by-voxel basis with the spatial adaptive weights to achieve distance-weighted suppression, which can be achieved through the following formula (5): (5) in, The final response value of voxel x after distance-weighted inhibition (i.e., weighted vessel edge response data); The multi-scale fusion edge response value of voxel x; voxels x The corresponding spatial adaptive weights. This formula can effectively suppress artifacts at the ROI boundary while preserving the edge response of real blood vessels within the ROI to the greatest extent, further purifying the vascular signal.

[0051] This embodiment addresses the technical problem of boundary artifact interference after multi-scale fusion by constructing a distance field to generate spatial adaptive weights and combining them with a voxel-by-voxel weighted multiplication method to specifically suppress ROI boundary artifacts. This method achieves continuous and smooth attenuation control of boundary false positive responses, significantly suppressing false positive responses within 2-5 voxels of the skull inner wall, while leaving normal vascular responses farther from the boundary unaffected. This processing effectively improves the purity of vascular edge response data, reduces false positive interference in subsequent segmentation processes, and provides more accurate input data for subsequent adaptive binarization, multimodal fusion, and other steps.

[0052] In another embodiment of this disclosure, the adaptive binarization processing of the weighted blood vessel edge response data to obtain a blood vessel binary mask can be achieved through the following technical means: determining the statistical distribution characteristics of the weighted blood vessel edge response data in the core region of the region of interest; determining an adaptive segmentation threshold based on the statistical distribution characteristics; and performing threshold segmentation on the weighted blood vessel edge response data based on the adaptive segmentation threshold to obtain a blood vessel binary mask.

[0053] First, a relatively stable core region is determined within the region of interest (ROI). This core region is preferably selected from areas within the ROI that are far from the boundary, minimally affected by artifacts, and have a uniform distribution of vascular structures. This core region can more realistically and accurately reflect the response level of the effective vessel edge, avoiding interference from residual boundary artifacts or local noise on the statistical results. Then, the numerical distribution of weighted vessel edge response data is statistically analyzed within this core region, and corresponding statistical distribution features are extracted. These can be quantiles, mean, variance, etc., with quantiles being the preferred implementation method. For example, selecting the 95th or 97th percentile can effectively avoid the influence of extreme outliers and better reflect the true distribution of vessel edge response.

[0054] Based on the above statistical distribution characteristics, an adaptive segmentation threshold is automatically calculated, denoted as . This allows the threshold to accurately match the actual response distribution of the current 3D vascular image data, adapting to differences between individuals and images. Finally, this adaptive segmentation threshold is used to perform threshold segmentation on the overall weighted vascular edge response data. The specific rule is: weighted vascular edge response values ​​higher than the threshold are segmented. The voxels are identified as vascular foreground and assigned a value of 1; the weighted vascular edge response values ​​are set to be lower than or equal to the threshold. The voxel is determined as the background and assigned a value of 0, and finally the binary mask of blood vessels is obtained, which can be expressed as the following formula (6): (6) This embodiment's solution effectively avoids the problem of insufficient adaptability of fixed thresholds across different data and locations by calculating an adaptive segmentation threshold based on the statistical characteristics of the vascular edge response in the core region. Medical images are affected by various factors such as contrast agent type, injection dosage, and individual differences, resulting in significant variations in grayscale distribution. A fixed threshold may perform well under one condition but fail completely under others. The adaptive quantile threshold used in this solution dynamically determines the segmentation threshold based on the actual response distribution of the core region for each data point, eliminating the need for manual parameter presets and achieving cross-scenario adaptive adaptation. Under three different contrast conditions—iohexol contrast (500 mgI / ml), HDVC iodine pool contrast, and barium-based contrast agent—it consistently generates high-quality vascular segmentation results without requiring separate threshold parameter adjustments for different conditions. This method ensures complete segmentation of both large and small vessels while further reducing false positive areas caused by noise and residual artifacts, significantly improving the accuracy, stability, and robustness of the initial binary mask for vessels.

[0055] In another embodiment of this disclosure, the vascular binary mask and the contrast agent-specific signal map are subjected to multimodal fusion processing to obtain a vascular fusion mask, including: normalizing the contrast agent-specific signal map to obtain a normalized contrast agent-specific signal map; obtaining an initial fusion mask based on the vascular binary mask and the contrast agent-specific signal map; and performing connected component analysis on the initial fusion mask to obtain the vascular fusion mask.

[0056] First, the original contrast agent-specific signal map is normalized, mapping the signal values ​​uniformly to the [0,1] interval, resulting in a normalized contrast agent-specific signal map, denoted as . This operation eliminates signal amplitude differences caused by different scanning protocols, individuals, and contrast agent types, ensuring numerical stability and data adaptability during multimodal fusion. Subsequently, based on a vascular binary mask... Normalized contrast agent specificity signal map Joint discrimination is performed, and only voxels that simultaneously satisfy "the binary mask of blood vessels is determined to be the foreground" and "have effective contrast agent signal" are retained to obtain the initial fusion mask, which can be achieved through formula (7): (7) in, Characterizes the initial fusion mask value corresponding to voxel x; The binary mask value for blood vessels (1 represents the foreground of blood vessels, and 0 represents the background). For the contrast agent signal discrimination function, a threshold discrimination form is preferred: when hour, =1, indicating a valid contrast agent signal region; when hour, =0 indicates no valid iodine signal. By multiplying voxels to achieve joint screening of morphological and signal characteristics, false positive areas without contrast agent enrichment (such as morphologically similar tissues like the skull) can be directly eliminated.

[0057] Finally, connected component analysis was performed on the initial fusion mask. Connected components that conform to the characteristics of blood vessel structure were selected based on indicators such as volume and extensibility. Small isolated regions caused by noise were removed to further purify the mask and finally obtain the blood vessel fusion mask.

[0058] This embodiment combines the morphological features of the vessel edge with the specific signal features of the contrast agent to achieve dual constraints of structure and material, which can significantly improve the reliability of vessel region recognition, effectively remove artifacts and false positive interference, and further optimize the mask quality through connected component analysis, providing a high-precision and robust data foundation for subsequent geometric morphology constraints and final vessel segmentation.

[0059] In another embodiment of this disclosure, the geometric morphological constraint of the connected regions in the vascular fusion mask to obtain the target vascular segmentation mask can be achieved through the following technical means: extracting the voxel space coordinates of each connected region in the vascular fusion mask and constructing a coordinate set; performing principal component analysis on the coordinate set of each connected region to obtain principal component eigenvalues; filtering each connected region according to the principal component eigenvalues ​​to obtain candidate connected regions; and filtering the candidate connected regions based on a skeleton length threshold to obtain the target vascular segmentation mask.

[0060] First, connected component labeling is performed on the vascular fusion mask. The three-dimensional spatial coordinates of all voxels contained in each independent connected component are extracted, forming a coordinate set for that connected component. Then, principal component analysis (PCA) is performed on the coordinate set of each connected component to obtain principal component eigenvalues ​​in three orthogonal directions. The magnitude of the eigenvalues ​​reflects the extent of the connected component's extension in different dimensions: typical blood vessels are elongated tubular, and their eigenvalues ​​show a significant differential distribution; while non-vascular artifacts are mostly clumpy or dotted, with a relatively uniform distribution of eigenvalues. Based on this difference in characteristics, connected components can be morphologically screened, retaining only regions that meet the tubular characteristic condition as candidate connected components.

[0061] Furthermore, the skeleton of the candidate connected regions is extracted and the skeleton length of each connected region is calculated. The skeleton length is compared with a preset skeleton length threshold, and small noise fragments or short rod-shaped artifacts with lengths less than the threshold are removed. Only connected regions with lengths that conform to the physiological structure of blood vessels are retained, and finally a target blood vessel segmentation mask with regular shape and no artifact interference is obtained.

[0062] The scheme in this embodiment achieves quantitative discrimination of the tubular geometric features of blood vessels through principal component analysis, and performs secondary screening by combining skeleton length constraints. This can effectively remove residual non-tubular artifacts and minor noise, making the segmentation results more consistent with the geometric shape of real blood vessels, and significantly improving the accuracy, completeness and robustness of the final blood vessel segmentation results.

[0063] In another embodiment of this disclosure, the method further includes: using a three-dimensional sliding window to perform block reasoning on the segmentation method described in the foregoing embodiments to obtain multiple block segmentation results; and splicing and fusing the block segmentation results to obtain a target blood vessel segmentation mask.

[0064] Specifically, the size of the 3D sliding window can be adaptively adjusted based on the voxel size of the 3D vascular image data, the minimum diameter of the target vessel, and hardware computing resources. The window sliding step size can be set according to the sub-block overlap requirements, preferably using a partial overlap method to reduce segmentation errors at the sub-block edges. During block inference, each sub-block independently performs steps such as edge enhancement, multi-scale fusion, distance-weighted suppression, adaptive binarization, multi-modal fusion, and geometric constraints to ensure that the segmentation accuracy of each sub-block is consistent with that of direct global inference. During stitching and fusion, for overlapping sub-block areas, the final voxel label is determined by averaging or voting. For non-overlapping areas, the corresponding block segmentation results are directly retained, effectively eliminating stitching artifacts at sub-block boundaries and ensuring the global continuity and accuracy of the target vessel segmentation mask.

[0065] The technical effects achieved by this solution are systematically explained through comparative analysis, as shown in Table 1. Compared with related technical solutions, the technical solution disclosed herein has achieved systematic optimization at the mathematical model layer, algorithm architecture layer, and system application layer to adapt to the 3D medical imaging vascular segmentation scenario.

[0066] At the mathematical model level, the relevant technical solutions are designed for two-dimensional single-channel brightness fields, using standard two-dimensional separable Gaussian kernels and fixed thresholds. However, this solution extends signal representation to three-dimensional volume data, constructs a unified filtering framework that supports layer-by-layer two-dimensional processing or true three-dimensional separable convolution, and introduces mask-normalized convolution to solve the signal leakage problem at the boundaries of irregular regions of interest (ROIs). At the same time, an adaptive quantile threshold is adopted based on the response distribution of the core region, replacing the original global fixed threshold.

[0067] At the algorithm architecture level, related technical solutions aim at artistic style control, employing limited multi-scale processing, boundaryless processing, and multimodal fusion capabilities, but do not support large-volume data. This solution, on the other hand, constructs an adaptive multi-scale space for vessel diameter, covering a vessel diameter range of 23–120 μm with 7 scales. It suppresses false positives at ROI boundaries through distance transformation boundary weighting and introduces a spectral-edge joint fusion strategy, utilizing contrast agent-specific signals to verify the authenticity of vessels at the edges. Simultaneously, this solution supports block inference based on a 3D sliding window, which can efficiently process volume data of arbitrary sizes. Furthermore, it eliminates non-tubular artifacts through geometric morphological constraints such as principal component analysis (PCA) for elongation screening and skeleton length verification.

[0068] At the system application level, the relevant technical solutions are mainly used for non-photorealistic rendering (NPR) and artistic stylization in the field of computer graphics, with evaluation primarily based on subjective visual assessment. This solution, however, is geared towards 3D medical image vascular segmentation, achieving effective decoupling of vascular and skeletal signals. It provides objective evaluation through quantitative indicators such as the Dice Similarity Coefficient (Dice) and the Centerline Dice (clDice), and can be applied to research scenarios such as the construction and quantitative analysis of small animal brain vascular networks.

[0069] Table 1

[0070] A second aspect of this disclosure provides a blood vessel image segmentation apparatus, such as... Figure 3 As shown, the device includes: The edge enhancement module 201 is used to perform vascular edge enhancement processing on the three-dimensional vascular image data to obtain vascular edge enhancement data. The multi-scale fusion module 202 is used to perform multi-scale fusion on the enhanced blood vessel edge data to obtain multi-scale blood vessel edge response data. Suppression module 203 is used to perform distance-weighted suppression processing on the multi-scale blood vessel edge response data to obtain weighted blood vessel edge response data; The adaptive binarization module 204 is used to perform adaptive binarization processing on the weighted blood vessel edge response data to obtain a blood vessel binary mask. The multimodal discrimination module 205 is used to perform multimodal fusion processing on the binary vascular mask and the contrast agent-specific signal map to obtain a vascular fusion mask; The geometric constraint module 206 is used to perform geometric morphological constraints on the connected regions in the blood vessel fusion mask to obtain the target blood vessel segmentation mask.

[0071] In another embodiment of this disclosure, the edge enhancement module 201 is further configured to perform Gaussian filtering on voxels in the region of interest of the three-dimensional vascular image data to obtain filtered data; and to perform edge response enhancement calculation on the filtered data to obtain vascular edge enhancement data.

[0072] In another embodiment of this disclosure, the multi-scale fusion module 202 is further configured to determine a Gaussian kernel scale set based on the diameter range of the target blood vessel and the voxel size of the three-dimensional blood vessel image data; for each scale in the Gaussian kernel scale set, perform edge response calculation on the blood vessel edge enhancement data to obtain blood vessel edge response data corresponding to the scale; and for each voxel, take the maximum value among the blood vessel edge response data corresponding to all scales to obtain multi-scale blood vessel edge response data.

[0073] In another embodiment of this disclosure, the suppression module 203 is further configured to: determine the straight-line distance from each voxel in the multi-scale vascular edge response data to the boundary of the region of interest; construct a distance field based on the straight-line distances from all voxels to the boundary of the region of interest; generate spatial adaptive weights based on the distance field; and multiply the multi-scale vascular edge response data with the spatial adaptive weights voxel by voxel to obtain weighted vascular edge response data.

[0074] In another embodiment of this disclosure, the adaptive binarization module 204 is further configured to determine the statistical distribution characteristics of the weighted blood vessel edge response data within the core region of the region of interest; determine an adaptive segmentation threshold based on the statistical distribution characteristics; and perform threshold segmentation on the weighted blood vessel edge response data based on the adaptive segmentation threshold to obtain a blood vessel binary mask.

[0075] In another embodiment of this disclosure, the multimodal discrimination module 205 is further configured to normalize the contrast agent-specific signal map to obtain a normalized contrast agent-specific signal map; obtain an initial fusion mask based on the vascular binary mask and the normalized contrast agent-specific signal map; and perform connected component analysis on the initial fusion mask to obtain a vascular fusion mask.

[0076] In another embodiment of this disclosure, the geometric constraint module 206 is further configured to extract the voxel space coordinates of each connected region in the vascular fusion mask and construct a coordinate set; perform principal component analysis on the coordinate set of each connected region to obtain principal component eigenvalues; filter each connected region according to the principal component eigenvalues ​​to obtain candidate connected regions; and filter the candidate connected regions based on the skeleton length threshold to obtain the target vascular segmentation mask.

[0077] In another embodiment of this disclosure, the device further includes a block processing module, which is used to perform block reasoning on the segmentation method described in the above embodiment using a three-dimensional sliding window to obtain multiple block segmentation results; and to stitch and fuse the block segmentation results to obtain a target blood vessel segmentation mask.

[0078] According to embodiments of this disclosure, this disclosure also provides an electronic device and a readable storage medium.

[0079] Figure 4 A schematic block diagram of an example electronic device that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0080] like Figure 4 As shown, device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 802 or a computer program loaded from storage unit 808 into random access memory (RAM) 803. RAM 803 may also store various programs and data required for the operation of device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via bus 804. Input / output (I / O) interface 805 is also connected to bus 804.

[0081] Multiple components in device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of monitors, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0082] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as the vascular image segmentation method. For example, in some embodiments, the vascular image segmentation method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed on device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the vascular image segmentation method described above can be performed. Alternatively, in other embodiments, the computing unit 801 can be configured to perform the vascular image segmentation method by any other suitable means (e.g., by means of firmware).

[0083] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transferring data and instructions to the storage system, the at least one input device, and the at least one output device.

[0084] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0085] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0086] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0087] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0088] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0089] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this disclosure can be achieved, and this is not limited herein.

[0090] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means two or more, unless otherwise explicitly specified.

[0091] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. A method for segmenting blood vessel images, characterized in that, The method includes: Three-dimensional vascular image data is processed to enhance the vascular edge, resulting in enhanced vascular edge data. Multi-scale fusion of the enhanced blood vessel edge data is performed to obtain multi-scale blood vessel edge response data; Distance-weighted suppression processing is performed on the multi-scale blood vessel edge response data to obtain weighted blood vessel edge response data; The weighted blood vessel edge response data is subjected to adaptive binarization to obtain a blood vessel binary mask; The binary vascular mask and the contrast agent-specific signal map are fused using multimodal processing to obtain a vascular fusion mask; Geometric constraints are applied to the connected regions in the vascular fusion mask to obtain the target vascular segmentation mask.

2. The method according to claim 1, characterized in that, The process of enhancing the vascular edges of the three-dimensional vascular image data to obtain enhanced vascular edge data includes: Gaussian filtering is applied to voxels within the region of interest in the three-dimensional vascular image data to obtain filtered data. Edge response enhancement calculations are performed on the filtered data to obtain enhanced blood vessel edge data.

3. The method according to claim 1, characterized in that, The multi-scale fusion of enhanced blood vessel edge data yields multi-scale blood vessel edge response data, including: Based on the diameter range of the target blood vessel and the voxel size of the three-dimensional vascular image data, determine the Gaussian kernel scale set; For each scale in the Gaussian kernel scale set, edge response calculation is performed on the enhanced blood vessel edge data to obtain the blood vessel edge response data corresponding to the scale. For each voxel, the maximum value among the blood vessel edge response data corresponding to all scales is taken to obtain multi-scale blood vessel edge response data.

4. The method according to claim 1, characterized in that, The distance-weighted suppression processing of the multi-scale vessel edge response data to obtain weighted vessel edge response data includes: For each voxel in the multi-scale blood vessel edge response data, determine the straight-line distance from the voxel to the boundary of the region of interest; A distance field is constructed based on the straight-line distances from all voxels to the boundary of the region of interest; Based on the distance field, spatial adaptive weights are generated; The multi-scale blood vessel edge response data is multiplied by the spatial adaptive weights voxel-by-voxel to obtain weighted blood vessel edge response data.

5. The method according to claim 1, characterized in that, The adaptive binarization processing of the weighted blood vessel edge response data to obtain a blood vessel binary mask includes: Determine the statistical distribution characteristics of weighted vascular edge response data within the core region of the region of interest; Based on the aforementioned statistical distribution characteristics, an adaptive segmentation threshold is determined; The weighted blood vessel edge response data is segmented based on the adaptive segmentation threshold to obtain a binary blood vessel mask.

6. The method according to claim 1, characterized in that, The step of performing multimodal fusion processing on the binary vascular mask and the contrast agent-specific signal map to obtain a vascular fusion mask includes: The contrast agent-specific signal map was normalized to obtain the normalized contrast agent-specific signal map. Based on the aforementioned binary vascular mask and the normalized contrast agent-specific signal map, an initial fusion mask is obtained; Connectivity analysis was performed on the initial fusion mask to obtain the blood vessel fusion mask.

7. The method according to claim 1, characterized in that, The step of geometrically constraining the connected regions in the vascular fusion mask to obtain the target vascular segmentation mask includes: For each connected region in the vascular fusion mask, extract its voxel space coordinates and construct a coordinate set; Principal component analysis is performed on the coordinate sets of each connected domain to obtain the principal component eigenvalues; Candidate connected components are obtained by filtering each connected component based on the principal component eigenvalues. The candidate connected components are filtered based on the skeleton length threshold to obtain the target blood vessel segmentation mask.

8. The method according to claim 1, characterized in that, The method further includes: A three-dimensional sliding window is used to perform block-based reasoning on the segmentation method described in claim 1 to obtain multiple block segmentation results; The segmented results are spliced ​​and fused to obtain the target blood vessel segmentation mask.

9. A blood vessel image segmentation device, characterized in that, The device includes: The edge enhancement module is used to perform edge enhancement processing on three-dimensional vascular image data to obtain vascular edge enhancement data. A multi-scale fusion module is used to perform multi-scale fusion on the enhanced blood vessel edge data to obtain multi-scale blood vessel edge response data. The suppression module is used to perform distance-weighted suppression processing on the multi-scale blood vessel edge response data to obtain weighted blood vessel edge response data. An adaptive binarization module is used to perform adaptive binarization processing on the weighted blood vessel edge response data to obtain a blood vessel binary mask; The multimodal discrimination module is used to perform multimodal fusion processing on the binary vascular mask and the contrast agent-specific signal map to obtain a vascular fusion mask; The geometric constraint module is used to impose geometric constraints on the connected regions in the vascular fusion mask to obtain the target vascular segmentation mask.

10. An electronic device, characterized in that, include: At least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-8.

Images