Medical Image Processing Methods and Systems

Abstract

This application provides a medical image processing method and system. The method includes: acquiring vascular image volume data from photon-counting computed tomography scans of a target region; based on the vascular image volume data, performing skeletonization processing on the vascular network, extracting multiple skeleton points in the vascular network, and forming a vascular centerline network according to the connection relationships between the skeleton points; identifying endpoints and bifurcation points from the vascular centerline network, thereby dividing the vascular centerline network into several continuous and unbifuzzy vascular segments, and constructing a vascular centerline diagram structure based on each vascular segment; establishing and solving a blood flow analysis model based on the vascular centerline diagram structure to obtain the blood flow analysis results of the vascular network; wherein, the blood flow analysis results include the blood flow rate and / or pressure difference between the two ends of each vascular segment. This application improves the accuracy of the blood flow analysis results.

Description

Technical Field

[0001] This application relates to the field of medical image processing technology, and in particular to a medical image processing method and system. Background Technology

[0002] The vascular system is characterized by large variations in vessel diameter, complex bifurcation structures, and diverse spatial orientations. In certain diseased areas, supplying, functional, and draining vessels intertwine to form complex vascular networks. When assessing these networks, researchers not only need accurate knowledge of the vessel geometry but also require an understanding of blood flow distribution among different branches, pressure changes at bifurcation points, and the overall blood flow load to develop appropriate treatment plans.

[0003] However, existing blood flow analysis methods usually rely on simplified vascular connections, making it difficult to automatically extract complete vascular topology from actual images. As a result, when faced with densely branched and complex vascular networks, they cannot fully and accurately reflect the connection relationships between various vascular branches and blood flow pathways in complex vascular networks. Summary of the Invention

[0004] In view of the shortcomings of the prior art described above, the purpose of this application is to provide a medical image processing method and system, which improves the technical problem that the prior art is limited by conventional image resolution and imaging quality, making it difficult to accurately extract small blood vessels and complex bifurcation areas, resulting in inaccurate blood flow analysis results.

[0005] To achieve the above and other related objectives, one embodiment of this application provides a medical image processing method, comprising: acquiring vascular image volume data from photon-counting computed tomography scans of a target region; performing skeletonization processing on the vascular network based on the vascular image volume data, extracting multiple skeleton points in the vascular network, and forming a vascular centerline network according to the connection relationship between the skeleton points; identifying endpoints and bifurcation points from the vascular centerline network, thereby dividing the vascular centerline network into several continuous and unbifuzzy vascular segments, and constructing a vascular centerline diagram structure based on each vascular segment; establishing a blood flow analysis model based on the vascular centerline diagram structure and solving it to obtain the blood flow analysis results of the vascular network; wherein, the blood flow analysis results include the blood flow rate and / or pressure difference of each vascular segment.

[0006] In one embodiment of this application, the steps of performing skeletonization processing on the vascular network based on the vascular image volume data, extracting multiple skeleton points in the vascular network, and forming a vascular centerline network according to the connection relationship between the skeleton points include: inputting the vascular image volume data into a vascular segmentation model to extract the vascular region; extracting multiple skeleton points located at the central axis of the vascular region based on the skeletonization algorithm, and calculating the distance from each skeleton point to the corresponding vascular wall, using it as the local radius of the corresponding skeleton point; and connecting the skeleton points carrying the local radius to each other according to the connection relationship of each skeleton point in the vascular region to form a vascular centerline network.

[0007] In one embodiment of this application, the step of inputting the vascular image volume data into a vascular segmentation model to extract the vascular region includes: inputting the vascular image volume data into the vascular segmentation model, segmenting an initial vascular region from the vascular image volume data; and performing region correction on the initial vascular region to obtain the final vascular region.

[0008] In one embodiment of this application, the blood vessel segmentation model is a three-dimensional blood vessel segmentation model including an encoder and a decoder. The step of inputting the blood vessel image volume data into the blood vessel segmentation model and segmenting an initial blood vessel region from the blood vessel image volume data includes: inputting the blood vessel image volume data into the encoder of the three-dimensional blood vessel segmentation model to downsample the blood vessel image volume data step by step to obtain blood vessel features of various different scales; inputting the blood vessel features of various different scales into the decoder of the three-dimensional blood vessel segmentation model to obtain decoded features of various different scales through step by step upsampling, and fusing each decoded feature with the corresponding scale blood vessel feature across layers to obtain the initial blood vessel region; wherein the encoder and the decoder have a corresponding scale relationship.

[0009] In one embodiment of this application, the step of performing region correction on the initial vascular region to obtain the final vascular region includes: performing connectivity analysis on the initial vascular region to identify multiple connected regions in the initial vascular region; and selecting the largest one from each connected region as the final vascular region.

[0010] In one embodiment of this application, the step of extracting multiple skeleton points located at the central axis of the blood vessel region based on the skeletonization algorithm, and calculating the distance from each skeleton point to the corresponding blood vessel wall, and using it as the local radius of the corresponding skeleton point, includes: based on the skeletonization algorithm, calculating the distance from each voxel in each blood vessel cross-section of the blood vessel region to the blood vessel wall, selecting the voxel with the largest distance as the skeleton point corresponding to the blood vessel cross-section, and using the distance from each skeleton point to the blood vessel wall as the local radius of the corresponding skeleton point.

[0011] In one embodiment of this application, the step of identifying endpoints and bifurcation points from the vascular centralline network, dividing the vascular centralline network into several continuous and unbifuzzy vascular segments accordingly, and constructing a vascular centralline diagram structure based on each of the vascular segments includes: identifying endpoints and bifurcation points from the vascular centralline network, dividing the vascular centralline network into several continuous and unbifuzzy vascular segments accordingly; using the connection points at both ends of each vascular segment as nodes, and using each vascular segment as an edge connecting the nodes to form a vascular centralline diagram structure; wherein each edge of the vascular centralline diagram structure carries the coordinates of the corresponding skeleton point and the local radius.

[0012] In one embodiment of this application, the steps of establishing and solving a blood flow analysis model based on the vascular centerline diagram structure to obtain the blood flow analysis results of the vascular network include: for each vascular segment that serves as an edge: based on the coordinates of each skeleton point in the vascular segment, determining the distance from each skeleton point to the first skeleton point in the vascular segment, and using it as the arc length of each skeleton point; weighting the local radii corresponding to each skeleton point in the vascular segment according to the arc length of each skeleton point to obtain the equivalent radius of the vascular segment; obtaining the hydraulic resistance of the vascular segment based on the equivalent radius of the vascular segment and the arc length of the last skeleton point; in the vascular centerline diagram structure, based on the hydraulic resistance of each vascular segment, establishing and solving a blood flow analysis model according to preset constraints to obtain the blood flow rate and / or pressure difference between the two ends of each vascular segment, and using it as the blood flow analysis result.

[0013] In one embodiment of this application, in the vascular centerline diagram structure, the step of establishing and solving a blood flow analysis model based on the hydraulic resistance of each vascular segment and according to preset constraints to obtain the blood flow rate and / or the pressure difference between the two ends of each vascular segment, and using these as the blood flow analysis results, includes: establishing a blood flow analysis model based on the hydraulic resistance of each vascular segment and according to preset constraints in the vascular centerline diagram structure; solving the blood flow analysis model to obtain the pressure of each node in the vascular centerline diagram structure and the blood flow rate of the corresponding vascular segment; and using the blood flow rate of each vascular segment and / or the pressure difference between the two ends of each vascular segment calculated from the pressure of each node as the blood flow analysis results of each vascular segment.

[0014] In one embodiment of this application, in the vascular centerline diagram structure, the step of establishing a blood flow analysis model based on the hydraulic resistance of each vascular segment and according to preset constraints includes: for each vascular segment serving as an edge in the vascular centerline diagram structure: based on the hydraulic resistance of the vascular segment, determining the mapping relationship between the pressure difference between the two ends of the vascular segment and the blood flow rate. ;in, The blood flow rate of the aforementioned blood vessel segment e. The hydraulic resistance of the blood vessel segment e is... The pressure at the inlet node of the blood vessel segment e. The pressure at the outlet node of the blood vessel segment e is determined; based on the structure of the blood vessel centerline diagram, the network inlet node and network outlet node of the blood vessel network are determined, and preset inlet constraints and outlet constraints are set at the network inlet node and the network outlet node respectively; based on the mapping relationship between the pressure difference at both ends of each blood vessel segment and blood flow, the inlet constraints and the outlet constraints, a blood flow analysis model is established.

[0015] In one embodiment of this application, the ingress constraint is a preset pressure and blood flow of the network node, and the egress constraint is... ,in, The preset reference pressure, The preset export resistance parameters, The pressure of the network egress node to be solved. The blood flow rate of the blood vessel segment corresponding to the network exit node to be solved is denoted as .

[0016] In one embodiment of this application, the method further includes: calculating the blood flow velocity of each of the blood vessel segments based on the blood flow rate and equivalent radius of each blood vessel segment, and visually displaying the blood flow velocity in the blood vessel network.

[0017] In one embodiment of this application, a medical image processing system is also provided. The system includes: a volume data acquisition module for acquiring vascular image volume data from photon-counting computed tomography scans of a target region; a centralization module for performing skeletonization processing on the vascular network based on the vascular image volume data, extracting multiple skeleton points in the vascular network, and forming a vascular centerline network according to the connection relationship between the skeleton points; wherein each skeleton point in the vascular centerline network includes its coordinates in the vascular image volume data and the distance from the skeleton point to the corresponding vessel wall; a graph construction module for identifying endpoints and bifurcation points from the vascular centerline network, dividing the vascular centerline network into several continuous and unbifurated vascular segments, and constructing a vascular centerline graph structure based on each vascular segment; and a blood flow analysis module for establishing a blood flow analysis model based on the vascular centerline graph structure and solving it to obtain the blood flow analysis results of the vascular network; wherein the blood flow analysis results include the blood flow rate and / or pressure difference of each vascular segment.

[0018] The beneficial effects of this application are as follows: The medical image processing method and system proposed in this application perform skeletonization processing on the vascular image volume data generated by photon counting computed tomography of the target area to obtain a vascular centerline network. Based on this, the vascular network is divided into continuous and unbranched vascular segments, forming a vascular centerline map structure. This not only preserves the true morphology and connection relationship of the blood vessels but also reduces the representation complexity of the vascular structure. On this basis, a blood flow analysis model is established and solved based on the vascular centerline map structure. This allows for accurate acquisition of blood flow and / or pressure difference between the two ends of each vascular segment without the need for complex three-dimensional fluid simulation, resulting in more accurate blood flow analysis results. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0020] In the attached diagram: Figure 1 A schematic flowchart of a medical image processing method provided in an embodiment of this application; Figure 2 This is a schematic diagram of three-dimensional vascular image volume data provided in an embodiment of this application; Figure 3 This is a schematic diagram of a vascular centerline network provided in an embodiment of this application; Figure 4 A schematic diagram of a three-dimensional visualization of a vascular centerline network provided in an embodiment of this application; Figure 5 This is a schematic diagram of the visualization results of blood flow velocity distribution based on the blood vessel centerline provided in one embodiment of this application; Figure 6 This is a structural block diagram of a medical image processing system provided in one embodiment of this application; Figure 7 This is a schematic diagram of the structure of an electronic device provided in one embodiment of this application. Detailed Implementation

[0021] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0022] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. The drawings only show the components related to this application and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0023] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the present application. However, it will be apparent to those skilled in the art that embodiments of the present application may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the present application.

[0024] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution of this application all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0025] The inventors of this application have discovered that complex vascular networks, exemplified by intracranial vessels, often exhibit characteristics such as small vessel diameters, numerous bifurcations, and dense interweaving in actual clinical imaging. This is particularly true in cerebral arteriovenous malformations (AVMs), where feeding arteries, multiple abnormal vessels within the malformation cluster, and draining veins interconnect to form a structurally complex vascular network. In this clinical setting, existing vascular images acquired using conventional computed tomography angiography (CTA) or energy-integrating computed tomography (EICT) are often limited by spatial resolution, imaging noise, and local volume effects. This often results in difficulties in accurately extracting the boundary structures of small vessels and complex bifurcations, leading to incomplete or discontinuous vascular extraction results. This instability in vascular extraction further affects subsequent blood flow analysis, severely reducing the accuracy of blood flow calculations. Furthermore, traditional 3D computational fluid dynamics (CFD) often requires constructing high-quality volume meshes based on a complete 3D vascular model and pre-setting complex boundary conditions at the inlet and outlet of the 3D vascular model. This computational process demands significant computational power and time, making it difficult to support rapid assessment in clinical scenarios and repeated simulations of multiple treatment plans. Moreover, in clinical practice, researchers focus more on blood flow characteristics such as shunts between blood supply branches, pressure changes at key vascular bifurcation points, and the overall load state of the drainage pathway. However, 3D CFD-based analysis methods typically focus on calculating local blood flow fields, often resulting in low computational efficiency and complex modeling processes when rapidly simulating the entire vascular network.

[0026] To address the aforementioned issues, this application provides a medical image processing method. This method involves skeletonizing the vascular image data generated from photon-counting computed tomography (CT) scans of the target region to obtain a vascular centerline network. Based on this network, the vascular network is divided into several continuous, unbranched vascular segments, forming a vascular centerline diagram structure. This not only preserves the true morphology and connectivity of the blood vessels but also reduces the complexity of vascular structure representation. Furthermore, a blood flow analysis model is established and solved based on this vascular centerline diagram structure. This allows for accurate acquisition of blood flow and / or pressure differences between the two ends of each vascular segment without the need for complex three-dimensional fluid simulation, resulting in more accurate blood flow analysis results.

[0027] like Figure 1 As shown, the medical image processing method includes the following steps: S100: Acquire vascular image data of the target area using photon counting computed tomography.

[0028] Photon-Counting Computed Tomography (PCCT) was used to scan the target region to obtain raw projection data recorded by a photon-counting detector. Compared to traditional energy-integrating computed tomography (ECT), PCCT achieves higher spatial resolution and effectively reduces imaging noise by counting incident X-ray photons photon-by-photon and distinguishing their energy information. This allows for clearer visualization of small blood vessels and complex vascular structures. Based on the raw projection data, volumetric data reconstruction of the target region was performed according to preset reconstruction parameters to generate three-dimensional vascular volumetric data. This vascular volumetric data characterizes the distribution and morphological features of blood vessels within the target region. The target region refers to the area where the blood vessels to be analyzed are located. The target region can be adaptively selected based on the actual medical application scenario to limit the scanning range of PCCT and generate corresponding vascular volumetric data. For example, the target region is the brain region.

[0029] Preferably, the vascular image volume data is ultra-high resolution (UHR) reconstructed volume data. Compared to ordinary photon-counting computed tomography (CT), UHR reconstructed volume data has a smaller voxel size and a higher spatial sampling density, thus more clearly presenting small blood vessels and complex bifurcation structures, facilitating subsequent vessel segmentation and centerline extraction. In this application, ultra-high resolution reconstructed volume data refers to vascular image volume data reconstructed during photon-counting computed tomography imaging in ultra-high resolution mode through a combination of ultra-high resolution acquisition and / or ultra-high resolution image reconstruction, which has higher spatial resolution and clearer vascular boundary details. The ultra-high resolution mode may include, but is not limited to: using detectors / acquisition configurations with higher spatial sampling density, using smaller reconstructed voxel sizes (e.g., thinner reconstruction slice thickness and smaller pixel pitch), and / or using reconstruction kernel functions / reconstruction parameter settings that enhance edge details, thereby improving the characterization ability of small vascular structures and vessel wall boundaries.

[0030] In a preferred embodiment, the reconstruction slice thickness of the vascular image volume data reconstructed in ultra-high resolution mode can be set to less than or equal to 0.2 mm (e.g., 0.2 mm) to obtain finer vascular edge details and more complete display of small vessels. Optionally, the ultra-high resolution mode can also be combined with one or more of the following parameter settings: the reconstruction matrix adopts a matrix size not lower than a preset threshold to increase the in-plane sampling density; a reconstruction kernel function / reconstruction filter setting is adopted to enhance edge details to improve the gradient response of the vessel wall boundary; a smaller voxel size (e.g., a combination of smaller pixel pitch and thinner slice thickness) is adopted to reduce intravoxel mixing effects. Based on the above settings, the ultra-high resolution mode can enable the vascular image volume data to meet the preset spatial resolution requirements and can reduce the minimum vessel scale threshold for stable segmentation, thereby improving the visibility and segmentability of small vessel segments and bifurcation structures.

[0031] It should be noted that in this application, whether it is the vascular image volume data generated by photon counting computed tomography or the ultra-high resolution reconstructed volume data, since this volume data itself has high spatial resolution and low noise level, no preprocessing operations such as image enhancement or image denoising are required before vascular segmentation.

[0032] S200. Based on the vascular image data, the vascular network is processed into a skeleton, multiple skeleton points in the vascular network are extracted, and a vascular centerline network is formed according to the connection relationship between each skeleton point.

[0033] Skeletonization is performed on vascular imaging data. Voxels located at the center of the vessel lumen are extracted to obtain skeleton points at the center of the vessel cross-section, thus forming a skeleton point set. Further, a vascular centerline network is constructed from the skeleton point set based on the spatial adjacency relationships between adjacent skeleton points. This vascular centerline network, composed of multiple skeleton points, characterizes the vessel's orientation, connectivity, and bifurcation structure in three-dimensional space; therefore, it can be considered as the central axis of the vessel.

[0034] In an optional embodiment of this application, step S200 includes the following steps S210 to S230: S210. Input the vascular image data into the vascular segmentation model and extract the vascular region.

[0035] Vascular image data is input into a trained vascular segmentation model. Voxel-by-voxel analysis is performed on the vascular image data to identify voxel regions belonging to blood vessels, which are then designated as vascular regions. The vascular segmentation model can be any deep learning-based segmentation model, including but not limited to 3D segmentation models or a combination of 2D segmentation models and 3D reconstruction. For example, 3D segmentation models include, but are not limited to, 3D U-Net, nnU-Net, and V-Net; 2D segmentation models include, but are not limited to, U-Net, DeepLab series models, and FCN, and can be combined with slice stacking to form 3D vascular regions. The specific model structure and implementation method are not limited. It should be noted that the process of obtaining the vascular regions described above is the same as the process of obtaining the initial vascular regions based on the vascular segmentation model described below; to avoid repetition, the specific processes will not be detailed further.

[0036] Furthermore, to adapt to the available video memory resources and the size of the vascular image volume data, the vascular image volume data can be segmented into several volume data blocks using a sliding window or block input method. These blocks are then sequentially input into the vascular segmentation model for processing, resulting in vascular sub-regions for each volume data block. These sub-regions are then stitched together according to their corresponding volume data positions within the complete vascular image volume data to form a complete vascular region. It should be noted that the above block input and result stitching process only involves spatial division and combination; it does not alter the intensity distribution of the original vascular image volume data and therefore does not constitute image preprocessing operations such as denoising or enhancement.

[0037] In an optional embodiment of this application, step S210 includes the following steps S211 and S212: S211. Input the vascular image volume data into the vascular segmentation model, and segment the initial vascular region from the vascular image volume data.

[0038] Vascular image data is input into a vascular segmentation model. Voxel-by-voxel analysis is performed on the vascular image data to identify voxel regions belonging to blood vessels within the image data. The identified voxel regions are then aggregated to form an initial vascular region. This initial vascular region is used to characterize the approximate distribution range of blood vessels in three-dimensional space.

[0039] In an optional embodiment of this application, the blood vessel segmentation model is a three-dimensional blood vessel segmentation model including an encoder and a decoder. Step S211 includes the following process: inputting the blood vessel image volume data into the encoder of the three-dimensional blood vessel segmentation model, downsampling the blood vessel image volume data step by step to obtain blood vessel features of various different scales; inputting the blood vessel features of various different scales into the decoder of the three-dimensional blood vessel segmentation model, upsampling step by step to obtain decoded features of various different scales, and fusing each decoded feature with the corresponding scale blood vessel features across layers to obtain an initial blood vessel region; wherein, the encoder and the decoder have a corresponding scale relationship.

[0040] The vascular image volume data is input into the encoder, where it undergoes progressive downsampling through multi-layer 3D convolutional operations to extract vascular features at different spatial scales, resulting in vascular features of various scales. The decoder progressively upsamples these multi-scale vascular features and, through skip connections, fuses the decoded features at the corresponding scale from the upsampling process with the corresponding vascular features from the encoder across layers. This fused feature is then used as the input to the next upsampling layer of the decoder, progressively restoring the spatial resolution of the fused features until they match the spatial resolution of the input vascular image volume data. The final fused feature output from the decoder is used as the prediction result, and classification prediction is applied to determine whether each voxel in the vascular image volume data is a blood vessel or not, thus obtaining the initial vascular region. This skip connection approach not only restores the spatial resolution of the fused features but also preserves the detailed structure and connectivity of the blood vessels, effectively reducing the missed detection of small blood vessels and vascular breaks, significantly improving the accuracy of vascular segmentation results.

[0041] It should be noted that the above prediction results can be either a vascular probability map or a multi-class probability map. Specifically, when the prediction result is a vascular probability map, it can be represented as follows: ,in, voxels The predicted probability of belonging to a vascular region can be determined by thresholding based on the vascular probability map, as shown in formula (1): (1) in, This indicates whether voxel x is a vascular voxel, for example... If the value is 1, then voxel x is a vascular voxel, or If the value is 0, then voxel x is a vascular voxel. The preset mask threshold, This is an indicator function; it takes a value of 1 when the condition within the parentheses is true, and a value of 0 otherwise. Through the above process, we can select vessels from the blood vessel probability map whose predicted probability is greater than or equal to the mask threshold. The voxels are used to form the corresponding vascular mask, which serves as the initial vascular region.

[0042] In another alternative embodiment, when the prediction result is a multi-class probability map, it can be represented as follows: ,in, Let x be the multi-class prediction probability vector corresponding to voxel x. This represents the number of different types of vascular structures. The above multi-class prediction probability vector... The kth component is denoted as , where is the predicted probability that voxel x belongs to the k-th type of vascular structure. In this embodiment, the category mask of voxel x can be determined by the maximum probability principle, as shown in formula (2): (2) in, This indicates whether voxel x is a mask voxel corresponding to the k-th type of vascular structure. For example, If the value is 1, then voxel x is the mask voxel corresponding to the kth type of blood vessel structure, or If 0, then voxel x is the mask voxel corresponding to the k-th type of vascular structure, and argmax is the voxel that makes the prediction probability among all vascular structure categories 0. The highest category index. For example, vascular structures include, but are not limited to, feeding arteries, draining veins, or malformations. Using the above method, a single-category vascular mask or a multi-category vascular structure mask can be generated according to the output format of the vascular segmentation model, providing a data foundation for subsequent extraction of the vascular centerline. Specifically, a single-category vascular mask can be directly used as the initial vascular region, while for a multi-category vascular structure mask, at least one vascular structure can be selected as the initial vascular region.

[0043] Furthermore, to address the issues of the small proportion of blood vessel voxels in the overall vascular image data and the thinness of blood vessel boundaries, the parameters of the blood vessel segmentation model can be optimized and adjusted during the training process using a combination of one or more loss functions. In one embodiment, the Dice loss function as shown in formula (3) can be used. : (3) in, For example, a real blood vessel mask label for voxel x. If the value is 1, then voxel x is a vascular voxel. If the value is 0, then voxel x is a non-vascular voxel. voxels Predicted probability of belonging to a vascular region These are preset smoothing parameters. Dice loss, by measuring the degree of overlap between the predicted result and the real mask, can effectively alleviate the training bias problem caused by the imbalance in the number of blood vessels and background voxels.

[0044] In another embodiment, or when used in conjunction with the Dice loss, the binary cross entropy (BCE) loss as shown in Equation (4) may also be used. : (4) The meanings of the parameters mentioned above are consistent with those of the Dice loss described earlier, and will not be elaborated upon further. Binary cross-entropy loss, by measuring the classification error of each voxel, can improve the recognition accuracy of the blood vessel segmentation model in the blood vessel boundary region. Optionally, focal loss or Tversky loss can also be introduced to further enhance the blood vessel segmentation model's learning ability for difficult-to-separate samples, small blood vessels, or boundary regions; the specific form is not limited.

[0045] In a preferred embodiment, the training objective function of the blood vessel segmentation model can be expressed as a weighted combination of multiple loss functions: (5) Where L is the total loss, and These are the preset weighting coefficients. The blood vessel segmentation model trained in the above manner can improve the ability to identify small blood vessel structures while ensuring overall segmentation stability, thus enabling more accurate extraction of the blood vessel centerline in subsequent steps.

[0046] S212. Perform region correction on the initial vascular region to obtain the final vascular region.

[0047] Considering that when performing voxel-by-voxel analysis on vascular image volume data, the vascular segmentation model may be affected by factors such as imaging noise and voxel resolution limitations, resulting in local missegmentation and discontinuous boundaries in the generated vascular region. To improve this situation, region correction can be performed on the vascular region generated by the model. Region correction is used to eliminate missegmented voxels and smooth and unify vascular boundaries, thereby obtaining a more continuous vascular region that conforms to the true morphology of the blood vessel. Region correction methods include, but are not limited to, removing small-volume pseudo-regions based on connected component analysis and filling the vascular region based on morphological operations. Specifically, in an optional embodiment of this application, step S212 includes the following process: performing connectivity analysis on the initial vascular region to identify multiple connected regions within the initial vascular region; selecting the largest of these connected regions as the final vascular region.

[0048] After obtaining the initial vascular region, connectivity analysis is performed to identify multiple connected regions. Connectivity analysis determines the spatial connectivity between voxels within the vascular region, thus dividing the initial vascular region into several disconnected connected regions. Each connected region corresponds to a spatially independent vascular region. The identified connected regions are then filtered based on the number of voxels they contain, and the region with the largest number of voxels is selected as the final vascular region. This method effectively eliminates scattered vascular regions caused by noise, missegmentation, or local artifacts, retaining vascular regions corresponding to the main vascular structures within the target area.

[0049] In another optional embodiment of this application, step S212 includes the following process: performing connectivity analysis on the initial vascular region to identify multiple connected regions in the initial vascular region; screening each connected region to delete connected regions with a voxel content less than a preset voxel threshold, and using the remaining connected regions as the final vascular region.

[0050] After obtaining the initial vascular region, connectivity analysis is performed to determine the spatial connections between voxels within the region, thereby dividing the initial vascular region into multiple independent connected regions. Each connected region corresponds to a spatially independent vascular structure. The identified connected regions are then filtered based on the number of voxels they contain. Connected regions with fewer than a preset voxel threshold are deleted, and the remaining connected regions are collectively used as the final vascular region. This method effectively removes scattered, isolated non-target regions, retaining the vascular region corresponding to the main vascular structures within the target region.

[0051] In another optional embodiment of this application, the above-mentioned connectivity analysis of the initial vascular region to identify multiple connected regions in the initial vascular region includes the following process: performing connectivity analysis on the initial vascular region to identify multiple initial connected regions in the initial vascular region; performing morphological closing operation or hole filling on each initial connected region to obtain multiple final connected regions.

[0052] After obtaining multiple initial connected regions through connectivity analysis, the following morphological closing operation is performed on each initial connected region: First, the initial connected region is dilated to connect small broken or gap regions caused by incomplete segmentation. Then, an erosion operation is performed on the dilated connected region to restore the overall outline of the connected region while maintaining connectivity, thereby improving the continuity of the vascular region and obtaining the final connected region. Furthermore, the non-vascular voxel regions completely surrounded by vascular voxels within the aforementioned initial connected regions can be filled to eliminate the voids inside the connected regions, thereby obtaining morphologically continuous connected regions, which are then used as the final connected regions.

[0053] S220. Based on the skeletonization algorithm, extract multiple skeleton points located on the central axis of the blood vessel region, and calculate the distance from each skeleton point to the corresponding blood vessel wall, using it as the local radius of the corresponding skeleton point.

[0054] In an optional embodiment of this application, step S220 includes the following process: based on the skeletonization algorithm, in each cross-section of the blood vessel region, the distance from each voxel in the cross-section of the blood vessel to the blood vessel wall is calculated, the voxel with the largest distance is selected as the skeleton point of the corresponding cross-section of the blood vessel, and the distance from each skeleton point to the blood vessel wall is used as the local radius of the corresponding skeleton point.

[0055] Specifically, for the binary mask of the vascular region ,For example =1 indicates a voxel It belongs to the vascular area. =0 indicates a voxel Not belonging to the vascular region. A voxel in the vascular region that is adjacent to at least one non-vascular voxel is defined as a vascular boundary voxel, and the set of all vascular boundary voxels is denoted as . In discrete voxel space, based on the following formula (6), the set of voxels from any voxel x within the vascular region to the vascular boundary is calculated. shortest distance : (6) in, voxel x to the set of voxels at the blood vessel boundary The shortest distance can be considered as the distance from voxel x to the vessel wall, used to characterize the local maximum inscribed sphere radius of the vessel at the location of voxel x. Based on the calculated shortest distance, the vessel region is skeletonized. Voxels with the largest distance to the vessel wall are selected as skeleton points in each cross-sectional direction of the vessel, thus obtaining a skeleton point set of one voxel width composed of multiple skeleton points. In this context, each voxel in the skeleton set is a skeleton point, and the distance from each skeleton point to the blood vessel wall is also considered. This serves as the local radius of the skeleton point. Since the skeleton point is located near the geometric center of the vessel cross-section, the set of skeleton points can be used as an approximate representation of the vessel centerline.

[0056] S230. Based on the connection relationship of each skeleton point in the vascular region, connect each skeleton point carrying a local radius to form a vascular centerline network.

[0057] Based on the adjacency and connectivity relationships of each skeleton point within the vascular region, the skeleton points are connected in order of their positions to construct a vascular centerline network. This network consists of multiple skeleton points and their connections. Each skeleton point not only represents its specific position on the vascular centerline but also carries local radius information at that position. Therefore, the vascular centerline network can represent both the overall spatial orientation and bifurcation of the blood vessel and reflect the local diameter changes along the centerline.

[0058] S300. Identify endpoints and bifurcation points from the vascular centerline network, and divide the vascular centerline network into several continuous and unbifuzzy vascular segments accordingly, and construct a vascular centerline diagram structure based on each of the vascular segments.

[0059] In an optional embodiment of this application, step S300 includes the following process: identifying endpoints and bifurcation points from the vascular centerline network, and dividing the vascular centerline network into several continuous and unbifuzzy vascular segments accordingly; using the connection points at both ends of each vascular segment as nodes, and using each vascular segment as an edge connecting the nodes to form a vascular centerline diagram structure; wherein each edge of the vascular centerline diagram structure carries the coordinates of the corresponding skeleton point and the local radius.

[0060] In the 26-neighborhood connectivity of 3D voxel space, for any skeleton point, its status as an endpoint or bifurcation point is determined based on the number of its neighboring skeleton points as follows: when a skeleton point is connected to only one neighboring skeleton point, it is identified as an endpoint; when a skeleton point is connected to three or more neighboring skeleton points, it is identified as a bifurcation point. After identifying endpoints and bifurcation points, the vascular centerline network is divided into several continuous and unbifuzzy vascular segments using the endpoints and bifurcation points as cutting points. Using the endpoints and bifurcation points as nodes and the vascular segments located between adjacent nodes as connecting edges, a vascular centerline graph structure is constructed, denoted as […]. In this system, the node set V represents the endpoints and branching points in the vascular network, and the edge set E represents the continuous vascular segments. For each vascular segment, its constituent skeleton points are recorded according to the spatial orientation of the vessel, forming a corresponding skeleton point sequence. ,in, The number of skeletal points in vascular segment e, and each skeletal point This process involves creating a network of nodes and their corresponding local radii, which describes the spatial geometry of the corresponding vascular segment. Through this process, the original vascular centerline network is transformed into a vascular topology containing nodes, vascular segments, and their geometric parameters.

[0061] Furthermore, to reduce radius fluctuations caused by local noise or segmentation glitch, the aforementioned local radii can be smoothed in one dimension to obtain the final local radius. One-dimensional smoothing methods include, but are not limited to, moving average, Savitzky-Golay filtering, or spline fitting.

[0062] S400. Establish a blood flow analysis model based on the vascular centerline diagram structure and solve it to obtain the blood flow analysis results of the vascular network; wherein, the blood flow analysis results include the blood flow rate of each vascular segment and / or the pressure difference between the two ends.

[0063] In an optional embodiment of this application, step S400 includes steps S410 to S440: For each of the blood vessel segments that serve as edges, steps S410 to S430 are performed: S410, based on the coordinates of each skeleton point in the blood vessel segment, the distance from each skeleton point to the first skeleton point in the blood vessel segment is determined, and this distance is used as the arc length of each skeleton point.

[0064] For each blood vessel segment that serves as an edge, let the corresponding skeleton points be arranged sequentially according to the direction of the blood vessel to form a skeleton point sequence. ,in, Let K be the coordinates of the k-th skeleton point in the skeleton point sequence. Let be the number of skeleton points in vascular segment e. Based on the skeleton point sequence, the spatial distance between adjacent skeleton points is accumulated segment by segment according to the connection order between the skeleton points to determine the cumulative path length of each skeleton point relative to the first skeleton point of the vascular segment, and the cumulative path length is used as the arc length of the corresponding skeleton point, as shown in formula (7): (7) in, Let be the arc length corresponding to the k-th skeleton point. For Euclidean distance, when k = 1, the arc length corresponding to the first skeleton point. The value is 0. Using the above method, each skeletal point in a blood vessel segment can be mapped to the arc length along the blood vessel's centerline, thus achieving a parametric description of the blood vessel segment.

[0065] S420 calculates the equivalent radius of the blood vessel segment by weighting the local radii corresponding to each skeleton point in the blood vessel segment according to the arc length of each skeleton point.

[0066] For each blood vessel segment that serves as an edge, its skeleton point sequence is denoted as... The arc length sequence formed by the arc lengths corresponding to each skeleton point is denoted as... The arc length corresponding to the last skeleton point is... The total length of vascular segment e Let the local radius of the k-th skeleton point be denoted as It can be obtained from the shortest distance from the skeleton point to the vessel wall using the above formula (6). In order to convert the local radius that varies along the vessel segment into a single scalar radius, the local radius can be weighted and averaged based on the arc length direction to obtain the equivalent radius of vessel segment e as shown in formula (8). : (8) in, The total length of vessel segment e is... For the position of blood vessel segment e at the arc length The local radius at point k, i.e., the local radius corresponding to the k-th skeleton point. This represents the arc length increment between the k-th skeleton point and the (k-1)-th skeleton point. = In this way, the local radius that varies along the blood vessel segment can be equivalent to a single equivalent radius, which can be used to characterize the overall geometric scale features of the blood vessel segment.

[0067] S430. Based on the equivalent radius of the blood vessel segment and the arc length of the last skeleton point, the hydraulic resistance of the blood vessel segment is obtained.

[0068] For each vessel segment e, its equivalent radius can be obtained through the above process. and the total length of the vessel segment ,in The total length of the vessel segment e can be calculated using the formula (7) above. In this embodiment, the vessel segment e can be equivalent to a circular pipe with a constant radius, and the hydraulic resistance can be calculated based on Poiseuille's law, as shown in formula (9): (9) in, The hydraulic resistance of segment e of the blood vessel. For the preset blood viscosity, The equivalent radius of vascular segment e can be obtained using the above formula (8). This is represented by pi. By using the above method, the local geometric information varying along the blood vessel segment is converted into the corresponding radius and arc length, which can be used to quickly calculate the hydraulic resistance of the blood vessel segment.

[0069] In another embodiment of this application, in order to more accurately characterize the influence of the changes in the local radii of each segment of the blood vessel on the hydraulic resistance, the hydraulic resistance can be obtained by integrating the local radii along the arc length direction of the blood vessel segment based on Poiseuille's law, as shown in formula (10): (10) in, Let be the local radius function of blood vessel segment e at arc length position s. This can be obtained by linear interpolation of various local radii. As an example, for any blood vessel segment e, within its corresponding arc length interval […]. , [Inside (where k)] The result can be obtained by linear interpolation as shown in formula (11). : = ( (11) in, and The positions of the arc lengths are respectively and The local radius corresponding to the skeleton point. It should be noted that, depending on the different accuracy requirements, the technicians may choose the approximate calculation method based on the equivalent radius shown in the above formula (9), or the integral calculation method based on the local radius distribution shown in the above formula (10), to obtain the hydraulic resistance of the blood vessel segment. No specific limitation is made.

[0070] S440. In the vascular centerline diagram structure, based on the hydraulic resistance of each vascular segment, a blood flow analysis model is established and solved according to preset constraints to obtain the blood flow rate and / or pressure difference between the two ends of each vascular segment, which is used as the blood flow analysis result.

[0071] In an optional embodiment of this application, step S440 includes steps S441 to S443: S441. In the vascular centerline diagram structure, a blood flow analysis model is established based on the hydraulic resistance of each vascular segment and according to preset constraints.

[0072] In the vascular centerline diagram structure, each vascular segment is considered as a pipe connecting its two ends, and the hydraulic resistance of the corresponding vascular segment is used as a parameter for analysis. Specifically, according to preset inlet and outlet constraints, and based on the mapping relationship between pressure and blood flow at the two ends of the vascular segment, a blood flow analysis model is constructed based on the vascular centerline diagram structure to obtain the blood flow and corresponding pressure distribution of each vascular segment in the vascular network.

[0073] In an optional embodiment of this application, step S441 includes the following process: for each blood vessel segment serving as an edge in the blood vessel centerline diagram structure: based on the hydraulic resistance of the blood vessel segment, determine the mapping relationship between the pressure difference between the two ends of the blood vessel segment and the blood flow. ;in, The blood flow rate of the aforementioned blood vessel segment e. The hydraulic resistance of the blood vessel segment e is... The pressure at the inlet node of the blood vessel segment e. The pressure at the outlet node of the blood vessel segment e is determined; based on the structure of the blood vessel centerline diagram, the network inlet node and network outlet node of the blood vessel network are determined, and preset inlet constraints and outlet constraints are set at the network inlet node and the network outlet node respectively; based on the mapping relationship between the pressure difference at both ends of each blood vessel segment and blood flow, the inlet constraints and the outlet constraints, a blood flow analysis model is established.

[0074] In the vascular centerline diagram structure, the vascular network is abstracted into a one-dimensional blood flow network model. Nodes in the vascular centerline diagram structure represent the connection or bifurcation points of blood vessels, and each vascular segment serves as an edge connecting the nodes. For any vascular segment e=(i,j) as an edge, i and j are the inlet and outlet nodes of vascular segment e, respectively. The inlet node is the node corresponding to the end where blood flows into the vascular segment, and the outlet node is the node corresponding to the end where blood flows out of the vascular segment. This vascular segment is considered a one-dimensional pipe, and the hydraulic resistance calculated above is introduced. Based on Poiseuille's law, according to the hydraulic resistance of the vascular segment, the blood flow through this vascular segment and the node pressure at both ends of the vascular segment satisfy the linear mapping relationship shown in formula (12): (12) in, This represents the blood flow rate through segment e of the blood vessel. Let be the pressure of blood vessel segment e at the inlet node i. Let e ​​be the pressure at the outlet node j of blood vessel segment e. Let be the hydraulic resistance of vascular segment e. Based on the vascular centerline diagram, the network entry and exit nodes of the entire vascular network can be identified, and entry and exit constraints can be set respectively. The network entry node is located upstream of the entire vascular network, connected to an external blood supply system, and inputting blood flow into the network. The network exit node is located downstream of the entire vascular network, connected to an external vascular bed or microcirculation system, and outflowing blood from the network. For example, when performing blood flow analysis for cerebral arteriovenous malformations, the network entry node can be the arterial end node supplying blood to the malformation area, and the network exit node can be the venous end node outflowing blood from the malformation area. Entry constraints are pre-set boundary conditions at the network entry node, used to limit the blood flow state into the entire vascular network. Exit constraints are pre-set boundary conditions at the network exit node, used to limit the pressure or hydraulic resistance when blood flows out of the vascular network. For example, an entry pressure and entry blood flow rate can be pre-set at the network entry node, and an exit pressure can be pre-set at the network exit node. Based on the mapping relationship between blood flow and pressure in each blood vessel segment in formula (12) and the constraints at the network entry and exit nodes, a blood flow analysis model can be constructed. The blood flow analysis model uses the pressure at each node and the blood flow in the corresponding blood vessel segment as unknowns. Through the mapping relationship between pressure and blood flow within the blood vessel segment and the boundary constraints of the entire blood vessel network, it is used to characterize the blood flow distribution and pressure transmission of the blood vessel network under given constraints.

[0075] Further, in an optional embodiment, the ingress constraint is a preset pressure and blood flow of the network node, and the egress constraint is... ,in, The preset reference pressure, The preset export resistance parameters, The pressure of the network egress node to be solved. The blood flow rate of the blood vessel segment corresponding to the network exit node to be solved is denoted as .

[0076] Inlet constraints are used to characterize the relationship between the vascular network and external blood sources. Specifically, at the network inlet node, the pressure value and blood flow into that node can be preset as known conditions to simulate cardiac output or upstream blood supply. By specifying inlet constraints at the inlet node, specific input information can be provided to the blood flow analysis model. At the network outlet node, to more realistically reflect the coupling relationship between the vascular distality and the downstream microcirculation or peripheral vascular bed, the pressure and blood flow at the outlet node can be set to satisfy... ,in, and Given quantities and Let be the quantity to be solved.

[0077] S442. Solve the blood flow analysis model to obtain the pressure of each node in the vascular centerline diagram structure and the blood flow of the corresponding vascular segment.

[0078] The pressure of all nodes in the vascular centerline diagram structure is taken as an unknown quantity, and the blood flow of each vascular segment is uniformly expressed as a function of the pressure of its two ends, as shown in the above formula (12). For each internal node v except for the network entry node and the network exit node, the node constraint relationship is established according to the principle of blood flow conservation, that is, in each vascular segment connected to the same node, the sum of the blood flow flowing into the node is equal to the sum of the blood flow flowing out of the node, as shown in formula (13): (13) in, This represents the blood flow rate through segment e of the blood vessel. This represents the set of blood vessel segments connected to node v and flowing into that node. This represents the set of blood vessel segments connected to and flowing out of node v. It is understood that the inflow and outflow directions mentioned above can be automatically determined after the blood flow analysis model is solved based on the pressure relationship between the nodes at both ends of the blood vessel segment. Combining the pressure or blood flow constraints given at the inlet node and the constraints set at the outlet node based on the reference pressure and outlet resistance parameters, the above constraints, the mapping relationship between blood flow and pressure shown in formula (12), and the blood flow conservation equation shown in formula (13) can form a set of equations about the pressure of each node, thus forming the blood flow analysis model corresponding to the blood vessel centerline diagram structure. After establishing the set of equations for each node, the set of equations can be solved based on the known quantities given in the constraints to obtain the pressure values ​​corresponding to each node in the blood vessel centerline diagram structure. After obtaining the node pressure, the blood flow in each blood vessel segment can be calculated according to formula (12) based on the known hydraulic resistance of the blood vessel segment and the pressure difference between the nodes at both ends of the blood vessel segment. Through the above solution process, the node pressure distribution of the vascular network under given boundary conditions and the distribution of blood flow in each vascular segment can be obtained, thereby characterizing the hemodynamic state corresponding to the vascular centerline diagram structure.

[0079] S443. The blood flow rate of each blood vessel segment and / or the pressure difference between the two ends of each blood vessel segment calculated from the pressure of each node are used as the blood flow analysis results of each blood vessel segment.

[0080] Based on the pressure at each node, the pressure difference between the two ends of the corresponding blood vessel segment can be calculated, and this pressure difference is used as the blood flow analysis result for the vascular network. Alternatively, the blood flow rate of each blood vessel segment can also be used as the blood flow analysis result. Furthermore, the pressure difference between the two ends of each blood vessel segment and the blood flow rate can be combined as the blood flow analysis result. The specific method is not limited. Specifically, suppose there are N nodes in the vascular centerline diagram structure, and the node pressure is denoted as... , ,…, By combining the blood flow conservation equations of the internal nodes with the constraints at the network entry and exit nodes, the pressure of each node is solved. After the pressure of the nodes is solved, the blood flow of each blood vessel segment and the pressure difference between the two ends are calculated based on the pressure of the nodes.

[0081] Furthermore, at the bifurcation point, in addition to satisfying the blood flow continuity constraint, the pressure relationship at the bifurcation point can also be constrained. In one embodiment, it can be assumed that the node pressure of each connected vessel segment at the bifurcation point is continuous, that is, the parent branch segment and each daughter branch segment have the same node pressure at the bifurcation point. In another embodiment, in order to more realistically reflect the local energy loss of blood flow at the bifurcation point, a local loss term is introduced at the bifurcation point, so that the pressure relationship between the parent branch segment and the daughter branch segment at the node is as shown in formula (14): (14) in, This is the preset local loss coefficient at the bifurcation node. For the preset blood density, This represents the node pressure of the parent branch vessel segment at the bifurcation node v. The node pressure at the bifurcation node v of the i-th daughter branch segment connected to the parent branch segment. The blood flow rate through the i-th sub-branch segment is... Let be the local cross-sectional area of ​​the i-th subbranch segment at the bifurcation node v, obtained as follows. Let the local radius of segment e at arc length position s be denoted as . ,in, The set of voxels at the blood vessel boundary can be derived from the skeletal point corresponding to that location. shortest distance Therefore, the shortest distance can be determined using the above formula (6). The arc length along the centerline of the blood vessel. can be and recursive relations Calculated. When the above arc length When corresponding to the bifurcation node v, the local cross-sectional area at that location can be calculated using formula (15): (15) By combining the pressure difference between the two ends of the above-mentioned blood vessel segment with the flow rate, the blood flow continuity constraint at the node, and the constraint conditions, a complete blood flow analysis model is established. The blood flow analysis model is then solved to obtain the distribution information of blood flow between different blood vessel segments and / or the pressure difference between the two ends of each blood vessel node in the blood vessel centerline diagram structure.

[0082] Furthermore, this application can also determine the proportion of blood flow splitting to each blood vessel segment at each bifurcation point based on the blood flow of each blood vessel segment in the vascular network, and generate blood flow distribution results accordingly.

[0083] It should be noted that since the establishment and solution of the blood flow analysis model depend on vascular geometric parameters (such as the structure of the vascular centerline map, the length of each vascular segment, the local radius of each skeleton point, the equivalent radius of the vascular segment, and the hydraulic resistance obtained therefrom), the accuracy of the vascular image volume data in representing the vascular morphology directly affects the reliability of the blood flow calculation results. In view of the above factors, in this application, vascular image volume data acquired in ultra-high resolution mode is preferably used to provide clearer vascular boundaries and more complete display of small vessels, thereby improving the boundary accuracy and connectivity integrity of vascular segmentation and reducing missed segmentation, breakage, and boundary offset. This makes the generated vascular centerline network more accurate, thereby improving the stability of endpoint and bifurcation point identification, and thus improving the accuracy of vascular segment division and vascular centerline map structure construction. Furthermore, the vascular image volume data obtained in ultra-high resolution mode also helps to more accurately estimate the local radius and the equivalent radius of the vascular segment, thereby reducing the propagation of geometric errors in hydraulic resistance calculation. Therefore, the established blood flow analysis model is more reliable in terms of input geometric parameters, which helps to reduce the calculation deviation of blood flow and pressure difference caused by geometric modeling errors. Based on the more refined vascular morphology characterization in the ultra-high resolution mode, more reliable geometric modeling input is obtained, which helps to improve the accuracy and reliability of blood flow analysis results.

[0084] like Figure 2 As shown, it displays the vascular network corresponding to three-dimensional vascular image volume data. By performing skeletonization processing on this volume data, the spatial main trunk direction is extracted along the center of the vessel lumen, thus obtaining the centerline result that only reflects the topological structure and spatial connectivity of the vessel. Figure 3 The diagram shows the centerline of the blood vessels. Further, Figure 4 This is a diagram showing the intracranial vascular network and its corresponding vascular regions. The blue areas are vascular regions segmented from vascular imaging data, and the gray-white areas are... Figure 2The 3D vascular image volume data shown presents a vascular network. Planar bounding boxes are used to represent the spatial reference position or cropping range of the vascular network in the 3D volume data. The vascular network is displayed in comparison with vascular regions in mask form to demonstrate the correspondence between vascular regions and vascular networks.

[0085] In an optional embodiment of the present invention, the method further includes: calculating the blood flow velocity of each of the blood vessel segments based on the blood flow rate and equivalent radius of each segment, and visually displaying the blood flow velocity in the vascular network. For example, as... Figure 5 As shown, it demonstrates Figure 2 This is a visualization combining vascular network and blood flow analysis results. The vascular network serves as the carrier, onto which blood flow velocity information is overlaid, and the magnitude of the blood flow velocity is encoded using color mapping (from blue to red indicates a gradual increase in blood flow velocity). The maximum blood flow velocity is shown at the location corresponding to the center line of the blood vessel. Based on blood flow in the vessel segment and equivalent radius It is calculated as follows: = The color bars in the diagram represent the range of blood flow velocities, thus visually reflecting the non-uniformity of blood flow distribution within different vascular segments and the location of high-velocity regions. This is used to analyze the hemodynamic characteristics and potential abnormal areas in the vascular network. For example, when the blood flow velocity exceeds a preset threshold, it indicates that the vascular segment is experiencing greater shear stress, and its risk of rupture increases accordingly. It is understandable that... Figure 4 and Figure 5 The colors shown are for illustrative purposes only; specific colors can be customized and are not limited here.

[0086] like Figure 6 As shown, the medical image processing system includes: a volume data acquisition module 610, a centralization module 620, a graph construction module 630, and a blood flow analysis module 640. The volume data acquisition module 610 acquires vascular image volume data from photon-counting computed tomography scans of a target region. The centralization module 620 performs skeletonization processing on the vascular network based on the vascular image volume data, extracting multiple skeleton points from the vascular network and forming a vascular centerline network according to the connection relationships between the skeleton points. The graph construction module 630 identifies endpoints and bifurcation points from the vascular centerline network, dividing the vascular centerline network into several continuous and unbifurated vascular segments, and constructs a vascular centerline graph structure based on each vascular segment. The blood flow analysis module 640 establishes and solves a blood flow analysis model based on the vascular centerline graph structure to obtain the blood flow analysis results of the vascular network; wherein the blood flow analysis results include the blood flow rate and / or pressure difference between the two ends of each vascular segment.

[0087] Specific limitations regarding the medical image processing system can be found in the limitations on medical image processing methods described above, and will not be repeated here. Each module in the aforementioned medical image processing system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of the computer device in hardware format or independent of it, or stored in the memory of the computer device in software format, so that the processor can call the corresponding operations of each module.

[0088] It should be noted that, in order to highlight the innovative aspects of this application, this embodiment does not include modules that are not closely related to solving the technical problems proposed in this application, but this does not mean that there are no other modules in this embodiment.

[0089] This application also provides a photon-counting CT scanner, comprising: a data acquisition module for performing photon-counting computed tomography on a target area to obtain vascular image volume data of the target area; and a processing module for obtaining a corresponding vascular centerline map structure based on the vascular image volume data, and establishing and solving a blood flow analysis model based on the vascular centerline map structure and pre-acquired vascular segment parameters to obtain the blood flow analysis results of the vascular network. It should be noted that the processing module of this application can be implemented based on the computing unit in existing medical imaging equipment, or integrated into the equipment as an independent image processing unit; the specific implementation method is not limited. As long as the processing module possesses the functionality of this application, that is, based on existing image processing functions, further utilizing the vascular image volume data for skeletonization processing to establish and solve a blood flow analysis model to obtain the blood flow analysis results of the vascular network, the photon-counting CT scanner with the above-mentioned processing module can be considered to fall within the protection scope of this application.

[0090] like Figure 7 As shown, the electronic device 7 may include a memory 71, a processor 72 and a bus, and may also include a computer program, such as a medical image processing program, stored in the memory 71 and capable of running on the processor 72.

[0091] The memory 71 includes at least one type of readable storage medium, including flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 71 can be an internal storage unit of the electronic device 7, such as a portable hard drive. In other embodiments, the memory 71 can be an external storage device of the electronic device 7, such as a plug-in portable hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 7. Furthermore, the memory 71 can include both internal and external storage units of the electronic device 7. The memory 71 can be used not only to store application software and various types of data installed on the electronic device 7, such as code for medical image processing, but also to temporarily store data that has been output or will be output.

[0092] In some embodiments, the processor 72 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits packaged with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 72 is the control unit of the electronic device 7, connecting various components of the entire electronic device 7 via various interfaces and lines. It executes programs or modules (such as medical image processing programs) stored in the memory 71, and calls data stored in the memory 71 to perform various functions and process data in the electronic device 7.

[0093] The processor 72 executes the operating system of the electronic device 7 and various installed application programs. The processor 72 executes the application programs to implement the steps in the above-described medical image processing method.

[0094] For example, a computer program can be divided into one or more modules, one or more of which are stored in memory 71 and executed by processor 72 to complete this application. One or more modules can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in electronic device 7. For example, the computer program can be divided into a volume data acquisition module 610, a centralization module 620, a graph construction module 630, and a blood flow analysis module 640.

[0095] The integrated unit implemented as a software functional module described above can be stored in a computer-readable storage medium, which can be non-volatile or volatile. The software functional module stored in the storage medium includes several instructions to cause a computer device (which may be a personal computer, computer equipment, or network device, etc.) or processor to execute some functions of the medical image processing methods of the various embodiments of this application.

[0096] In summary, this application utilizes ultra-high resolution vascular image volume data from photon-counting computed tomography (CT) for vascular segmentation, effectively reducing segmentation instability caused by unclear boundaries of small vessels. By adopting a centerline-first approach, the complex vascular network is abstracted into a centerline graph structure composed of a topological network and the parameters of the vessel segments, which better reflects the blood flow distribution and transmission characteristics at the vascular network scale. Based on this, a one-dimensional blood flow analysis model is used to solve the vascular network, significantly reducing computational complexity and cost compared to traditional three-dimensional computational fluid dynamics methods. This enables rapid assessment of vascular blood flow status and multi-scheme extrapolation analysis.

[0097] The present application has been shown and described in detail above with reference to the accompanying drawings and preferred embodiments. However, the present application is not limited to these disclosed embodiments, and other solutions derived by those skilled in the art are also within the scope of protection of the present application. The above embodiments are merely illustrative of the principles and effects of the present application and are not intended to limit the present application. Any person skilled in the art can modify or change the above embodiments without departing from the spirit and scope of the present application. Therefore, all equivalent modifications or changes made by those skilled in the art without departing from the spirit and technical concept disclosed in the present application should still be covered by the claims of the present application.

Claims

1. A medical image processing method, characterized in that, The medical image processing method includes: Acquire photon-counted computed tomography (CT) data of blood vessels in the target region; Based on the vascular image data, the vascular network is skeletonized, multiple skeleton points in the vascular network are extracted, and a vascular centerline network is formed according to the connection relationship between each skeleton point. Endpoints and bifurcation points are identified from the vascular centerline network, and the vascular centerline network is divided into several continuous and unbifuzzy vascular segments. A vascular centerline diagram structure is constructed based on each of the vascular segments. A blood flow analysis model is established and solved based on the aforementioned vascular centerline diagram structure to obtain the blood flow analysis results of the vascular network; wherein, the blood flow analysis results include the blood flow rate and / or pressure difference between the two ends of each vascular segment.

2. The medical image processing method according to claim 1, characterized in that, Based on the vascular image data, the steps of performing skeletonization processing on the vascular network, extracting multiple skeleton points from the vascular network, and forming a vascular centerline network according to the connection relationships between the skeleton points include: The vascular image data is input into the vascular segmentation model to extract the vascular region; Based on the skeletonization algorithm, multiple skeleton points located on the central axis of the vascular region are extracted, and the distance from each skeleton point to the corresponding vascular wall is calculated and used as the local radius of the corresponding skeleton point. Based on the connection relationship of each skeleton point in the vascular region, the skeleton points carrying local radii are connected to each other to form a vascular centerline network.

3. The medical image processing method according to claim 2, characterized in that, The steps of inputting the vascular image data into the vascular segmentation model and extracting the vascular region include: The vascular image volume data is input into the vascular segmentation model to segment the initial vascular region from the vascular image volume data; The initial vascular region is modified to obtain the final vascular region.

4. The medical image processing method according to claim 3, characterized in that, The vessel segmentation model is a three-dimensional vessel segmentation model including an encoder and a decoder. The step of inputting the vessel image volume data into the vessel segmentation model and segmenting the initial vessel region from the vessel image volume data includes: The vascular image volume data is input into the encoder of the three-dimensional vascular segmentation model, and the vascular image volume data is downsampled step by step to obtain vascular features at various scales. The vascular features at various scales are input into the decoder of the three-dimensional vascular segmentation model. By upsampling at each level, various decoded features at different scales are obtained. Each decoded feature is then fused with the corresponding vascular features across layers to obtain the initial vascular region. The encoder and the decoder have a corresponding scale relationship.

5. The medical image processing method according to claim 3, characterized in that, The steps for refining the initial vascular region to obtain the final vascular region include: Connectivity analysis was performed on the initial vascular region to identify multiple connected regions within the initial vascular region; Select the largest one from all connected regions as the final vascular region.

6. The medical image processing method according to claim 2, characterized in that, Based on the skeletonization algorithm, the steps of extracting multiple skeleton points located at the central axis of the vascular region and calculating the distance from each skeleton point to the corresponding vascular wall, and using this distance as the local radius of the corresponding skeleton point, include: Based on the skeletonization algorithm, the distance from each voxel in each cross-section of the blood vessel in the blood vessel region to the blood vessel wall is calculated. The voxel with the largest distance is selected as the skeleton point of the corresponding blood vessel cross-section, and the distance from each skeleton point to the blood vessel wall is used as the local radius of the corresponding skeleton point.

7. The medical image processing method according to claim 1, characterized in that, The steps of identifying endpoints and bifurcation points from the vascular centerline network, dividing the vascular centerline network into several continuous and unbifuzzing vascular segments, and constructing a vascular centerline map structure based on each of the vascular segments include: The endpoints and bifurcation points are identified from the vascular centerline network, and the vascular centerline network is divided into several continuous and unbifled vascular segments accordingly. The connection points at both ends of each of the aforementioned blood vessel segments are used as nodes, and the various blood vessel segments are used as edges connecting the nodes to form a blood vessel centerline diagram structure; wherein, each edge of the blood vessel centerline diagram structure carries the coordinates of the corresponding skeleton point and the local radius.

8. The medical image processing method according to claim 7, characterized in that, The steps for establishing and solving a blood flow analysis model based on the aforementioned vascular centerline diagram structure to obtain the blood flow analysis results of the vascular network include: For each of the blood vessel segments that serve as edges: Based on the coordinates of each skeleton point in the blood vessel segment, the distance from each skeleton point to the first skeleton point in the blood vessel segment is determined, and this distance is used as the arc length of each skeleton point. The equivalent radius of the blood vessel segment is obtained by weighting the local radii corresponding to each skeleton point in the blood vessel segment according to the arc length of each skeleton point. Based on the equivalent radius of the blood vessel segment and the arc length of the last skeleton point, the hydraulic resistance of the blood vessel segment is obtained; In the aforementioned vascular centerline diagram structure, based on the hydraulic resistance of each vascular segment, a blood flow analysis model is established and solved according to preset constraints to obtain the blood flow rate and / or pressure difference between the two ends of each vascular segment, which is then used as the blood flow analysis result.

9. The medical image processing method according to claim 8, characterized in that, In the aforementioned vascular centerline diagram structure, based on the hydraulic resistance of each vascular segment, a blood flow analysis model is established and solved according to preset constraints to obtain the blood flow rate and / or pressure difference between the two ends of each vascular segment, and the steps of using this as the blood flow analysis result include: In the aforementioned vascular centerline diagram structure, a blood flow analysis model is established based on the hydraulic resistance of each vascular segment and according to preset constraints. The blood flow analysis model is solved to obtain the pressure at each node in the vascular centerline diagram structure and the blood flow of the corresponding vascular segment; The blood flow rate of each blood vessel segment and / or the pressure difference between the two ends of each blood vessel segment calculated from the pressure at each node are used as the blood flow analysis results for each blood vessel segment.

10. The medical image processing method according to claim 9, characterized in that, In the aforementioned vascular centerline diagram structure, the steps for establishing a blood flow analysis model based on the hydraulic resistance of each vascular segment and according to preset constraints include: For each vessel segment that serves as an edge in the aforementioned vessel centerline diagram structure: based on the hydraulic resistance of the vessel segment, determine the mapping relationship between the pressure difference between the two ends of the vessel segment and the blood flow. ;in, The blood flow rate of the aforementioned blood vessel segment e. The hydraulic resistance of the blood vessel segment e is... The pressure at the inlet node of the blood vessel segment e. The pressure at the outlet node of the blood vessel segment e; Based on the vascular centerline diagram structure, the network entry node and network exit node of the vascular network are determined, and preset entry constraints and exit constraints are set at the network entry node and the network exit node respectively. A blood flow analysis model is established based on the mapping relationship between the pressure difference at both ends of each blood vessel segment and the blood flow, the inlet constraint conditions, and the outlet constraint conditions.

11. The medical image processing method according to claim 10, characterized in that, The ingress constraint is a preset pressure and blood flow of the network node, and the egress constraint is... ,in, The preset reference pressure, The preset export resistance parameters, The pressure of the network egress node to be solved. The blood flow rate of the blood vessel segment corresponding to the network exit node to be solved is given.

12. The medical image processing method according to claim 8, characterized in that, The method further includes: calculating the blood flow velocity of each of the blood vessel segments based on the blood flow rate and equivalent radius of each blood vessel segment, and visually displaying the blood flow velocity in the blood vessel network.

13. A medical image processing system, characterized in that, The system includes: The volume data acquisition module is used to acquire vascular image volume data from photon-counting computed tomography scans of the target area; The centralized module is used to perform skeletonization processing on the vascular network based on the vascular image volume data, extract multiple skeleton points in the vascular network, and form a vascular centerline network according to the connection relationship between each skeleton point. The graph construction module is used to identify endpoints and bifurcation points from the vascular centerline network, thereby dividing the vascular centerline network into several continuous and unbifled vascular segments, and constructing a vascular centerline graph structure based on each of the vascular segments. The blood flow analysis module is used to establish a blood flow analysis model based on the vascular centerline diagram structure and solve it to obtain the blood flow analysis results of the vascular network; wherein, the blood flow analysis results include the blood flow rate and / or pressure difference of each vascular segment.

Images