Anatomical structure three-dimensional reconstruction method, device and equipment applied to cardiovascular surgery

Abstract

The application provides an anatomic structure three-dimensional reconstruction method, device and equipment applied to cardiovascular surgery, obtains a two-dimensional angiography image sequence of a target cardiovascular region, and performs time-space synchronous correction to eliminate time-space errors caused by heartbeats and flow of contrast agents. Based on geometric deformation characteristics of the multi-angle corrected images, an initial three-dimensional blood vessel tree model is generated through space intersection matching and surface contour reconstruction. Preset hemodynamic parameters are combined with the model for joint optimization, the model is compared with simulated blood flow contrast agent distribution and measured data, local areas with abnormal blood flow in the model are iteratively corrected, and the geometric shape of the local areas is matched with the real blood flow path. Vessel center lines, surface contours and topological connection relationships of the corrected model are extracted to generate structured guide data used for surgery planning. The application integrates image correction, three-dimensional reconstruction and hemodynamic optimization, and improves anatomic and functional accuracy of a cardiovascular three-dimensional model.

Description

Technical Field

[0001] This application relates to the field of data processing, and in particular to a method, apparatus, and equipment for three-dimensional reconstruction of anatomical structures applied in cardiovascular surgery. Background Technology

[0002] In cardiovascular surgery, accurate three-dimensional reconstruction of the anatomical structure of the lesion area is a crucial prerequisite for complex surgical planning and simulation. Currently, the commonly used method in clinical practice is to acquire a sequence of two-dimensional angiographic images of the target cardiovascular region and generate a static anatomical model based on the vascular tree projection contour information presented in these images using a three-dimensional reconstruction algorithm. In actual clinical applications, three-dimensional models constructed solely based on the static morphological information presented by angiographic images often fail to fully reflect the dynamic geometric characteristics of blood vessels under real physiological conditions. For example, the compliance changes of the vessel lumen under blood flow impact and the deformation effects of blood flow paths on the vessel wall. This results in a certain deviation between the reconstructed three-dimensional model and the actual anatomical structure and physiological state within the patient, thus affecting the accuracy and reliability of the surgical planning scheme to some extent. Summary of the Invention

[0003] This invention provides a method, apparatus, and equipment for three-dimensional reconstruction of anatomical structures applied in cardiovascular surgery.

[0004] In a first aspect, embodiments of the present invention provide a method for three-dimensional reconstruction of anatomical structures applied in cardiovascular surgery, comprising: The raw image data sequence of the target cardiovascular region is obtained. The raw image data sequence consists of two-dimensional angiography image units acquired at consecutive time points. Each two-dimensional angiography image unit contains the vascular tree projection contour information and blood flow contrast agent distribution density information at the corresponding time point. Spatiotemporal synchronization correction is performed on the original image data sequence. Based on the propagation delay time parameter of blood flow contrast agent distribution density information between adjacent two-dimensional angiography image units, a corrected image data set with time alignment markers is generated. The spatial dimension of the corrected image data set is expanded, and an initial three-dimensional vascular tree model of the target cardiovascular region is generated based on the geometric deformation characteristics of the vascular tree projection contour information under multiple preset projection angles. The preset hemodynamic simulation parameters are jointly optimized with the initial three-dimensional vascular tree model. The spatial topology of the initial three-dimensional vascular tree model is corrected by using the distribution density information of blood flow contrast agent in the flow path distribution of the initial three-dimensional vascular tree model, resulting in a corrected three-dimensional vascular tree model. Cardiovascular surgical planning guidance data is generated based on the modified 3D vascular tree model, which includes a sequence of vascular lumen surface contour coordinates and descriptors of vascular branch connection relationships.

[0005] Secondly, embodiments of the present invention provide a three-dimensional reconstruction apparatus, comprising: The data acquisition module is used to acquire the original image data sequence of the target cardiovascular region. The original image data sequence consists of two-dimensional angiography image units acquired at consecutive time points. Each two-dimensional angiography image unit contains the vascular tree projection contour information and blood flow contrast agent distribution density information at the corresponding time point. The data correction module is used to perform spatiotemporal synchronization correction on the original image data sequence. Based on the propagation delay time parameter of the blood flow contrast agent distribution density information between adjacent two-dimensional angiography image units, it generates a corrected image data set with time alignment markers. The model generation module is used to expand the spatial dimension of the corrected image data set and generate an initial three-dimensional vascular tree model of the target cardiovascular region based on the geometric deformation characteristics of the vascular tree projection contour information under multiple preset projection angles. The joint optimization module is used to jointly optimize the preset hemodynamic simulation parameters with the initial three-dimensional vascular tree model. By using the blood flow contrast agent distribution density information to determine the flow path distribution in the initial three-dimensional vascular tree model, the spatial topology of the initial three-dimensional vascular tree model is corrected to obtain the corrected three-dimensional vascular tree model. The data generation module is used to generate cardiovascular surgical planning guidance data based on the modified three-dimensional vascular tree model, which includes a sequence of vascular lumen surface contour coordinates and a descriptor of vascular branch connection relationships.

[0006] Thirdly, embodiments of this application provide a computer device, including a processor and a memory, wherein a computer program is stored in the memory, and the computer program is loaded and executed by the processor to implement the three-dimensional reconstruction method of anatomical structures applied to cardiovascular surgery as described above.

[0007] The embodiments of this application have the following beneficial effects: This invention first performs spatiotemporal synchronization correction on the acquired raw image data sequence, determining the time alignment benchmark based on the actual propagation delay of the contrast agent between adjacent image units. This effectively eliminates the temporal misalignment of image units caused by cardiac pulsation or respiratory movements, ensuring that the image data used for subsequent processing has a consistent physiological phase in time. Based on this, a multi-plane reconstruction model is invoked to expand the spatial dimension of the corrected image data set. An initial three-dimensional vascular tree model is generated based on the geometric deformation characteristics of the vascular tree projection contour at different projection angles, achieving preliminary reconstruction from two-dimensional angiographic images to three-dimensional anatomical structures. Subsequently, preset hemodynamic simulation parameters are introduced as independent constraints, along with the initial three-dimensional vascular tree model. By performing joint optimization processing and comparing the flow path distribution of blood flow contrast agent density information in the model with the actual situation recorded in the images, the spatial topology of the model is specifically corrected. This process ensures that the final three-dimensional vascular tree model not only accurately restores the static anatomical morphology of blood vessels but also incorporates dynamic physiological information reflecting the actual blood flow state. As a result, it highly matches the actual patient situation in terms of geometric accuracy and hemodynamic characteristics. Based on the corrected model, the surgical planning guidance data generated includes the contour coordinate sequence of the vascular lumen surface and the descriptor of the connection relationship of vascular branches. This data can provide reliable three-dimensional navigation basis for cardiovascular surgery that has both fine anatomical structure and conforms to individual blood flow characteristics, effectively improving the accuracy and safety of surgical planning. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the application environment provided in the embodiments of this application; Figure 2 This is a flowchart illustrating the three-dimensional reconstruction method for anatomical structures applied in cardiovascular surgery provided in the embodiments of this application; Figure 3 This is a block diagram of the three-dimensional reconstruction apparatus provided in the embodiments of this application.

[0009] Figure 4 This is a structural block diagram of the computer device provided in the embodiments of this application. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0011] In some embodiments, the three-dimensional reconstruction method for anatomical structures in cardiovascular surgery provided in this application is applied to, for example... Figure 1 The application environment shown. For example, as... Figure 1As shown, the application environment includes an angiography device 10 (e.g., a digital subtraction angiography device) and a computer device 20. The computer device 20 can be a server or a computer system. Data transmission between the angiography device 10 and the computer device 20 occurs via a network.

[0012] One point that needs to be clarified is that the above Figure 2 The descriptions provided are merely exemplary and illustrative. In exemplary embodiments, the functions of the imaging device 10 and the computer device 20 can be flexibly configured and adjusted, and the embodiments of this application do not limit this.

[0013] Please refer to Figure 2 This document illustrates a flowchart of a three-dimensional reconstruction method for anatomical structures applied in cardiovascular surgery, according to an embodiment of this application. The steps in this method can be performed by the methods described above. Figure 1 The method is executed by computer device 20. The method may include the following steps: Step S100: Obtain the original image data sequence of the target cardiovascular region. The original image data sequence consists of two-dimensional angiography image units acquired at consecutive time points. Each two-dimensional angiography image unit contains the vascular tree projection contour information and blood flow contrast agent distribution density information at the corresponding time point.

[0014] The target cardiovascular region is a specific cardiovascular anatomical location requiring 3D reconstruction, such as the left anterior descending artery, the circumflex artery, and their branches, or the aortic root and aortic arch. The raw image data sequence is a chronologically ordered collection of images, composed of consecutive 2D angiographic image units acquired at different time points. Each 2D angiographic image unit is a single frame of 2D image acquired at a specific moment. These image units, arranged chronologically, collectively depict the dynamic process of contrast agent flow and diffusion within the target cardiovascular region over time. The vascular tree projection contour information refers to the boundary line of the high-density image formed along the X-ray penetration path of the vascular lumen in each frame of the 2D angiographic image unit due to the iodine-containing contrast agent filling the lumen. This boundary line delineates the projected morphology of the vascular tree on the 2D plane, specifically including the vascular pathway, changes in lumen diameter, confluence points of vascular branches, and the curvature of the vessels. The contrast agent distribution density information refers to the relative concentration of contrast agent at each pixel location in each frame of the 2D angiographic image unit. This concentration information is typically represented in images by pixel grayscale values. Areas with high contrast agent concentration, such as the main flow in the center of a blood vessel, have higher grayscale values; areas with low contrast agent concentration, such as the edges of a blood vessel or the ends of its branches, have lower grayscale values. This density information can reflect the flow state of blood within the blood vessel, such as the position of the contrast agent leading wave, the speed of blood flow, and the presence of eddies or stagnation areas.

[0015] Step S200: Perform spatiotemporal synchronization correction on the original image data sequence. Based on the propagation delay time parameter of blood flow contrast agent distribution density information between adjacent two-dimensional angiography image units, generate a corrected image data set with time alignment markers.

[0016] In this embodiment, step S200 specifically includes steps S210 to S260: Step S210: Extract the acquisition timestamp corresponding to each two-dimensional angiography image unit in the original image data sequence, determine the time interval distribution characteristics between adjacent image units according to the order of acquisition timestamps, and store the time interval distribution characteristics as a time interval sequence.

[0017] Specifically, from the DICOM file generated by the DSA system, the metadata tag of each image unit is parsed, such as the tag, which records the time offset of the image relative to the start of the sequence, in milliseconds. For a raw image data sequence consisting of N image units, a list containing N timestamps can be obtained. Then, starting from the second image unit, the timestamp of the previous image unit is subtracted from the timestamp of the current image unit to calculate N-1 time intervals. For example, subtracting the timestamp of the first image unit from the timestamp of the second image unit gives the first time interval, subtracting the timestamp of the second image unit from the timestamp of the third image unit gives the second time interval, and so on. These calculated time interval values, arranged in the order of the corresponding image units, constitute the time interval sequence. This sequence is stored as a one-dimensional array, and the correspondence between the array index and adjacent image units is recorded; for example, the index corresponds to the interval between the first and second image units.

[0018] Step S220: Perform correlation analysis on the blood flow contrast agent distribution density information of adjacent two-dimensional angiography image units, determine the propagation delay time parameter of blood flow contrast agent between adjacent time points based on the correlation analysis results, and associate and store the propagation delay time parameter with the corresponding adjacent image units.

[0019] In this embodiment, step S220 specifically includes steps S221 to S226: Step S221: Extract the subset of blood flow contrast agent distribution density corresponding to the vascular tree projection contour region in adjacent two-dimensional angiography image units to form a first density distribution sequence and a second density distribution sequence. The first density distribution sequence corresponds to the previous image unit, and the second density distribution sequence corresponds to the next image unit.

[0020] For two adjacent two-dimensional angiography image units, a region-growing-based segmentation algorithm is used to extract the vascular tree projection contour region. Taking the previous image unit as an example, a seed point located within the blood vessel is manually or automatically selected in the image. Then, a grayscale threshold range is set, and all pixels connected to the seed point and whose grayscale values ​​are within the threshold range are classified as vascular regions, thus obtaining a binary vascular mask. The same operation is applied to the next image unit to obtain its vascular mask. Due to the influence of heartbeat, the vascular masks of the two images may have overall displacement. Therefore, it is necessary to further align the vascular mask of the next image unit with the vascular mask of the previous image unit using image registration techniques, such as rigid registration based on mutual information. After alignment, sampling points are set along the vascular centerline from the vascular inlet to the distal branch in both image units with a fixed step size. At each sampling point, the grayscale value in the previous image unit is used as an element of the first density distribution sequence, and the grayscale value at the corresponding registration position in the next image unit is used as an element of the second density distribution sequence. Based on this, we obtained two density distribution sequences of equal length and one-to-one spatial location.

[0021] Step S222: Perform cross-correlation operation on the first density distribution sequence and the second density distribution sequence to generate a cross-correlation function curve. The cross-correlation function curve represents the similarity at different time offsets, and record the sequence of ordinate values ​​of the cross-correlation function curve.

[0022] Assume the first density distribution sequence is A, and the second density distribution sequence is B, both with length M. Define a maximum possible displacement range, for example, from [from [from M] to [from M]]. For each displacement d within this range, shift sequence B forward or backward by d positions, padding the portion exceeding the original sequence length with zeros. Then, calculate the sum of the products of corresponding elements of the shifted sequence B and sequence A; this sum is the cross-correlation value corresponding to displacement d. By iterating through all possible values ​​of d, a series of cross-correlation values ​​are obtained. Construct a two-dimensional coordinate system with displacement d as the x-axis and the corresponding cross-correlation value as the y-axis. Connecting all the points generates the cross-correlation function curve. The sequence of y-axis values ​​of this curve is the set of all calculated cross-correlation values ​​arranged in order of displacement d.

[0023] Step S223: Determine the peak position of the cross-correlation function curve, use the time offset corresponding to the peak position as the candidate propagation delay time parameter, and record the ordinate value of the peak position as the correlation confidence level.

[0024] Peak detection is performed on the cross-correlation function curve generated in step S222. A one-dimensional extreme value search algorithm is used to traverse all points on the curve and find the point with the largest ordinate value. The abscissa of this point is the displacement corresponding to the peak position, which is assumed to be K units of sampling interval. Since each sampling point corresponds to a spatial position on the blood vessel and the frame interval of image acquisition is known, it is necessary to convert the spatial displacement into a time delay. This requires combining the estimation of blood flow velocity. Since the blood flow velocity is unknown, a time unit can be directly defined using the frame interval. However, a more refined approach is to understand the displacement K as the number of spatial movement steps of the contrast agent blobs. Since the time interval between each frame image is fixed, the displacement K can be multiplied by the frame interval time to obtain a preliminary time delay value, which serves as a candidate propagation delay time parameter. At the same time, the ordinate value of the peak point is extracted as the correlation confidence of this candidate parameter.

[0025] Step S224: Based on the ratio between the candidate propagation delay time parameter and the difference in acquisition timestamps of adjacent image units, the rationality of the candidate propagation delay time parameter is verified. When the ratio exceeds the preset range, the candidate propagation delay time parameter is weighted and adjusted based on the correlation confidence level so that the adjusted parameter falls within the preset range.

[0026] A preset ratio range, such as from 0 to 1.5, is established, indicating that the propagation delay time parameter should not exceed 1.5 times the acquisition time interval and cannot be negative. The ratio of the candidate propagation delay time parameter to the difference in acquisition timestamps between the corresponding adjacent image units obtained from step S210 is calculated. If this ratio is within the preset range, the candidate parameter is directly retained. If the ratio is less than zero, it indicates that the calculated delay time is negative, which does not conform to the physiological fact of positive contrast agent flow. In this case, the candidate propagation delay time parameter is forcibly set to a very small positive value, such as one-tenth of the acquisition interval. If the ratio is greater than the upper limit of the preset range, such as exceeding 1.5, a weighted adjustment is initiated. A weighting factor is constructed based on the correlation confidence calculated for the pair of image units and the average correlation confidence corresponding to the candidate parameters calculated for other adjacent image units with ratios within a reasonable range. This weighting factor makes the adjusted new parameter equal to an interpolation between the preset upper limit and the candidate parameter, with the interpolation weight determined by the comparison between the candidate parameter's own confidence and the average confidence. The higher the confidence level, the closer the adjusted parameter is to the original candidate value; the lower the confidence level, the closer the adjusted parameter is to the preset upper limit value, thus ensuring that the final parameter falls within a reasonable range.

[0027] Step S225: The verified candidate propagation delay time parameter is weighted and averaged with other candidate parameters calculated from multiple groups of adjacent image units to generate the final propagation delay time parameter. The weight of the weighted average is determined based on the correlation confidence of each group.

[0028] Assuming the original image data sequence contains N image units, steps S221 to S224 yield N-1 verified candidate propagation delay time parameters and the corresponding correlation confidence score for each parameter. These N-1 candidate parameters and their confidence scores are used as input for a weighted average calculation. For each pair of adjacent image units, the final weighting coefficient is equal to the confidence score of that pair divided by the sum of the confidence scores of all N-1 pairs. Then, each pair of candidate propagation delay time parameters is multiplied by its corresponding weighting coefficient, and finally, all products are summed to obtain the final propagation delay time parameter for the entire sequence. This final parameter is not a single numerical value, but a feature value representing the average level of the entire sequence, which will be used in subsequent steps to construct a unified spatiotemporal mapping relationship.

[0029] Step S226: Associate and store the final propagation delay time parameter with the timestamp information of the corresponding adjacent image unit, and establish a mapping table between the propagation delay time parameter and the image unit index.

[0030] Create a two-dimensional data table with the same number of rows as the number of image units N in the original image data sequence. The first column stores the index number of the image unit, for example, from zero to N-1. The second column stores the original acquisition timestamp of the corresponding image unit extracted from step S210. For the first image unit, since it has no previous image to pair with, the associated final propagation delay time parameter can be set to zero or a default value. For the second and subsequent image units, such as the image unit with index i, the associated final propagation delay time parameter is the unique global average calculated in step S225. Based on this, the image unit with index i corresponds to a row in the table, containing its index, original timestamp, and global average delay time. This table is the mapping table between the propagation delay time parameter and the image unit index, which is stored in the system's memory or hard disk for subsequent steps to access.

[0031] Step S230: Establish a spatiotemporal mapping relationship based on the time interval distribution characteristics and propagation delay time parameters, map each two-dimensional angiography image unit in the original image data sequence to a unified time reference coordinate system, and generate the mapped time coordinates of each image unit in the unified time reference coordinate system.

[0032] In this embodiment, step S230 specifically includes steps S231 to S236: Step S231: Using the acquisition timestamp of the first two-dimensional angiography image unit in the original image data sequence as the time reference origin, the time intervals in the time interval distribution feature are accumulated with the time reference origin to generate a reference time coordinate sequence. The reference time coordinate sequence contains the reference time coordinates of each image unit.

[0033] Specifically, from the list of original acquisition timestamps obtained in step S210, the timestamp of the first image unit is extracted and its value is set to zero, serving as the time reference origin. Then, the time interval sequence generated in step S210 is read, which contains N-1 time interval values. A reference time coordinate sequence array of length N is created. The first element of the array is set to zero. Next, for the second to Nth image units, their reference time coordinate is equal to the reference time coordinate of the previous image unit plus the corresponding previous time interval. For example, the reference time coordinate of the second image unit is equal to zero plus the first time interval, the reference time coordinate of the third image unit is equal to the reference time coordinate of the second image unit plus the second time interval, and so on. Finally, a reference time coordinate sequence is obtained that corresponds one-to-one with the original image data sequence, with each coordinate value precisely reflecting the acquisition time sequence recorded by the device.

[0034] Step S232: Based on the deviation between the propagation delay time parameter and the corresponding time point in the reference time coordinate sequence, determine the offset correction amount of each two-dimensional angiography image unit on the time axis, and associate the offset correction amount with the corresponding image unit.

[0035] For example, the final propagation delay time parameter, denoted as T, is obtained from the mapping table established in step S226. delay From the reference time coordinate sequence generated in step S231, obtain the reference time coordinates for each image unit. For the i-th image unit, its reference time coordinates are denoted as t. ref_i Define the offset correction Δi for this image unit. Calculating the offset correction requires constructing a cumulative deviation function. First, based on T... delay This allows us to calculate the theoretically required total duration of the entire sequence, which is N multiplied by T. delay The last value t of the actual reference time coordinate sequence. ref_N This is the actual total acquisition time. The ratio between the two, i.e., T, is the total acquisition time. delay Multiply by N and divide by t ref_N This reflects the difference in average time scale. For each image unit in the middle, its offset correction Δi can be achieved by adjusting the reference time coordinate t. ref_i Multiply by this ratio, then subtract t ref_iThe value of Δi is obtained from the image unit itself. Based on this, the value of Δi may differ for different i, allowing subsequent adjustments to maintain overall scale consistency while also enabling local fine-tuning. The calculated Δi is associated with and stored with the index i of the image unit, forming a list of offset correction amounts.

[0036] Step S233: Based on the offset correction, locally adjust the reference time coordinate sequence to generate a corrected time coordinate sequence, so that the propagation delay of blood flow contrast agent distribution density information between adjacent image units matches the actual physiological flow process.

[0037] Specifically, a new array is created to store the corrected time coordinate sequence, with a length equal to the number of image units N. The array is iterated from the first to the Nth image unit. For the i-th image unit, its reference time coordinate t is obtained from step S231. ref_i The offset correction amount Δi is obtained from step S232. Then, the corrected time coordinate t of the i-th image unit... corr_i It equals t ref_i Add Δi. After performing this operation on all i, a corrected time coordinate sequence is obtained. By observing this new sequence, the time difference between adjacent image units, i.e., t, can be identified. corr_i+1 Subtract t corr_i This is closer to the actual propagation delay characteristics of the pair of image units obtained from step S220, thus achieving physiological matching on the time axis.

[0038] Step S234: Associate and store the vascular tree projection contour information of each two-dimensional angiography image unit with the corresponding time coordinates in the corrected time coordinate sequence to generate a spatiotemporally aligned data unit with a unified time reference.

[0039] For example, a new data structure is created, such as an array of structures or a database table. Each element of this data structure corresponds to an image unit in the original image data sequence. Each element contains two main parts: the first part is the vascular tree projection contour information extracted and stored from the original image unit, which can be represented as a set of two-dimensional coordinates of a series of contour points; the second part is the corrected time coordinates corresponding to the image unit index obtained from the corrected time coordinate sequence generated in step S233. In this way, the original image unit is encapsulated into new spatiotemporally aligned data units. These data units are arranged in ascending order of their corrected time coordinates, thus forming a temporally aligned data set.

[0040] Step S235: Arrange the spatiotemporally aligned data units in the order of the corrected time coordinate sequence to construct the basic elements of the intermediate corrected image set.

[0041] Using the spatiotemporally aligned data unit set generated in step S234 as input, the entire set is sorted according to the corrected time coordinate values ​​stored in each data unit. Algorithms such as quicksort or mergesort can be used, with the corrected time coordinate as the sort key, to rearrange all data units in ascending order of time coordinate. After sorting, a new ordered list is generated, where each element is still a complete spatiotemporally aligned data unit. This ordered list is defined as the basic element of the intermediate corrected image set, which has achieved temporal alignment and is ready for the next step of spatial location correction.

[0042] Step S236: Perform continuity verification on the corrected time coordinates in the spatiotemporal aligned data unit. When there are time coordinate inversions or jump anomalies, re-interpolate and adjust the time coordinates of the abnormal area to generate a set of corrected intermediate images.

[0043] For example, the basic element list of the intermediate corrected image set generated in step S235, sorted by corrected time coordinates, is traversed. Starting from the second data unit, the corrected time coordinate of the current data unit is checked sequentially to see if it is greater than the corrected time coordinate of the previous data unit. If it is less than or equal to, it is marked as a time coordinate inversion anomaly. At the same time, the time difference between adjacent data units is calculated. If this time difference is less than a preset lower threshold or greater than a preset upper threshold, it is marked as a jump anomaly. When an anomaly is detected, the continuous interval in which the anomaly occurs is determined. Then, the time coordinates of the previous normal data unit and the first normal data unit after the interval, as well as the index positions of these two normal data units in the list, are extracted. Using these two normal data points as endpoints, linear interpolation or spline interpolation algorithms are used to calculate new time coordinate values ​​for all data units within the interval, which are uniformly increasing from the start point to the end point. These new time coordinate values ​​replace the original anomaly values. After completing the interpolation correction of all anomaly areas, the final time coordinate sequence guarantees strict monotonic increase and a reasonable gradient. The data unit set at this time constitutes the corrected intermediate corrected image set.

[0044] Step S240: Based on the spatiotemporal mapping relationship, the spatial position coordinates of the two-dimensional angiography image unit are offset and corrected to eliminate the position offset of the vascular tree projection contour caused by physiological motion, and an intermediate corrected image set with time alignment marks is generated. The intermediate corrected image set contains the two-dimensional angiography image unit with corrected spatial position coordinates.

[0045] First, from the corrected intermediate image set, a time point is selected as the reference phase, for example, the spatiotemporally aligned data unit corresponding to the moment when the heart is most fully diastolic and the motion is slowest. Then, using image registration techniques, the vascular tree projection contours of all other time point data units in the set are registered with the data units of the reference phase. The registration process employs a B-spline-based free deformation model, which can simulate non-rigid physiological motion. For each image unit to be corrected, the algorithm optimizes an energy function to find a set of optimal B-spline control point displacements, such that the vascular tree projection contour of that image unit, after undergoing a deformation field defined by the control point displacements, achieves the maximum match with the vascular tree projection contour of the reference phase image unit. This energy function typically includes two parts: one is a similarity metric, such as mutual information or normalized cross-correlation coefficient, used to measure the similarity between the deformed image and the reference image; the other is a smoothness constraint, used to ensure the physical rationality of the deformation field. After optimization, a deformation field from the image to be corrected to the reference image is obtained. Applying this deformation field to each pixel coordinate of the image unit to be corrected yields the corrected spatial coordinates of that image unit in the reference phase space. Performing this operation on all non-reference phase image units results in a complete set of intermediate corrected images, where all image units have been corrected to the same spatial location, and each image unit in this set carries the corrected vascular tree projection contour coordinates.

[0046] Step S250: Continuously adjust the blood flow contrast agent distribution density information of each two-dimensional angiography image unit in the intermediate correction image set to maintain a smooth transition of the density distribution gradient between adjacent image units, thereby generating a transition correction image set with a smooth density gradient.

[0047] First, all image units in the intermediate-corrected image set are considered as a three-dimensional image volume, where two dimensions are the x and y coordinates of the image plane, and the third dimension is time t. For each fixed pixel location in the image plane, the gray values ​​at all time points are extracted to form a one-dimensional temporal intensity curve. Since spatial location correction has been performed previously, the same pixel location at different time points corresponds to the same anatomical point on the vascular tree. Then, a low-pass filter, such as a Gaussian filter or a Savitzky-Golay smoothing filter, is applied to this one-dimensional temporal intensity curve. The filter parameters need to be set according to the typical frequency of contrast agent flow to preserve the main hemodynamic changes while filtering out high-frequency noise. After filtering, each pixel location obtains a new, smoother temporal intensity curve. Finally, these smoothed intensity values ​​are refilled back into the corresponding time points and pixel locations to generate a new set of two-dimensional angiography image units. These new image units maintain the original spatial structure, but the gray values ​​of each pixel have become continuously smooth in the temporal dimension, thus forming an intermediate-corrected image set with a smooth density gradient.

[0048] Step S260: Perform contour consistency verification on the vascular tree projection contour information of each two-dimensional angiography image unit in the transition correction image set. When there is contour breakage or overlap anomaly, perform interpolation repair based on the contour information of adjacent image units to obtain a complete corrected image data set.

[0049] Specifically, for each image unit in the overcorrected image set, an active contour model algorithm is employed, using its existing vascular tree projection contour as the initial contour. This contour iteratively evolves on the image to precisely match the gradient edges. During the evolution process, the algorithm automatically detects contour breaks and overlaps. Contour breaks typically manifest as the active contour failing to converge to strong edges in certain areas, while contour overlap manifests as self-intersections or intersections of contour lines from different branches. When a break is detected, its location is recorded. Using temporally adjacent image units, such as the previous and next frames, complete contour segments at the corresponding break locations are extracted. Then, using the time coordinate as the independent variable, linear interpolation is performed on the point coordinates of these two contour segments to generate repaired contour points for the broken region in the current frame. The repaired contour points are connected to the original, correct contour portion in the current frame to form a complete contour. When an overlap is detected, the overlapping contour segments are first identified. Similarly, referring to adjacent frames, the correct topological relationship of the corresponding region contours in adjacent frames is analyzed. Based on this correct topological relationship, overlapping contours are separated and resampled in the current frame, redundant points are removed, and each contour line is ensured to be a simple closure. After repairing all image units, the final set of corrected image data with intact contours is obtained.

[0050] Step S300: Spatial dimension expansion is performed on the corrected image data set, and an initial three-dimensional vascular tree model of the target cardiovascular region is generated based on the geometric deformation characteristics of the vascular tree projection contour information under multiple preset projection angles.

[0051] In this embodiment, step S300 specifically includes the following steps S310 to S360: Step S310: Select at least two two-dimensional angiography image units acquired from different projection angles from the corrected image data set, extract the boundary curve coordinates and vessel centerline trajectory coordinates of the vascular tree projection contour in each image unit, and classify and store the boundary curve coordinates and vessel centerline trajectory coordinates according to the projection angle.

[0052] First, based on the DICOM metadata recorded by the DSA system, such as tags, image sequences with different projection angles are selected. Assume two angles are chosen: 45 degrees left anterior oblique and 30 degrees right anterior oblique. From each angle sequence, the two-dimensional angiography image unit corresponding to the time point where the contrast agent filling is most complete and the blood vessels are most clearly displayed is selected. Then, these two image units are processed separately. A region-growing-based segmentation algorithm, combined with manual correction, is used to accurately extract the boundary curve coordinates of the vascular tree projection contour, resulting in a series of two-dimensional points, which are then connected sequentially to form a closed contour. Based on this, a distance-transform-based skeleton extraction algorithm is used to perform a distance transformation on the segmented vascular region binary image. Then, by tracing the ridges of the distance map, the coordinates of the vascular centerline trajectory are extracted, resulting in a series of two-dimensional points, which are then connected sequentially to form a single-pixel-wide line. Finally, the boundary curve coordinates and centerline trajectory coordinates extracted from the left anterior oblique image are stored in a data structure named "LAO45," and the data extracted from the right anterior oblique image is stored in another data structure named "RAO30."

[0053] Step S320: Perform spatial intersection matching on the coordinates of the vessel centerline trajectory. Based on the geometric perspective relationship under different projection angles, determine the intersection position of the vessel centerline trajectory coordinates in three-dimensional space, generate the initial spatial coordinate set of the vessel bifurcation point, and associate the initial spatial coordinate set with the corresponding vessel bifurcation point identifier.

[0054] In this embodiment, step S320 specifically includes steps S321 to S325: Step S321: Extract the trajectory coordinates of the first blood vessel centerline in the two-dimensional angiography image unit under the first projection angle, extract the trajectory coordinates of the second blood vessel centerline in the two-dimensional angiography image unit under the second projection angle, and store the trajectory coordinates of the first blood vessel centerline and the second blood vessel centerline as the first coordinate set and the second coordinate set, respectively.

[0055] Following the example from step S310, the first projection angle is set to 45 degrees at the left front oblique angle, and the second projection angle is set to 30 degrees at the right front oblique angle. From the data structure already categorized and stored in step S310, the centerline trajectory coordinates of the 45-degree left front oblique angle image unit are directly read, resulting in a list of two-dimensional points, named the first coordinate set C1. Each point in the list contains its x and y coordinates in the image coordinate system. Similarly, the centerline trajectory coordinates of the 30-degree right front oblique angle image unit are read, resulting in the second coordinate set C2. These two sets will be used for subsequent ray generation and intersection calculations.

[0056] Step S322: Based on the imaging geometric parameters corresponding to the first projection angle and the second projection angle, the trajectory coordinates of the first blood vessel centerline and the second blood vessel centerline are converted into the first projection ray set and the second projection ray set in three-dimensional space, respectively. Each ray in the first projection ray set and the second projection ray set is defined by the starting point and the direction vector.

[0057] First, the imaging geometry parameters for the 45-degree left anterior oblique acquisition are obtained from the DICOM metadata. These include the coordinates S1 of the X-ray source focus in the world coordinate system, as well as the spatial position and orientation of the detector. This allows the construction of a projection matrix from 3D world coordinates to 2D image coordinates. Using the inverse relationship of this projection matrix, for each 2D point p1 in the first coordinate set C1, a ray originating from focus S1 and passing through point p1 can be calculated. The starting point of this ray is S1, and its direction vector is a unit vector pointing from S1 to an intermediate 3D point calculated by backprojection from p1. This intermediate 3D point is typically taken as a point on the detector plane. This process is repeated for all points in C1 to obtain the first projection ray set R1. Similarly, using the imaging geometry parameters for the 30-degree right anterior oblique acquisition, focus S2, the same operation is performed for each point p2 in the second coordinate set C2 to obtain the second projection ray set R2.

[0058] Step S323: Determine the shortest spatial distance between each ray in the first set of projection rays and the second set of projection rays. When the shortest spatial distance is less than the preset matching tolerance threshold, determine the midpoint of the corresponding two rays as the candidate intersection point spatial coordinates, and associate and store the candidate intersection point spatial coordinates with the corresponding two ray identifiers.

[0059] For each ray r1i in the first set of projected rays R1 and each ray r2j in the second set of projected rays R2, pair them together. For each pair of rays, calculate the shortest spatial distance between them. This can be achieved by solving an optimization problem: find a point a on r1i and a point b on r2j such that the length of the vector ab is minimized. This minimum length is the shortest distance d between the two rays. ij If d ij If the distance is less than a preset matching tolerance threshold, such as a physical length equivalent to the average diameter of a blood vessel, then the pair of rays is considered matched. The midpoint m of the line connecting points a and b is taken. ij Let m be the spatial coordinates of a candidate intersection point generated by rays r1i and r2j. Record this candidate point m. ij The coordinates of the point and the identifiers of the two rays that generated it, such as the index i of r1i and the index j of r2j, are stored as an entry in the candidate intersection list.

[0060] Step S324: Cluster and merge the spatial coordinates of candidate intersection points. Based on the topological connectivity of the bifurcation region in the vascular tree projection contour information, merge the spatial coordinates of multiple candidate intersection points belonging to the same vascular bifurcation point into initial spatial coordinates representing the unique position of the bifurcation point. The merging process is based on the spatial distance between candidate intersection points and the vascular topological connectivity.

[0061] Using the candidate intersection list generated in step S323 as input, a density-based spatial clustering algorithm, such as the DBSCAN algorithm, is employed to cluster all candidate points. The key parameters of the algorithm are the neighborhood radius and the minimum number of points. The neighborhood radius can be set based on the average diameter of the blood vessels to ensure that candidate points belonging to the same bifurcation point are clustered together. The minimum number of points can be set to a small integer, such as 2 or 3, to exclude isolated candidate points that may be generated by noise. After clustering, each cluster represents a potential blood vessel bifurcation point. For each cluster, the average coordinates of all candidate points within the cluster are calculated as the initial spatial coordinates of that bifurcation point. Then, the blood vessel tree projection contour information extracted in step S310, especially the connectivity at the bifurcation point, is used to verify and adjust these clusters. For example, if two clusters that are close in three-dimensional space show that they connect completely different blood vessel segments in the two-dimensional projection image, then they may indeed be two independent bifurcation points and should not be merged. In this way, a unique initial spatial coordinate is finally determined for each bifurcation point and recorded.

[0062] Step S325: Sort the initial spatial coordinates according to the order of the bifurcation points recorded in the vascular tree projection contour information, and associate and store the connecting vascular segment identifiers corresponding to each bifurcation point to generate an initial spatial coordinate set of vascular bifurcation points. The initial spatial coordinate set contains the spatial coordinates and connection relationship information of each bifurcation point.

[0063] Based on the two-dimensional vascular tree projection contour information extracted in step S310, the topological relationships of the bifurcation points can be analyzed. For example, starting from the beginning of the trunk, the first bifurcation point divides the trunk into two branches. An order is assigned to all bifurcation points according to the blood flow direction from proximal to distal. Then, the initial spatial coordinates of each bifurcation point obtained after clustering and merging in step S324 are arranged in this order to form a list. For each bifurcation point in the list, a connection table is also stored, recording all vascular segments connected to that bifurcation point. Each vascular segment is represented by a unique identifier, such as "LAD-Proximal" or "LAD-Diagonal1". In this way, a data structure is finally generated, which contains the sequentially arranged three-dimensional coordinates of the bifurcation points and the vascular segment connection relationships between these bifurcation points, i.e., the initial spatial coordinate set of the vascular bifurcation points.

[0064] Step S330: Based on the initial set of spatial coordinates of the bifurcation points of blood vessels and the boundary curve coordinates of the projection contour of the blood vessel tree, determine the spatial extension direction of the blood vessel segment, and perform curve fitting on the blood vessel segments between adjacent bifurcation points according to the spatial extension direction to generate the spatial trajectory curve of the blood vessel segment. The spatial trajectory curve contains the three-dimensional coordinates of each point on the curve.

[0065] For two adjacent bifurcation points A and B, such as the proximal bifurcation point of the left anterior descending branch and the bifurcation point of the first diagonal branch, their coordinates P in three-dimensional space are known. A and P B Simultaneously, from the two-dimensional projection image extracted in step S310, the centerline trajectory coordinates of the blood vessel segment at at least one projection angle can be obtained. These two-dimensional centerline points are then back-projected into rays in three-dimensional space based on their imaging geometry parameters. These rays are correlated with those obtained by P... A and P B The defined straight line segments together constitute the constraints. Using the B-spline curve fitting algorithm, P... A and P BThe algorithm forces points to pass through the beginning and end points of the curve. Then, for the rays corresponding to the intermediate points obtained from the projected image, the algorithm optimizes the control points of the B-spline curve so that the points on the curve are as close as possible to these rays, while ensuring the smoothness of the curve. By solving this constrained optimization problem, a smooth three-dimensional B-spline curve is finally obtained. The points on this curve can be sampled in the parameter domain to obtain a series of three-dimensional coordinates, thereby generating a complete spatial trajectory curve of the blood vessel segment.

[0066] Step S340: The spatial trajectory curves of the blood vessel segments are spliced ​​together according to the blood vessel branch connection order recorded in the blood vessel tree projection contour information to form a preliminary three-dimensional skeleton structure containing the topological connection relationship of blood vessels. The preliminary three-dimensional skeleton structure consists of the spatial trajectory curves of the blood vessel segments and the spatial coordinates of the blood vessel bifurcation points.

[0067] Specifically, a graph data structure is created, where the nodes are the initial spatial coordinates of the bifurcation points of the blood vessels generated in step S325, and the terminal points of the blood vessels. The terminal points can be determined by tracing the two-dimensional centerline to the endpoint where there are no more branches. Then, for each spatial trajectory curve of a blood vessel segment generated in step S330, the two nodes it connects to are determined based on the attributes of its two ends. For example, a curve of the left anterior descending artery connects proximally to the starting node of the aortic root and distally to the bifurcation point of the first diagonal branch. This curve segment is added to the graph as an edge, and the data on the edge is the complete three-dimensional coordinate point set of the curve. By traversing all blood vessel segments and gradually adding edges to the graph, a complete tree diagram is finally constructed. This graph is the preliminary three-dimensional skeleton structure, which contains the three-dimensional coordinates of all nodes (bifurcation points and terminal points) and the three-dimensional centerline trajectory curves of all edges (blood vessel segments).

[0068] Step S350: Reconstruct the surface contour of the preliminary three-dimensional skeleton structure, map the boundary curve coordinates of the vascular tree projection contour to the vertical plane of the spatial trajectory curve of the vascular segment, and generate a three-dimensional lumen surface mesh that wraps the spatial trajectory curve by contour stacking to obtain the initial three-dimensional vascular tree model.

[0069] In this embodiment, step S350 specifically includes the following steps S351 to S356: Step S351: Extract multiple sampling points along the spatial trajectory curve of the blood vessel segment with a fixed sampling step size. At each sampling point, construct a local projection plane perpendicular to the tangent direction of the spatial trajectory curve, and align and store the normal vector of the local projection plane with the tangent direction of the spatial trajectory curve.

[0070] For each vessel segment's spatial trajectory curve in the preliminary 3D skeleton structure generated in step S340, such as a segment of the left anterior descending artery, its total length is first calculated. Then, a fixed sampling step size is set, for example, a length related to the vessel diameter. Starting from the curve's origin, a sampling point is calculated at every step size. At each sampling point, the tangent vector of the curve at that point is calculated. According to differential geometry, the tangent vector can be obtained by differentiating the curve parameters. Then, a plane is constructed with the sampling point as the origin and the tangent vector as the normal vector. This plane is the local projection plane. The normal vector (i.e., the tangent direction) of this plane and two orthogonal basis vectors within the plane are recorded for subsequent coordinate transformations. This information is stored along with the spatial coordinates of the sampling point itself.

[0071] Step S352: Extract the two-dimensional angiography image unit corresponding to the time of each sampling point from the corrected image data set, and map the boundary curve coordinates of the vascular tree projection contour in the two-dimensional angiography image unit onto the local projection plane through inverse projection transformation to generate the vascular cross-section contour curve at the sampling point. The vascular cross-section contour curve is composed of a closed contour point sequence.

[0072] First, from the corrected image dataset, a suitable image unit is selected for the vessel segment being processed, such as the frame where the heart is in end-diastole and the contrast agent concentration is highest. From this image unit, the boundary curve coordinates of the vessel segment are extracted. For sampling point P... i For each point q on the two-dimensional boundary curve, based on its imaging geometry parameters, a ray originating from the focal point and passing through q is generated, along with its local projection plane. The intersection of this ray and plane P is then calculated. i The intersection of these points yields a three-dimensional point r. The coordinates of r are then transformed to coordinates relative to P. i In a local two-dimensional coordinate system with the origin at the origin and the two basis vectors of the plane as coordinate axes, the two-dimensional coordinates are obtained. By performing this operation on all points on the boundary curve, a series of two-dimensional points can be obtained, forming a point set in the local coordinate system. Due to the limitation of the projection angle, this point set may only cover a part of the cross-sectional profile. To obtain the complete closed profile, the profile information of the same blood vessel segment at other angles can be combined. Through similar mapping and fusion, a complete and closed sequence of two-dimensional points, i.e., sampling point P, is finally obtained. i The profile curve of the blood vessel cross section at that location.

[0073] Step S353: Perform contour point matching on the contour curves of blood vessel cross sections at adjacent sampling points. Based on the geometric similarity of the contour curves and the continuity constraint of the blood vessel centerline, establish vertex connection relationships between adjacent cross section contour curves. Vertex connection relationships are recorded through a connection edge index table.

[0074] For two adjacent sampling points Pi and P i+1 The corresponding blood vessel cross-sectional contour curves are contour C. i and outline C i+1 First, the point sequences of both contours are normalized according to the same starting direction, for example, both starting from the innermost point of the blood vessel. Then, a dynamic programming algorithm is used to solve for the optimal correspondence between the two contour points. The cost function of the algorithm consists of two parts: one part is the geometric feature difference between the point pair, such as the curvature at that point, the distance to the centroid of the contour, etc.; the other part is the length penalty of the connecting edge, that is, if C i Point a on C is connected to C i+1 If point b is on the surface of the blood vessel, then the length of the connecting edge ab should not be too long to satisfy the assumption of a smooth transition on the surface of the blood vessel. Through dynamic programming, a path from point b to point b on the surface of the blood vessel can be found. i From the starting point to C i+1 The path from the starting point minimizes the cumulative cost, and this path defines the correspondence between contour points. This correspondence is recorded in a connection edge index table, where each row contains C. i A point index on and its corresponding C i+1 A point index on the [theory / system].

[0075] Step S354: Generate triangular patches that connect adjacent cross-sectional contour curves based on vertex connection relationships. Continuously splice the triangular patches along the spatial trajectory curve of the blood vessel segment to form a three-dimensional lumen surface mesh unit covering the entire blood vessel segment. The three-dimensional lumen surface mesh unit is composed of a set of triangular patches.

[0076] In this step, generating triangular patches based on vertex connectivity is the core of constructing the 3D surface. For two adjacent cross-sectional profiles, a series of quadrilaterals can be naturally generated based on the established point correspondences, and then each quadrilateral is further subdivided into two triangles. For example, for C... i Point a on j and a j+1 and their presence in C i+1 The corresponding point b k and b k+1 These four points form a quadrilateral, which can be connected to point a. j -b k -a j+1 and b k -a j+1 -b k+1 Two triangles are generated. The triangular patches are then continuously spliced ​​along the blood vessel segment. This involves performing the above operation on all adjacent cross-sectional contour pairs to generate a series of interconnected triangles, forming a tubular mesh surface that covers the entire blood vessel segment.

[0077] In this embodiment, the specific implementation of generating triangular patches and forming mesh cells is as follows. Following step S353, for each pair of adjacent contours C i and C i+1 And a table of vertex connections between them. Traverse the connection table, for each pair of corresponding points recorded in the table, such as C i Point a on j Corresponding to C i+1 Point b on k Meanwhile, consider the next corresponding pair of points, namely a. j+1 Corresponding to b k+1 So, point a j a j+1 b k b k+1 This forms a spatial quadrilateral. Connecting aj and b... k+1 , or connect a j+1 and b k This can divide a quadrilateral into two triangles. Choose a division method, such as connecting aj and b. k+1 Generate triangles T1 and T2. Record the vertex indices of these two triangles. Repeat this process for all j to generate connection C. i and C i+1 All the triangular facets. Collecting the vertex indices and coordinate information of all these triangular facets constitutes the complete three-dimensional luminal surface mesh unit of this blood vessel segment.

[0078] Step S355: The three-dimensional lumen surface mesh units corresponding to different blood vessel segments are bounded together at the blood vessel bifurcation point to eliminate gaps and overlapping areas between mesh units, generating a continuous and closed three-dimensional lumen surface mesh. The binding process is based on contour matching and vertex merging at the bifurcation point.

[0079] Taking a bifurcation point as an example, suppose three blood vessels converge there: inflow vessel segment A, and two outflow vessel segments B and C. In the mesh cells generated in step S354, the distal mesh boundary of vessel segment A is an open profile composed of a series of vertices. Similarly, the proximal mesh boundaries of vessel segments B and C are also open profiles. The first step in suturing is to identify these boundary profiles. Then, it is necessary to determine which vertices on the boundary profile of vessel segment A should connect to which vertices on the boundary profile of vessel segment B, and which connect to the boundary of vessel segment C. This typically requires reference to the anatomical structure at the bifurcation point. By calculating the spatial distance and normal consistency between the boundary vertices, the profile point set of A can be divided into two parts, corresponding to the profiles of B and C respectively. Then, for the corresponding profile parts of A and B, a method similar to step S354 can be used to generate new triangular patches to connect the two open boundaries, instead of simply merging vertices. The same applies to A and C. Finally, the three independent mesh units A, B, and C are smoothly connected at the bifurcation point using the newly generated triangular facets, forming a continuous and seamless three-dimensional cavity surface mesh.

[0080] Step S356: Evaluate the surface smoothness of the three-dimensional lumen surface mesh. When there are local uneven areas on the surface, relax and adjust the mesh nodes in that area to make the surface curvature change more gradual, and generate the initial three-dimensional vascular tree model.

[0081] First, traverse every vertex on the continuous 3D vascular surface mesh generated in step S355. For each vertex, calculate the normal vector of all triangular patches formed by it and all its neighboring vertices, and then calculate the rate of change of these normal vectors. If the normal vector of a vertex differs significantly from the normal vectors of other vertices in its neighborhood, it indicates that there is unevenness on the surface at that location. For vertices marked as uneven regions, perform a relaxation operation. The relaxation operation is performed by calculating the average of the spatial coordinates of all its first-order neighbor vertices (i.e., vertices directly connected to the vertex by edges) for that vertex, obtaining a new position. Then, perform a weighted average of this new position and the vertex's original position, with the weight controlled by a smoothing intensity coefficient. The smoothing intensity coefficient is usually a positive number less than 1 to ensure that the adjustment is gradual and controllable. Update the vertex coordinates to the weighted average position. Perform one or more iterations of relaxation adjustment for all abnormal vertices until the rate of change of the normal vector of the entire surface is lower than a preset smoothness threshold. The mesh after all adjustments is confirmed as the initial 3D vascular tree model.

[0082] Step S360: Perform mesh quality detection on the three-dimensional lumen surface mesh in the initial three-dimensional vascular tree model. When a distorted or self-intersecting region is detected in the mesh, rearrange the mesh nodes in the corresponding region to generate an optimized initial three-dimensional vascular tree model. Then, compare and verify the optimized model with the preliminary three-dimensional skeleton structure.

[0083] For the surface mesh of the initial 3D vascular tree model, all triangular faces are traversed first, and the interior angles and quality indices of each triangle are calculated, such as minimum angle, maximum angle, and side length ratio. If the minimum angle of a triangle is less than a preset angle threshold, or the maximum angle is greater than a preset angle threshold, it is marked as a distorted triangle. Then, the self-intersection of the mesh is detected. By constructing a spatial index structure, such as an octree, potentially intersecting triangle pairs are quickly identified, and precise intersection tests are performed. For regions with distortion or self-intersection, a local mesh re-partitioning algorithm is used. For example, for a small patch of distorted triangles, they can all be removed to form a small hole. Then, based on the vertices on the boundary of the hole, the Delaunay triangulation algorithm is used to regenerate more regularly shaped triangles to fill the hole. For self-intersecting regions, the self-intersecting vertices or edges need to be identified, and the self-intersection is resolved by moving the vertices along a set direction. After optimization, a new mesh is obtained. To verify this, the centerline is re-extracted from the optimized mesh, for example, using a skeleton extraction algorithm, and compared with the preliminary 3D skeleton structure generated in step S340. The Hausdorff distance between the two is calculated. If the distance is less than a preset error threshold, the verification is successful, confirming that the optimization process did not introduce structural errors, and the optimized initial 3D vascular tree model is finally output.

[0084] Step S400: Combine the preset hemodynamic simulation parameters with the initial three-dimensional vascular tree model for joint optimization. By analyzing the distribution of blood flow contrast agent density in the flow path of the initial three-dimensional vascular tree model, the spatial topology of the initial three-dimensional vascular tree model is corrected to obtain the corrected three-dimensional vascular tree model.

[0085] In this embodiment, step S400 specifically includes steps S410 to S460: Step S410: Load the blood flow velocity distribution curve and blood flow pressure gradient parameters from the hemodynamic simulation parameters onto the vascular lumen surface mesh nodes of the initial three-dimensional vascular tree model to establish the initial boundary conditions for blood flow. The initial boundary conditions include the blood flow velocity vector and pressure value at each mesh node.

[0086] First, pre-defined hemodynamic simulation parameters are obtained from medical literature or clinical measurements. For example, the inlet blood flow velocity distribution curve of the coronary artery is set as a time-pulsating waveform, which can be described by a periodic function. The outlet pressure is set as constant, representing the resistance of the capillary bed. Then, on the initial 3D vascular tree model, the inlet surface (e.g., the aortic root) and various outlet surfaces (e.g., the ends of branch vessels) are identified. All grid nodes on the inlet surface are traversed, and a velocity vector is assigned to each node according to the blood flow velocity distribution curve. The direction of the velocity vector is perpendicular to the inlet surface and points inward into the model. The magnitude can be adjusted according to the node's position on the inlet section (faster at the center, slower at the edge) to simulate a realistic parabolic or plug flow profile. Similarly, a constant pressure value is assigned to all nodes on the outlet surface. For nodes on the vessel wall, a no-slip boundary condition is set, i.e., the velocity is zero. This set of grid nodes assigned velocity and pressure values ​​constitutes the initial boundary conditions of the entire model.

[0087] Step S420: Extract the blood flow contrast agent distribution density information at continuous time points from the corrected image data set, map the blood flow contrast agent distribution density information to the corresponding spatial location of the initial three-dimensional vascular tree model, and generate a contrast agent concentration spatiotemporal distribution field. The contrast agent concentration spatiotemporal distribution field contains the concentration value of each spatial location at different time points.

[0088] First, image units covering all time points throughout the contrast agent's passage are selected from the corrected image dataset. For each grid node in the initial 3D vascular tree model, such as a node located in the middle of the left anterior descending artery, its corresponding pixel in the 2D image unit at each time point needs to be found. This can be achieved by projecting the 3D node coordinates onto the 2D image plane according to the imaging geometry parameters of the image unit at each time point. Since the image unit has been spatially corrected, the projection point should fall exactly within the vascular region. The pixel grayscale value at this projection point is read; this value represents the contrast agent concentration at that node at that moment. Performing this operation for each grid node and each time point yields a 3D matrix. The three dimensions of the matrix are the spatial node index and the time index, respectively, and the elements of the matrix are the concentration values. This 3D matrix represents the spatiotemporal distribution field of the contrast agent concentration.

[0089] Step S430: Based on the concentration gradient changes of the contrast agent concentration spatiotemporal distribution field in the bifurcation and tortuous regions of blood vessels, identify spatial regions in the initial three-dimensional vascular tree model where blood flow stagnation or eddy currents exist as candidate correction regions, and associate the spatial range of the candidate correction regions with the corresponding vascular segment identifiers.

[0090] In this embodiment, step S430 specifically includes steps S431 to S436: Step S431: Perform gradient analysis on the spatiotemporal distribution field of contrast agent concentration along the direction of the blood vessel centerline to determine the concentration change trend along the centerline direction, and mark the locations where local decreases or stagnations occur in the concentration change trend as suspected points of blood flow stagnation, and record the spatial coordinates and concentration values ​​of the suspected points of blood flow stagnation.

[0091] For each vessel segment in the initial 3D vascular tree model, extract its centerline trajectory curve. Along this curve, sample a series of points with small step sizes. For each sampling point, extract the concentration change curve over the entire time series from the contrast agent concentration spatiotemporal distribution field generated in step S420. Analyze the characteristics of this curve, especially the time period from zero to peak concentration. Calculate the rate of change of concentration over time at that point, i.e., the first derivative. If a point's rate of change during its corresponding concentration rise phase is significantly lower than that of its upstream neighboring points, or even negative, then that point is marked as a suspected point of blood flow stagnation. Record the 3D spatial coordinates of this suspected point, as well as its concentration value at the peak time and the time required to reach the peak.

[0092] Step S432: Perform circumferential distribution analysis on the spatiotemporal distribution field of contrast agent concentration in the direction of blood vessel cross section to determine the concentration difference at different angle positions on the blood vessel cross section contour. When the circumferential concentration difference exceeds the preset fluctuation threshold, mark the corresponding cross section area as a suspected eddy region and record the cross section position and concentration difference distribution of the suspected eddy region.

[0093] For each cross section in the initial 3D vascular tree model, the section is defined by the local projection plane generated in step S351 and the vascular cross section contour curve on it. Multiple points are sampled along the contour curve on this section, representing different angular positions around the vascular wall. For each point at an angular position, the concentration-time curve is extracted from the contrast agent concentration spatiotemporal distribution field generated in step S420. At a specific time point, such as the peak of contrast agent filling, the concentration values ​​at all angular positions are compared. The standard deviation of these concentrations, or the difference between the maximum and minimum values, is calculated. If this difference exceeds a preset fluctuation threshold, the contrast agent distribution on that section is considered extremely uneven, and the section is marked as a suspected eddy region. The location of the section (i.e., the corresponding centerline sampling point) and the concentration distribution data as a function of angle on that section are recorded.

[0094] Step S433: Map the spatial coordinates of suspected blood flow stagnation points and suspected eddy current regions back to the initial three-dimensional vascular tree model, extract the spatial trajectory curve segments of the vascular segments where these coordinate points are located and the surface mesh units of the lumen, and associate the extracted mesh units with the corresponding vascular segment identifiers.

[0095] For each suspected point of blood flow stagnation marked in step S431, its spatial coordinates are known. In the surface mesh of the initial 3D vascular tree model, the mesh node closest to these spatial coordinates is searched, and the vascular segment identifier to which this node belongs is recorded. Simultaneously, a spatial trajectory curve of this vascular segment near the suspected point is extracted, for example, a curve segment extending a certain length upstream and downstream of the suspected point, along with the corresponding lumen surface mesh element. For each suspected eddy region marked in step S432, its corresponding cross-sectional location is also known. Similarly, the vascular segment identifier to which this cross-sectional location belongs is found, and a spatial trajectory curve near this cross-section and its corresponding surface mesh element are extracted. Finally, all extracted mesh elements are associated with their respective vascular segment identifiers, forming a list of candidate correction regions to be processed.

[0096] Step S434: Based on the anatomical features of the bifurcation points and bends of blood vessels, the spatial trajectory curve segments are clustered and segmented. Multiple suspected points or suspected regions that are continuously distributed are merged into candidate correction regions with the same blood flow abnormality type. The clustering and segmentation are based on spatial continuity and consistency of blood flow abnormality type.

[0097] In this step, clustering suspected points or regions is performed because blood flow abnormalities are usually not isolated points, but rather continuous intervals. For example, a vortex region may cover a section of tortuous blood vessel. Merging multiple spatially continuous suspected points or regions with the same abnormality type (such as all being sluggish or all being vortices) into a large correction region facilitates overall geometric adjustments. Anatomical features, such as bifurcation points and bends, can serve as natural boundaries to aid in segmentation.

[0098] Based on the candidate correction region list generated in step S433, the following steps are performed: First, preliminary clustering is conducted based on spatial continuity: if two suspected points are spatially separated by a threshold and there are no obvious anatomical structures (such as bifurcation points) obstructing their path, they are considered to belong to the same continuous region. Then, further subdivision is performed based on blood flow abnormality type: if points within the same continuous region contain both sluggish suspected points and eddy current suspected regions, merging or splitting based on the primary type may be necessary. Finally, multiple spatially continuous entries with the same abnormality type are merged into a new, larger candidate correction region. A new identifier is generated for each merged region, and the index of the original suspected points or regions it contains is recorded.

[0099] Step S435: Generate a correction region descriptor for each candidate correction region, which includes the region boundary coordinates, region center coordinates, blood flow abnormality type identifier, and associated contrast agent concentration gradient value. This descriptor will serve as the target object for subsequent geometric deformation adjustment. The correction region descriptor will be stored as a correction region list.

[0100] For each candidate correction region after clustering and merging in step S434, perform the following operations: Region boundary coordinates: Take the minimum and maximum coordinate values ​​of all grid nodes within the region to form a bounding box, or more precisely, extract the coordinates of the start and end centerline points of the region on the vessel segment. Region center coordinates: Calculate the average of the coordinates of all grid nodes within the region, or take the midpoint of the region's centerline point. Blood flow abnormality type identifier: Set an enumeration value based on the dominant abnormality type within the region, such as "TYPE". STASIS "or "TYPE" VORTEX The associated developer concentration gradient values ​​are then extracted: for stagnant regions, the average concentration rise rate of that region can be extracted; for eddy regions, the average circumferential concentration difference value of that region can be extracted. This information is combined into a structure, namely the corrected region descriptor. All region descriptors are stored in a list as input for subsequent steps.

[0101] Step S436: Associate the modified region descriptor with the topology of the initial 3D vascular tree model, and establish a connection index between the candidate modified region and the adjacent vascular segment. The connection index is used to maintain the continuity of the region boundary during geometric deformation adjustment.

[0102] For each region in the correction region list, starting from the vascular segment identifiers covered by its spatial extent, the upstream and downstream connectivity of that vascular segment is queried within the preliminary 3D skeleton structure generated in step S340. For example, if a region is located in the middle segment of the left anterior descending artery (LAD), its upstream segment might be the proximal segment of the LAD, and its downstream segment might be the distal segment of the LAD or a bifurcation point. The identifiers of these two adjacent normal vascular segments that do not belong to the current correction region are recorded. This connectivity information, such as the upstream segment ID and downstream segment ID, is stored in the correction region descriptor for that region. Based on this, each correction region descriptor contains its own spatial definition, anomaly type, and its connection interface information with the surrounding normal model.

[0103] Step S440: Based on the blood flow velocity distribution curve, perform local geometric deformation adjustment on the surface mesh of the blood vessel lumen in the candidate correction region. By changing the spatial coordinates of the mesh nodes, the lumen diameter and bending angle of the adjusted region are matched with the measured data of the blood flow contrast agent propagation path, thereby generating the adjusted local mesh.

[0104] In this embodiment, step S440 specifically includes steps S441 to S446: Step S441: Extract the coordinates of the grid nodes on the surface of the blood vessel lumen in the candidate correction region. Based on the blood flow velocity distribution curve in this region, determine the degree of blood flow impact on each grid node and quantify the degree of blood flow impact as a node displacement influence factor.

[0105] For each grid node within the candidate correction region, firstly, based on the initial boundary conditions loaded in step S410 and a preliminary blood flow simulation performed in that region, the blood flow velocity vector near the node can be obtained. Then, the local curvature of the vessel at the node's location is calculated. According to fluid dynamics principles, in a tortuous vessel, the pressure on the outer wall is greater than that on the inner wall. Therefore, a function related to curvature can be defined. The node displacement influence factor is proportional to the magnitude of the blood flow velocity at the node and is related to the combination of the node's distance from the vessel's centerline and the local curvature. For example, the influence factor is set to a larger value for nodes located on the outer side of the bend, and a smaller value for nodes on the inner side. The influence factors for all nodes are normalized to between 0 and 1.

[0106] Step S442: Generate displacement driving parameters for mesh nodes based on blood flow impact factors, combine the displacement driving parameters with the boundary constraints of the candidate correction region, and determine the displacement of each mesh node in the radial and axial directions. The displacement includes the displacement direction and displacement magnitude.

[0107] For each grid node within the candidate correction region, first calculate its nearest distance to the vessel centerline and the tangent direction of the centerline at that point. The radial direction is defined as the direction from the centerline towards the node, and the axial direction is the tangent direction of the centerline. The displacement driving parameter is a preset, desired pipe diameter change ratio; for example, for a stagnant region, it is desired to increase the pipe diameter by 10%. Based on the influence factor obtained in step S441, a specific radial displacement can be assigned to the node, i.e., the driving parameter multiplied by the influence factor, and then multiplied by the original radial distance of the node. The displacement direction is radially outward. For nodes on the boundary, their displacement is forced to zero. For nodes near the boundary, their displacement is transitioned through a decay function to ensure a smooth transition with the boundary. The axial displacement is usually set to zero unless the bending angle needs to be adjusted.

[0108] Step S443: Update the grid node coordinates according to the displacement, so that the lumen diameter of the candidate correction region changes in the direction of increasing or decreasing, and at the same time adjust the curvature angle of the blood vessel centerline in the region to change the blood flow direction, and generate the updated grid node coordinates.

[0109] For each grid node within the candidate correction region, obtain its original coordinates P. old And the displacement vector D (containing radial and axial components) calculated in step S442. New coordinate P new equals P old Adding D. Performing this operation on all nodes yields an updated set of node coordinates. For regions where the bend angle needs adjustment, such as to reduce eddies, it may be necessary to move the bend vertices to one side to reduce the sharpness of the bend. This can be achieved by moving several key points on the centerline of the region, then recalculating the radial direction of all mesh nodes based on the new centerline, and applying the radial displacement again to generate a lumen surface that matches the new bend angle.

[0110] Step S444: Extract the measured contrast agent propagation path coordinates of the corresponding candidate correction area from the corrected image data set, compare the contrast agent propagation path coordinates regenerated by hemodynamic simulation in the adjusted area with the measured contrast agent propagation path coordinates, and calculate the coordinate deviation between the two.

[0111] First, from the corrected image dataset, for the current candidate correction region, a series of three-dimensional spatial points are obtained by tracing the leading edge position of the developer in each frame of the image. Connecting these points gives the measured propagation path L. obs Then, the hemodynamic simulation is run again on the adjusted local mesh generated in step S443. In the simulation, a large number of massless virtual particles are released at the inlet, and their trajectories are tracked. The trajectories of those particles that pass through the corrected region are extracted, averaged, and the simulated propagation path L is obtained. sim In L obs and L sim The same number of points are uniformly sampled, and then the average or maximum Euclidean distance between corresponding points is calculated as the coordinate deviation.

[0112] Step S445: When the coordinate deviation between the simulated propagation path and the measured propagation path exceeds the preset accuracy threshold, adjust the displacement weight coefficient in the displacement driving parameter according to the deviation direction, and re-update the mesh node displacement until the coordinate deviation between the simulated propagation path and the measured propagation path converges to the threshold range, and output the adjusted vascular lumen surface mesh.

[0113] For example, a precision threshold is set, such as a physical length equivalent to the image resolution. After calculating the coordinate deviation in step S444, if the deviation is greater than the threshold, analysis is performed. If L sim The whole is located at L obsIf the simulated path is too close to the inside, it indicates that the simulated blood flow is too fast, requiring an increase in vessel curvature or a decrease in diameter to slow it down, and vice versa. Accordingly, the displacement driving parameters in step S442 are adjusted. For example, if the simulated path is too close to the inside, it indicates insufficient expansion on the outer side of the bend, so the radial displacement weight coefficient of the outer nodes of the bend can be increased. After adjusting the parameters, return to step S442 to recalculate the displacement and update the mesh. Then, perform the blood flow simulation and deviation calculation again. Repeat this process until the deviation meets the threshold requirement. The mesh generated in the last iteration is the adjusted vessel lumen surface mesh.

[0114] Step S446: Evaluate the mesh quality of the adjusted vascular lumen surface mesh. When mesh distortion or self-intersection exists, locally re-subdivide the mesh nodes in the distorted area to generate the final adjusted local mesh.

[0115] Specifically, the adjusted vascular surface mesh output after the iteration convergence in step S445 undergoes the same mesh quality inspection process as in step S360. This involves detecting the interior angles, side length ratios, and self-intersections of triangles. For detected distorted triangular regions, a local mesh re-division algorithm is used for repair. During re-division, it is necessary to ensure that the newly generated triangles are smoothly connected to the surrounding unre-divisioned regions. After repair, the final adjusted local mesh, which satisfies both hemodynamic matching and has good mesh quality, is obtained.

[0116] Step S450: After each local geometric deformation adjustment, regenerate the spatiotemporal distribution field of developer concentration in the candidate correction area, and iteratively compare it with the measured developer distribution information extracted from the image data until the difference between the two converges to the preset error range, and output the adjusted local area model.

[0117] For the adjusted local mesh output in step S446, run the hemodynamic simulation again, but this time not only track the particle paths, but solve the contrast agent transport equation for the entire region to obtain the simulated concentration value C of each mesh node at each time point. sim(x,t) From the spatiotemporal distribution field of developer concentration generated in step S420, the measured concentration values ​​C of all nodes within the corresponding region are extracted. obs(x,t) Calculate the difference between the two, such as the root mean square error. Compare this error to a preset error range. If the error exceeds the range, adjust the displacement driving parameters in step S442 again and re-deform the mesh based on the spatial pattern of the error distribution (e.g., generally low simulated concentration in a certain area). Repeat this process until the concentration field difference across the entire region converges. The model generated in the last iteration is the adjusted local region model.

[0118] Step S460: Merge the adjusted local region model with the unadjusted part of the initial 3D vascular tree model, and smooth the transition of the fusion boundary to generate a complete corrected 3D vascular tree model.

[0119] First, all mesh nodes and triangular faces belonging to candidate correction regions are removed from the initial 3D vascular tree model, leaving an unadjusted part of the model with multiple holes. Then, each adjusted local region model generated in step S450 is embedded as a patch into the corresponding hole. Since the boundaries of the holes and patches are carefully designed to remain continuous (through the boundary constraints in step S436), they are essentially aligned. Next, for each pair of adjacent nodes on the boundary (one from the unadjusted part, one from the local patch), and nodes within a few layers near the boundary, smoothing is performed. A Laplace smoothing algorithm can be used, iteratively shifting the coordinates of each node towards the average position of its neighboring nodes within a ring-shaped region near the boundary, thereby eliminating any possible microsteps or creases. The smoothing process requires keeping nodes outside the boundary fixed to ensure the overall structure remains unchanged. After fusion and smoothing, the resulting complete mesh is the corrected 3D vascular tree model.

[0120] Step S500: Generate cardiovascular surgical planning guidance data based on the corrected three-dimensional vascular tree model, which includes the vascular lumen surface contour coordinate sequence and vascular branch connection relationship descriptor.

[0121] In this embodiment, step S500 specifically includes the following steps S510 to S560: Step S510: Smooth the surface mesh of the corrected 3D vascular tree model to eliminate mesh wrinkles and discontinuities caused by local geometric deformation adjustments, and generate a final 3D vascular tree model with a smooth surface profile. The final 3D vascular tree model is composed of the smoothed surface mesh.

[0122] In this embodiment, step S510 specifically includes steps S511 to S516: Step S511: Traverse the surface mesh nodes of the corrected 3D vascular tree model, extract the set of neighboring nodes for each mesh node, calculate the neighborhood centroid coordinates of the node based on the spatial coordinate distribution of the neighboring node set, and use them as the initial smoothing target position of the node.

[0123] For each vertex v on the surface mesh of the corrected 3D vascular tree model, obtain a list of all its first-order neighbor vertices, assuming there are N neighbors with coordinates p1, p2, ..., pN. Calculate the arithmetic mean of these neighbor coordinates, i.e., (p1 + p2 + ... + pN) divided by N, to obtain a new 3D point g. This point g is the centroid coordinate of vertex v's neighborhood, and also the initial target position of vertex v in this smoothing iteration.

[0124] Step S512: Based on the curvature change of the area where the surface mesh node is located, assign a smoothing weight coefficient to each mesh node. The smoothing weight coefficient is inversely proportional to the curvature of the area, so that the high curvature area retains its original geometric features and is not over-smoothed. Then, associate the smoothing weight coefficient with the node and store it.

[0125] In this step, a smoothing weight coefficient is assigned to preserve feature smoothness. Regions with large curvature changes, such as ridges at vessel bifurcation or the tops of vessel bends, are important anatomical features that need to be preserved. The smoothing weight coefficient is inversely proportional to the curvature, meaning that in flat regions (small curvature), the weight coefficient is large, allowing for greater smoothing movement; in high-curvature regions, the weight coefficient is small, almost disallowing movement, thus protecting these features from being smoothed away. For each mesh vertex v, its local curvature is first estimated. Curvature can be approximated by calculating the difference between the vertex's normal vector and the normal vectors of its neighboring vertices. Points with drastic changes in normal vectors have large curvature. Then, a smoothing weight coefficient w is defined. v This coefficient is the curvature k v Functions, such as w v =1 / (1+α*|k v |), where α is a positive constant controlling the strength of feature retention. Based on this, when k v When it is very large, w v Approaching 0; when k v When I was very young, w v Approximately 1. This calculated w v Stored in association with vertex v.

[0126] Step S513: The initial smoothing target position and the original coordinates of the grid nodes are weighted and fused. The fusion weight is controlled by the smoothing weight coefficient to generate the updated spatial coordinates of each grid node. The updated spatial coordinates are located between the original coordinates and the smoothing target position.

[0127] For vertex v, its original coordinates are P. old The initial smoothing target position calculated in step S511 is g, and the smoothing weight coefficient calculated in step S512 is w. v Then, the updated coordinates P of vertex v new =(1-w v )*Pold +w v *g. This formula performs linear interpolation from the original position to the target position, with the interpolation ratio controlled by a smoothing weighting coefficient.

[0128] Step S514: After performing the replacement operation of the updated spatial coordinates on all mesh nodes, re-evaluate the distance and angle between adjacent nodes in the surface mesh, detect whether there are abnormal cells with mesh distortion or flipping, and record the location and type of abnormal cells.

[0129] In this step, after performing coordinate replacement, the mesh topology remains unchanged, but the node positions change. This can cause some triangles to become excessively long (distorted) or even have their normals reversed (flipped). Therefore, a quality check is needed immediately to identify these abnormal triangular elements. This is a necessary step to ensure that the smoothing process does not compromise mesh validity. After applying the update from step S513 to all vertices, all triangular faces are traversed. For each triangle, the lengths of its three sides and its normal vector are calculated. If a side is too long or too short, or if the minimum interior angle of the triangle is too small, it is marked as a distorted element. If the angle between the normal vector of the updated triangle and the normal vector of the triangle before the update is too large, for example, exceeding 90 degrees, it is marked as a flipped element. The index and type of these abnormal elements are recorded.

[0130] Step S515: When an abnormal cell is detected, the mesh nodes in the local area where the abnormal cell is located are adjusted in a second iteration. Smoothing constraints are introduced to limit the displacement direction of the nodes until all mesh cells meet the preset geometric quality index, and the intermediate smoothed three-dimensional vascular tree model is output.

[0131] For the anomalous units detected in step S514, the set of vertices covered by these units is determined. For these vertices, instead of using the global result of step S513, a constrained smoothing process is applied. For example, a small step size factor can be set, and steps S511 to S513 can be repeated, moving only a small step each time, with a check for new anomalous units after each step. Simultaneously, constraints can be added, such as allowing vertices to move only within their original tangent plane, to suppress drastic changes in the normal. This process is repeated multiple times until all anomalous units are eliminated. The model at this point is the intermediately smoothed 3D vascular tree model.

[0132] Step S516: Perform an overall smoothness test on the surface mesh of the intermediate smoothed 3D vascular tree model. Confirm that the mesh smoothing effect meets the expected requirements by calculating the gradient of the change of the surface normal vector, and generate the final 3D vascular tree model.

[0133] For the smoothed 3D vascular tree model, all vertices are traversed. For each vertex, the average angle between its normal vector and the normal vectors of all its neighboring vertices is calculated. The average of this average angle across all vertices can be used as the smoothness index of the entire model. If this index is lower than a preset threshold, the smoothing effect is considered satisfactory. Otherwise, the parameter α in step S512 can be adjusted, and one or more rounds of global smoothing can be performed again starting from step S511. Finally, when the smoothness index meets the requirements, the model is output as the final 3D vascular tree model.

[0134] Step S520: Extract the centerline trajectory coordinates of each blood vessel segment in the final three-dimensional blood vessel tree model, sort the centerline trajectory coordinates according to the blood flow direction, generate the centerline coordinate sequence of each blood vessel segment, and associate and store the centerline coordinate sequence with the identifier of the corresponding blood vessel segment.

[0135] For the final 3D vascular tree model, a skeleton extraction algorithm based on distance transform or a Voronoi diagram method can be used to extract single-pixel-wide centerlines from a closed surface mesh. Since the model is continuous, the extracted centerlines are also continuous. Then, based on the model's topology, the continuous centerlines are segmented at bifurcation points to obtain centerline segments belonging to each independent vascular segment. For each segment, the points on the line are sorted according to the blood flow direction, i.e., from the cardiac side to the peripheral side. Finally, the sorted 3D point coordinate sequence, along with its corresponding vascular segment identifier, is stored in a data structure.

[0136] Step S530: Based on the connection relationship of the blood vessel bifurcation points in the final three-dimensional blood vessel tree model, construct a tree-like topological connection graph describing the parent-child hierarchical structure between each blood vessel segment, and encode the tree-like topological connection graph into a blood vessel branch connection relationship descriptor containing node identifiers and connection edges. The node identifier corresponds to the blood vessel segment, and the connection edge corresponds to the blood vessel bifurcation point.

[0137] In this embodiment, step S530 specifically includes the following steps S531 to S536: Step S531: Using the heart outlet or aortic root as the root node of the tree-like topology connection diagram, mark the vascular segments directly connected to the root node as first-level branch nodes, and assign a unique branch identifier to each first-level branch node. The branch identifier contains hierarchical information and sequence information.

[0138] In this step, the root node is established as the starting point of the entire tree diagram. The vascular segments directly connected to the root node are the first-level branches, equivalent to the trunk of the tree. Identifiers containing hierarchical and sequential information are assigned, such as "L1-1" representing the first vessel of the first level and "L1-2" representing the second vessel of the first level, so that the position of the vascular segment in the tree can be seen from the identifier alone. In the final 3D vascular tree model, the opening of the aortic root is defined as the root node. Then, in the centerline extracted in step S520, all vascular segment centerline segments originating from or connected to the root node region are found. The vascular segments corresponding to these segments are marked as first-level branch nodes. Identifiers are assigned to them, such as "L1-1" for the left main coronary artery and "L1-2" for the right coronary artery. These identifiers are stored in a mapping table.

[0139] Step S532: Traverse all the blood vessel bifurcation points in the final three-dimensional blood vessel tree model, identify the inflow blood vessel segment and at least two outflow blood vessel segments at each bifurcation point, mark the inflow blood vessel segment as the parent node, mark the outflow blood vessel segment as the child node, and record the blood vessel segment connection relationship at each bifurcation point.

[0140] For each bifurcation point in the model, the vascular segment converging at that point can be found from the centerline data in step S520. Based on the direction of the centerline (from proximal to distal), the unique vascular segment flowing into that bifurcation point, i.e., the parent node, and at least two vascular segments flowing out of that bifurcation point, i.e., the child nodes, can be identified. For example, at the bifurcation point of the left main coronary artery, the parent node is the left main coronary artery, and the child nodes are the left anterior descending artery and the left circumflex artery. These relationships are recorded to form a temporary relationship table.

[0141] Step S533: Establish directed connections from the parent node to each child node, record the starting node identifier and ending node identifier of each connection, and assign geometric angle parameters at the bifurcation point to each connection. The geometric angle parameters include the bifurcation angle and the spatial azimuth angle.

[0142] For each parent-child relationship recorded in step S532, such as parent node A and child node B, at the bifurcation point, the tangent direction of the centerline of vessel segment A and the tangent direction of the centerline of vessel segment B at the bifurcation point can be calculated. The angle between these two directions is the bifurcation angle. Simultaneously, a local coordinate system can be established, using the tangent direction of the parent vessel as a reference, to calculate the azimuth angle of the tangent direction of the child vessel relative to this reference. These angle parameters are stored together with the connecting edge (A->B).

[0143] Step S534: Traverse downwards level by level according to the blood vessel hierarchy, and repeat the identification of parent and child nodes and the establishment of connecting edges for all blood vessel segments to generate a hierarchical topological connection graph covering the entire blood vessel tree. The hierarchical topological connection graph contains complete information of all nodes and connecting edges.

[0144] Specifically, starting from the first-level branch node marked in step S531, for each first-level branch node, trace along its centerline towards the distal end until the first branch point is encountered. At this branch point, its child nodes (second-level branch nodes) are identified. Then, using these second-level branch nodes as new starting points, continue tracing towards the distal end until the next branch point is encountered, identifying the third-level branch nodes. This process is recursively repeated until all blood vessel segments have been traversed. During this process, at each branch point encountered, the directed edges from the parent node to each child node are recorded, and the relevant geometric angle parameters are stored. Ultimately, all nodes and directed edges constitute a complete hierarchical topological connection graph.

[0145] Step S535: Serialize all node identifiers and connection edge information in the hierarchical topology connection graph into text format, and append the storage address pointer of the corresponding vascular segment's geometric centerline coordinate sequence to each node identifier to generate a vascular branch connection relationship descriptor.

[0146] In this step, serialization converts the graph structure into a compact, storable, and transmittable format, such as JSON or XML. Appending a pointer to the storage address of the centerline coordinate sequence after the node identifier connects the topological and geometric information. Based on this, by parsing the descriptor, not only can the connection relationships of the blood vessels be determined, but the specific shape data of each blood vessel can also be found through the pointers. For example, a data structure can be created, such as a JSON object. This object contains an array of "nodes" and an array of "edges". Each element in the "nodes" array corresponds to a blood vessel segment node, containing an "id" field (i.e., the branch identifier) ​​and a "geometry" field. ref The "edges" array contains a "source" field (parent node ID), a "target" field (child node ID), and an optional "angle" field. Each element in the "edges" array corresponds to a directed edge and includes a "source" field (parent node ID), a "target" field (child node ID), and an optional "angle" field. params The field stores information such as the branching angle. Converting this JSON object into a text string yields the descriptor for the branching relationship of blood vessels.

[0147] Step S536: Perform an integrity check on the vascular branch connection relationship descriptor to ensure that the node identifier of each vascular segment appears in the connection relationship and that there are no circular references in the parent-child relationship. Correct any missing or circular references to generate the final vascular branch connection relationship descriptor.

[0148] Specifically, the descriptor text generated in step S535 is parsed to obtain a list of nodes and a list of edges. First, a set of all nodes is constructed. Then, all edges are traversed, and the source and target nodes of each edge are marked as "connected". After checking, it is checked whether any nodes have never been marked as "connected". If so, it means that the node is an isolated island and its connection edges need to be re-established according to the model topology. Second, a depth-first search algorithm is used to traverse the entire graph starting from the root node. If a node that has already been visited is visited, it means that a circular reference exists. For the discovered circular references, the connection relationships of the branch points in the circular region of the model need to be re-examined and corrected. After all corrections are completed, the graph is serialized again to generate the final blood vessel branch connection relationship descriptor.

[0149] Step S540: Extend the centerline coordinate sequence of each blood vessel segment along the centerline normal, extract the blood vessel lumen surface contour coordinates at the corresponding position of each centerline point, generate the blood vessel lumen surface contour coordinate sequence, and associate the blood vessel lumen surface contour coordinate sequence with the centerline coordinate sequence through index pointers to form a one-to-one mapping relationship.

[0150] For the centerline coordinate sequence of each vessel segment extracted in step S520, operations are performed along this sequence either with a fixed step size or directly on the existing centerline points. For each centerline point C i First, calculate the tangent direction at that point. Then, construct a plane perpendicular to the tangent. On this plane, find the point at which the tangent is located within the surface mesh of the final 3D vascular tree model, at a distance C. i The nearest set of points located near this plane. Sort and connect these point sets to form a closed contour line, denoted as C. i The surface contour of the blood vessel lumen at that location. Store the coordinates of all points on this contour as an array. Then, for C... i A mapping relationship is established with this contour array, for example, by associating the index of the contour array with the index i of the centerline point. Ultimately, for the entire blood vessel segment, a sequence of centerline points and its corresponding contour sequence are obtained.

[0151] Step S550: Encapsulate the vascular lumen surface contour coordinate sequence and vascular branch connection relationship descriptor into a structured data file, and attach the identifier and geometric dimension parameters of each vascular segment to the structured data file to generate cardiovascular surgical planning guidance data.

[0152] Specifically, an HDF5 file is created, containing different datasets. For example, a dataset named "centerlines" stores the centerline coordinate sequences of all vessel segments, organized by segment. A dataset named "contours" stores the surface contour coordinate sequences corresponding to all centerlines, associated with the centerlines via indexes. The vessel branch connection relationship descriptor generated in step S536 is appended as a text attribute to the root directory of the file. Simultaneously, the geometric parameters of each vessel segment are calculated, such as the average diameter calculated from the contour sequence and the length calculated from the centerline, and these parameters are stored as attributes of that segment's dataset. Based on this, the HDF5 file contains all the geometric and topological information required for surgical planning, i.e., it serves as guidance data for cardiovascular surgical planning.

[0153] Step S560: Perform format standardization verification on the cardiovascular surgery planning guidance data to ensure that the vascular lumen surface contour coordinate sequence and vascular branch connection relationship descriptor conform to the preset surgical navigation system input specifications.

[0154] First, obtain the input specification document for the target surgical navigation system. According to the specification, check whether the dataset name, data type, attribute format, etc., in the HDF5 file are consistent with the requirements. For example, the specification may require all coordinates to use double-precision floating-point numbers in millimeters; it may require the vascular branch connection descriptor to be in a specific version of JSON format, and the key names to be fixed. Based on these requirements, check each item in the data file generated in step S550. If any discrepancies are found, such as coordinate units being meters instead of millimeters, perform the corresponding unit conversion and data type adjustment. After adjustment, verify again until it fully complies with the specification. The data file that passes the verification is the final cardiovascular surgery planning and guidance data that can be delivered to the surgical navigation system.

[0155] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0156] Please refer to Figure 3 The diagram illustrates a block diagram of a three-dimensional reconstruction apparatus 100 according to an embodiment of this application. This apparatus has the function of implementing the aforementioned three-dimensional reconstruction method for anatomical structures applied in cardiovascular surgery. This function can be implemented in hardware or by hardware executing corresponding software. The apparatus can be a computer device or can be installed within a computer device. The apparatus 100 may include: a data acquisition module 110, a data correction module 120, a model generation module 130, a joint optimization module 140, and a data generation module 150.

[0157] The data acquisition module 110 is used to acquire the original image data sequence of the target cardiovascular region. The original image data sequence consists of two-dimensional angiography image units acquired at consecutive time points. Each two-dimensional angiography image unit contains the vascular tree projection contour information and blood flow contrast agent distribution density information at the corresponding time point. The data correction module 120 is used to perform spatiotemporal synchronization correction on the original image data sequence and generate a corrected image data set with time alignment markers based on the propagation delay time parameter of the blood flow contrast agent distribution density information between adjacent two-dimensional angiography image units. The model generation module 130 is used to expand the spatial dimension of the corrected image data set and generate an initial three-dimensional vascular tree model of the target cardiovascular region based on the geometric deformation characteristics of the vascular tree projection contour information under multiple preset projection angles. The joint optimization module 140 is used to jointly optimize the preset hemodynamic simulation parameters with the initial three-dimensional vascular tree model. By using the blood flow contrast agent distribution density information to determine the flow path distribution in the initial three-dimensional vascular tree model, the spatial topology of the initial three-dimensional vascular tree model is corrected to obtain the corrected three-dimensional vascular tree model. The data generation module 150 is used to generate cardiovascular surgical planning guidance data based on the modified three-dimensional vascular tree model, which includes a sequence of vascular lumen surface contour coordinates and a descriptor of vascular branch connection relationships.

[0158] It should be noted that the apparatus provided in the above embodiments is only illustrated by the division of the above functional modules when implementing its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0159] Please refer to Figure 4 This diagram illustrates a structural block diagram of a computer device 20 provided in one embodiment of this application. This computer device can be used to implement the functions of the aforementioned three-dimensional reconstruction method for anatomical structures applied in cardiovascular surgery. Specifically: Computer device 20 includes a central processing unit (CPU) 21, a system memory 24 including random access memory (RAM) 22 and read-only memory (ROM) 23, and a system bus 25 connecting the system memory 24 and the CPU 21. Computer device 20 also includes a basic input / output system (I / O system) 26 that facilitates information transfer between various devices within the computer, and a mass storage device 27 for storing the operating system 271.

[0160] The input / output system 26 may include a display for showing information and input devices such as a mouse and keyboard for user input. Both the display and the input devices are connected to the central processing unit 21 via an input / output controller connected to the system bus 25.

[0161] Mass storage device 27 is connected to central processing unit 21 via a mass storage controller (not shown) connected to system bus 25. Mass storage device 27 and its associated computer-readable media provide non-volatile storage for computer device 20. That is, mass storage device 27 may include computer-readable media (not shown) such as hard disk or CD-ROM (Compact Disc Read-Only Memory) drive.

[0162] Without loss of generality, computer-readable media can include computer storage media and communication media. Computer storage media includes volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes RAM, ROM, EPROM (Erasable Programmable Read Only Memory), EEPROM (Electrically Erasable Programmable Read Only Memory), flash memory or other solid-state storage devices, CD-ROM, DVD (Digital Video Disc) or other optical storage, magnetic tape cassettes, magnetic tape, disk storage, or other magnetic storage devices. Of course, those skilled in the art will recognize that computer storage media are not limited to the above-mentioned types. The system memory 24 and mass storage device 27 described above can be collectively referred to as memory.

[0163] According to various embodiments of this application, computer device 20 can also be connected to a remote computer on a network, such as the Internet, for operation. That is, computer device 20 can be connected to network 29 via network interface unit 28 connected to system bus 25, or network interface unit 28 can be used to connect to other types of networks or remote computer systems (not shown).

[0164] The memory also includes a computer program stored in the memory and configured to be executed by one or more processors to implement the above-described method for three-dimensional reconstruction of anatomical structures applied in cardiovascular surgery.

Claims

1. A method for three-dimensional reconstruction of anatomical structures applied in cardiovascular surgery, characterized in that, include: The original image data sequence of the target cardiovascular region is obtained. The original image data sequence consists of two-dimensional angiography image units acquired at consecutive time points. Each two-dimensional angiography image unit contains vascular tree projection contour information and blood flow contrast agent distribution density information at the corresponding time point. The original image data sequence is spatiotemporally synchronized and corrected. Based on the propagation delay time parameter of the blood flow contrast agent distribution density information between adjacent two-dimensional angiography image units, a corrected image data set with time alignment markers is generated. The spatial dimension of the corrected image data set is expanded, and an initial three-dimensional vascular tree model of the target cardiovascular region is generated based on the geometric deformation characteristics of the vascular tree projection contour information under multiple preset projection angles. The preset hemodynamic simulation parameters are jointly optimized with the initial three-dimensional vascular tree model. The spatial topology of the initial three-dimensional vascular tree model is corrected by using the blood flow contrast agent distribution density information to determine the flow path distribution in the initial three-dimensional vascular tree model, resulting in a corrected three-dimensional vascular tree model. Based on the modified 3D vascular tree model, cardiovascular surgical planning guidance data is generated, which includes a sequence of vascular lumen surface contour coordinates and descriptors of vascular branch connection relationships.

2. The method according to claim 1, characterized in that, The process of performing spatiotemporal synchronization correction on the original image data sequence involves generating a corrected image data set with time alignment markers based on the propagation delay time parameter of the blood flow contrast agent distribution density information between adjacent two-dimensional angiography image units. Extract the acquisition timestamp corresponding to each two-dimensional angiography image unit in the original image data sequence, determine the time interval distribution characteristics between adjacent image units according to the order of the acquisition timestamps, and store the time interval distribution characteristics as a time interval sequence; Correlation analysis is performed on the distribution density information of blood flow contrast agent in adjacent two-dimensional angiography image units. Based on the correlation analysis results, the propagation delay time parameter of blood flow contrast agent between adjacent time points is determined, and the propagation delay time parameter is associated and stored with the corresponding adjacent image units. Based on the time interval distribution characteristics and the propagation delay time parameter, a spatiotemporal mapping relationship is established, and each two-dimensional angiography image unit in the original image data sequence is mapped to a unified time reference coordinate system, generating the mapped time coordinates of each image unit in the unified time reference coordinate system. Based on the spatiotemporal mapping relationship, the spatial position coordinates of the two-dimensional angiography image unit are offset and corrected to eliminate the position offset of the vascular tree projection contour caused by physiological motion, and an intermediate corrected image set with time alignment mark is generated. The intermediate corrected image set contains two-dimensional angiography image units with corrected spatial position coordinates. The blood flow contrast agent distribution density information of each two-dimensional angiography image unit in the intermediate correction image set is continuously adjusted to generate a transition correction image set with a smooth density gradient. The contour consistency of the vascular tree projection contour information of each two-dimensional angiography image unit in the transition correction image set is checked. When there is contour breakage or overlap abnormality, interpolation repair is performed based on the contour information of adjacent image units to obtain a complete set of corrected image data.

3. The method according to claim 2, characterized in that, The process involves performing correlation analysis on the distribution density information of blood flow contrast agent in adjacent two-dimensional angiography image units, determining the propagation delay time parameter of blood flow contrast agent between adjacent time points based on the correlation analysis results, and associating and storing the propagation delay time parameter with the corresponding adjacent image units, including: Extract the blood flow contrast agent distribution density subset corresponding to the vascular tree projection contour region in adjacent two-dimensional angiography image units to form a first density distribution sequence and a second density distribution sequence. The first density distribution sequence corresponds to the previous image unit, and the second density distribution sequence corresponds to the next image unit. Cross-correlation is performed on the first density distribution sequence and the second density distribution sequence to generate a cross-correlation function curve, which represents the similarity at different time offsets, and the sequence of ordinate values ​​of the cross-correlation function curve is recorded. Determine the peak position of the cross-correlation function curve, use the time offset corresponding to the peak position as the candidate propagation delay time parameter, and record the ordinate value of the peak position as the correlation confidence level. Based on the ratio between the candidate propagation delay time parameter and the difference in acquisition timestamps of adjacent image units, the rationality of the candidate propagation delay time parameter is verified. When the ratio exceeds the preset range, the candidate propagation delay time parameter is weighted and adjusted based on the correlation confidence level so that the adjusted parameter falls within the preset range. The verified candidate propagation delay time parameter is weighted and averaged with other candidate parameters calculated from multiple groups of adjacent image units to generate the final propagation delay time parameter. The weight of the weighted average is determined based on the correlation confidence of each group. The final propagation delay time parameter is associated with and stored with the timestamp information of the corresponding adjacent image units, and a mapping table between the propagation delay time parameter and the image unit index is established.

4. The method according to claim 2, characterized in that, The step of establishing a spatiotemporal mapping relationship based on the time interval distribution characteristics and the propagation delay time parameter, mapping each two-dimensional angiography image unit in the original image data sequence to a unified time reference coordinate system, and generating the mapped time coordinates of each image unit in the unified time reference coordinate system includes: Using the acquisition timestamp of the first two-dimensional angiography image unit in the original image data sequence as the time reference origin, the time intervals in the time interval distribution feature are accumulated with the time reference origin to generate a reference time coordinate sequence, which contains the reference time coordinates of each image unit. Based on the deviation between the propagation delay time parameter and the corresponding time point in the reference time coordinate sequence, the offset correction amount of each two-dimensional angiography image unit on the time axis is determined, and the offset correction amount is associated with the corresponding image unit; Based on the offset correction amount, the reference time coordinate sequence is locally adjusted to generate a corrected time coordinate sequence, so that the propagation delay of blood flow contrast agent distribution density information between adjacent image units matches the actual physiological flow process. The vascular tree projection contour information of each two-dimensional angiography image unit is associated and stored with the corresponding time coordinates in the corrected time coordinate sequence to generate a spatiotemporally aligned data unit with a unified time reference. The spatiotemporal aligned data units are arranged in the order of the corrected time coordinate sequence to construct the basic elements of the intermediate corrected image set; The continuity of the corrected time coordinates in the spatiotemporal aligned data unit is verified. When there is an anomaly of time coordinate inversion or jump, the time coordinates of the abnormal area are re-interpolated and adjusted to generate a set of corrected intermediate images.

5. The method according to claim 1, characterized in that, The step of spatially expanding the corrected image data set and generating an initial three-dimensional vascular tree model of the target cardiovascular region based on the geometric deformation characteristics of the vascular tree projection contour information under multiple preset projection angles includes: Select at least two two-dimensional angiography image units acquired from different projection angles from the corrected image data set, extract the boundary curve coordinates and the vessel centerline trajectory coordinates of the vessel tree projection contour in each image unit, and classify and store the boundary curve coordinates and the vessel centerline trajectory coordinates according to the projection angle. Spatial intersection matching is performed on the coordinates of the blood vessel centerline trajectory. Based on the geometric perspective relationship under different projection angles, the intersection position of the coordinates of the blood vessel centerline trajectory in three-dimensional space is determined, an initial spatial coordinate set of blood vessel bifurcation points is generated, and the initial spatial coordinate set is associated with the corresponding blood vessel bifurcation point identifier. Based on the initial set of spatial coordinates of the blood vessel bifurcation points and the boundary curve coordinates of the blood vessel tree projection contour, the spatial extension direction of the blood vessel segment is determined, and the blood vessel segments between adjacent blood vessel bifurcation points are curve fitted according to the spatial extension direction to generate the spatial trajectory curve of the blood vessel segment. The spatial trajectory curve contains the three-dimensional coordinates of each point on the curve. The spatial trajectory curves of the blood vessel segments are spliced ​​together according to the blood vessel branch connection order recorded in the blood vessel tree projection contour information to form a preliminary three-dimensional skeleton structure containing the topological connection relationship of blood vessels. The preliminary three-dimensional skeleton structure is composed of the spatial trajectory curves of the blood vessel segments and the spatial coordinates of the blood vessel bifurcation points. The surface contour of the preliminary three-dimensional skeleton structure is reconstructed, and the boundary curve coordinates of the vascular tree projection contour are mapped to the vertical plane of the spatial trajectory curve of the vascular segment. A three-dimensional lumen surface mesh that wraps the spatial trajectory curve is generated by contour stacking to obtain the initial three-dimensional vascular tree model. The mesh quality of the three-dimensional lumen surface mesh in the initial three-dimensional vascular tree model is checked. When the mesh is found to have distortion or self-intersection regions, the mesh nodes in the corresponding regions are rearranged to generate an optimized initial three-dimensional vascular tree model. The optimized model is then compared and verified with the preliminary three-dimensional skeleton structure.

6. The method according to claim 5, characterized in that, The step involves performing spatial intersection matching on the coordinates of the vessel centerline trajectory, determining the intersection points of the vessel centerline trajectory coordinates in three-dimensional space based on the geometric perspective relationships under different projection angles, generating an initial set of spatial coordinates for vessel bifurcation points, and associating the initial set of spatial coordinates with the corresponding vessel bifurcation point identifiers, including: Extract the trajectory coordinates of the first blood vessel centerline in the two-dimensional angiography image unit under the first projection angle, extract the trajectory coordinates of the second blood vessel centerline in the two-dimensional angiography image unit under the second projection angle, and store the trajectory coordinates of the first blood vessel centerline and the second blood vessel centerline as the first coordinate set and the second coordinate set, respectively. Based on the imaging geometric parameters corresponding to the first projection angle and the second projection angle, the trajectory coordinates of the first blood vessel centerline and the trajectory coordinates of the second blood vessel centerline are converted into a first set of projection rays and a second set of projection rays in three-dimensional space, respectively. Each ray in the first set of projection rays and the second set of projection rays is defined by a starting point and a direction vector. Determine the shortest spatial distance between each ray in the first set of projection rays and the second set of projection rays. When the shortest spatial distance is less than a preset matching tolerance threshold, determine the midpoint of the corresponding two rays as the candidate intersection point spatial coordinates, and associate and store the candidate intersection point spatial coordinates with the corresponding two ray identifiers. The spatial coordinates of the candidate intersection points are clustered and merged. Based on the topological connectivity of the blood vessel bifurcation region in the blood vessel tree projection contour information, the spatial coordinates of multiple candidate intersection points belonging to the same blood vessel bifurcation point are merged into initial spatial coordinates representing the unique position of the bifurcation point. The merging process is based on the spatial distance between candidate intersection points and the blood vessel topological connectivity. The initial spatial coordinates are sorted according to the order of the bifurcation points recorded in the vascular tree projection contour information, and the identifier of the connecting vascular segment corresponding to each bifurcation point is associated and stored to generate the initial spatial coordinate set of the vascular bifurcation points. The initial spatial coordinate set contains the spatial coordinates and connection relationship information of each bifurcation point. The process of reconstructing the surface contour of the preliminary three-dimensional skeleton structure involves mapping the boundary curve coordinates of the vascular tree projection contour to the vertical plane of the spatial trajectory curve of the vascular segment, and generating a three-dimensional lumen surface mesh that encloses the spatial trajectory curve through contour stacking to obtain the initial three-dimensional vascular tree model, including: Multiple sampling points are extracted along the spatial trajectory curve of the blood vessel segment with a fixed sampling step size. At each sampling point, a local projection plane perpendicular to the tangent direction of the spatial trajectory curve is constructed, and the normal vector of the local projection plane is aligned with the tangent direction of the spatial trajectory curve and stored. Two-dimensional angiography image units corresponding to the time of each sampling point are extracted from the corrected image data set. The boundary curve coordinates of the vascular tree projection contour in the two-dimensional angiography image unit are mapped onto the local projection plane through inverse projection transformation to generate the vascular cross-section contour curve at the sampling point. The vascular cross-section contour curve is composed of a closed contour point sequence. Contour point correspondence matching is performed on the blood vessel cross-sectional contour curves at adjacent sampling points. Based on the geometric similarity of the contour curves and the continuity constraint of the blood vessel centerline, the vertex connection relationship between adjacent cross-sectional contour curves is established. The vertex connection relationship is recorded through a connection edge index table. Based on the vertex connection relationship, triangular patches are generated to connect adjacent cross-sectional contour curves. The triangular patches are then continuously spliced ​​along the spatial trajectory curve of the blood vessel segment to form a three-dimensional lumen surface mesh unit covering the entire blood vessel segment. The three-dimensional lumen surface mesh unit is composed of a set of triangular patches. The three-dimensional lumen surface mesh units corresponding to different blood vessel segments are sutured at the blood vessel bifurcation point to eliminate gaps and overlapping areas between mesh units, thereby generating a continuous and closed three-dimensional lumen surface mesh. The suture process is based on contour matching and vertex merging at the bifurcation point. The surface smoothness of the three-dimensional lumen surface mesh is evaluated. When there are local uneven areas on the surface, the mesh nodes in that area are relaxed and adjusted to make the surface curvature change more gradual, thus generating the initial three-dimensional vascular tree model.

7. The method according to claim 1, characterized in that, The step involves jointly optimizing the preset hemodynamic simulation parameters with the initial three-dimensional vascular tree model. By using the blood flow contrast agent distribution density information to determine the flow path distribution within the initial three-dimensional vascular tree model, the spatial topology of the initial three-dimensional vascular tree model is corrected to obtain a corrected three-dimensional vascular tree model. This includes: The blood flow velocity distribution curve and blood flow pressure gradient parameters in the hemodynamic simulation parameters are loaded onto the vascular lumen surface mesh nodes of the initial three-dimensional vascular tree model to establish the initial boundary conditions for blood flow. The initial boundary conditions include the blood flow velocity vector and pressure value at each mesh node. Blood flow contrast agent distribution density information at continuous time points is extracted from the corrected image data set, and the blood flow contrast agent distribution density information is mapped to the corresponding spatial location of the initial three-dimensional vascular tree model to generate a contrast agent concentration spatiotemporal distribution field, which contains the concentration value of each spatial location at different time points. Based on the concentration gradient changes of the contrast agent concentration spatiotemporal distribution field in the bifurcation region and the tortuous region of the blood vessel, spatial regions in the initial three-dimensional blood vessel tree model that have blood flow stagnation or eddy current phenomena are identified as candidate correction regions, and the spatial range of the candidate correction regions is associated with the corresponding blood vessel segment identifiers. Based on the blood flow velocity distribution curve, the local geometric deformation of the vascular lumen surface mesh in the candidate correction region is adjusted. By changing the spatial coordinates of the mesh nodes, the lumen diameter and bending angle of the adjusted region are matched with the measured data of the blood flow contrast agent propagation path, thereby generating the adjusted local mesh. After each local geometric deformation adjustment, the spatiotemporal distribution field of developer concentration in the candidate correction area is regenerated and iteratively compared with the measured developer distribution information extracted from the image data until the difference between the two converges to the preset error range, and the adjusted local area model is output. The adjusted local region model is fused with the unadjusted portion of the initial 3D vascular tree model, and the fusion boundary is smoothly transitioned to generate a complete corrected 3D vascular tree model.

8. The method according to claim 7, characterized in that, The process involves identifying spatial regions in the initial three-dimensional vascular tree model exhibiting slowed blood flow or eddy currents based on the concentration gradient changes of the contrast agent concentration spatiotemporal distribution field in the vascular bifurcation and tortuous regions. These regions are then used as candidate correction regions, and their spatial extent is associated with corresponding vascular segment identifiers. This includes: Gradient analysis was performed on the spatiotemporal distribution field of the contrast agent concentration along the direction of the blood vessel centerline to determine the concentration change trend along the centerline direction. The locations where the concentration change trend showed local decreases or stagnation were marked as suspected points of blood flow stagnation. The spatial coordinates and concentration values ​​of the suspected points of blood flow stagnation were recorded. The spatiotemporal distribution field of the contrast agent concentration is analyzed in a circumferential direction along the cross-section of the blood vessel to determine the concentration difference at different angles on the blood vessel cross-section profile. When the circumferential concentration difference exceeds a preset fluctuation threshold, the corresponding cross-sectional area is marked as a suspected eddy region, and the cross-sectional position and concentration difference distribution of the suspected eddy region are recorded. The spatial coordinates of the suspected blood flow stagnation points and the suspected eddy current regions are mapped back to the initial three-dimensional vascular tree model. The spatial trajectory curve segments and lumen surface mesh units of the vascular segments where these coordinate points are located are extracted, and the extracted mesh units are associated with the corresponding vascular segment identifiers. Based on the anatomical features of vascular bifurcation points and vascular bend points, the spatial trajectory curve segments are clustered and segmented, and multiple suspected points or suspected regions that are continuously distributed are merged into candidate correction regions with the same blood flow abnormality type. The clustering and segmentation is based on spatial continuity and consistency of blood flow abnormality type. For each candidate correction region, a correction region descriptor is generated, which includes the region boundary coordinates, region center coordinates, blood flow abnormality type identifier, and associated contrast agent concentration gradient value. The correction region descriptors are then stored as a correction region list. The modified region descriptor is associated with the topology of the initial three-dimensional vascular tree model to establish a connection index between the candidate modified region and adjacent vascular segments. The connection index is used to maintain the continuity of the region boundary during geometric deformation adjustment.

9. A three-dimensional reconstruction device, characterized in that, include: The data acquisition module is used to acquire the original image data sequence of the target cardiovascular region. The original image data sequence consists of two-dimensional angiography image units acquired at consecutive time points. Each two-dimensional angiography image unit contains the vascular tree projection contour information and blood flow contrast agent distribution density information at the corresponding time point. The data correction module is used to perform spatiotemporal synchronization correction on the original image data sequence. Based on the propagation delay time parameter of the blood flow contrast agent distribution density information between adjacent two-dimensional angiography image units, it generates a corrected image data set with time alignment markers. The model generation module is used to expand the spatial dimension of the corrected image data set and generate an initial three-dimensional vascular tree model of the target cardiovascular region based on the geometric deformation characteristics of the vascular tree projection contour information under multiple preset projection angles. The joint optimization module is used to jointly optimize the preset hemodynamic simulation parameters with the initial three-dimensional vascular tree model. By using the blood flow contrast agent distribution density information to determine the flow path distribution in the initial three-dimensional vascular tree model, the spatial topology of the initial three-dimensional vascular tree model is corrected to obtain the corrected three-dimensional vascular tree model. The data generation module is used to generate cardiovascular surgical planning guidance data based on the modified three-dimensional vascular tree model, which includes a sequence of vascular lumen surface contour coordinates and a descriptor of vascular branch connection relationships.

10. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing a computer program that is loaded and executed by the processor to implement the three-dimensional reconstruction method for anatomical structures applied in cardiovascular surgery as described in any one of claims 1 to 9.

Images