A method for rapid reconstruction of variant anatomy of hepatic artery and recommendation of access point

Abstract

The present application belongs to the technical field of medical image guided interventional therapy planning, and proposes a method for rapid reconstruction of hepatic artery variation anatomy and recommendation of interventional access points. The method comprises the following steps: first, data preprocessing and hepatic artery candidate region positioning are performed based on multi-phase enhanced images and patient information; a hepatic artery tree model with semantic labels is constructed through three-dimensional blood vessel segmentation, centerline tracking and matching with anatomical prior rules; the feeding branch is identified in combination with lesion location, the target point is encoded, and the geometric characteristics of each candidate path are searched and calculated; finally, the path is scored by multi-index normalization weighting and three-dimensional visualization rendering to generate a planning scheme containing recommended interventional points and termination points. The present application realizes automatic and accurate reconstruction of hepatic artery variation structure and quantitative optimization of interventional path, effectively improving the accuracy and efficiency of surgical planning.

Description

Technical Field

[0001] This invention belongs to the field of medical image-guided interventional treatment planning technology, and particularly relates to a method for rapid reconstruction of hepatic artery variation anatomy and recommendation of interventional site points. Background Technology

[0002] In the field of preoperative planning technology for medical image-guided interventional therapy, existing solutions for hepatic artery anatomical reconstruction and intervention point recommendation driven by multi-phase enhanced hepatic artery images and patient basic information usually rely on manual interpretation of angiographic images combined with empirical lesion spatial location estimation and empirical selection of interventional entry points, or consist of a candidate path search module that only considers a single indicator of the vascular tree and a static three-dimensional display interface. These solutions have limitations such as difficulty in stably obtaining the mask structure of the candidate hepatic artery region and the anatomical annotation structure of the hepatic artery from multi-phase enhanced hepatic artery images, lack of systematic modeling that maps lesion spatial location data and the set of candidate interventional entry points to the interventional path candidate set structure, and lack of a unified and objective evaluation mechanism for various geometric features of different candidate paths. Existing methods often rely on the operator's subjective comparison between the hepatic artery tree centerline and the spatial location data of the lesion, or on a fixed preference for prioritizing the total path length to complete the candidate path search. In preoperative planning scenarios where the anatomy of the hepatic artery is complex, the spatial location data of the lesion is diverse, and the candidate interventional entry set is constrained by anatomical and operational conditions, the selection of intervention point and termination point information is highly dependent on experience and there are significant differences between different operators. Furthermore, it is difficult to maintain consistency between the interventional path candidate set structure and the superimposed rendering results of the 3D model across multiple examinations and multiple systems. This makes it difficult to achieve a stable realization of the interventional planning output structure based on the hepatic artery anatomical annotation structure and the interventional path candidate set structure. Regarding the joint processing of hepatic artery anatomical annotation structures, interventional pathway candidate sets, and intervention point and termination point information, existing technologies generally lack unified configurations in the path multi-indicator normalization and weighted scoring ranking stages. Furthermore, they lack dynamic annotation and path marking mechanisms corresponding to the hepatic artery anatomical annotation structures in the 3D model overlay and rendering stage. This makes it difficult to establish a consistent process for the acquisition, alignment, determination, and recording of interventional pathway candidate sets and output structures in collaborative application scenarios for interventional treatment planning and intraoperative navigation in patients with hepatic artery variations. This process starts from multi-phase enhanced hepatic artery phase images and basic patient information, constructs the interventional pathway candidate set structure through the hepatic artery candidate region mask structure and the hepatic artery anatomical annotation structure, and outputs the interventional planning output structure. Consequently, there is a lack of integrated modeling and reliable transmission between interventional pathway multi-indicator information, anatomical semantic information, and intervention point and termination point information, increasing the burden of preoperative discussions and intraoperative adjustments, and failing to support the actual clinical need for reproducible generation of interventional planning output structures. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a method for rapid reconstruction of hepatic artery variant anatomy and recommendation of interventional site points, comprising:

[0004] We acquire multi-phase enhanced hepatic artery phase images and patient basic information, perform format-unified registration, noise suppression, liver organ segmentation and hepatic artery candidate region localization processing, and generate hepatic artery candidate region mask structure.

[0005] Based on the candidate region mask structure of the hepatic artery, three-dimensional vessel segmentation, centerline tracing and anatomical prior classification annotation are performed to generate a hepatic artery anatomical annotation structure with variant labels and key branch annotations.

[0006] Based on the hepatic artery anatomical annotation structure with variant labels and key branch annotations, the system performs lesion blood supply branch identification, target encoding and candidate path search and geometric feature calculation and processing to generate an interventional path candidate set structure containing multiple geometric indicators.

[0007] Based on the candidate set structure of intervention paths, the system performs multi-indicator normalization, weighted scoring and sorting, and three-dimensional model overlay rendering operations to generate an intervention planning output structure containing information on intervention points and termination points.

[0008] Furthermore, the process of standardizing and registering formats also includes:

[0009] The format-unified registration process includes normalizing the differences between the arterial phase sequence and the portal venous phase sequence in terms of matrix size, voxel spacing, and coordinate origin; performing three-dimensional resampling to interpolate the original volume data onto a unified voxel grid to obtain multi-phase three-dimensional volume data with consistent scale; and mapping the portal venous phase sequence to the reference coordinate system through rigid or non-rigid spatial transformation, using the coordinate system of the arterial phase sequence as the registration reference.

[0010] Furthermore, the process of liver organ division also includes:

[0011] The liver organ segmentation process involves inferring the range of liver grayscale values ​​and abdominal spatial range based on the patient's gender and body weight information in the basic patient information, combined with the prior of the enhanced patterns in the arterial and portal venous phases. A pre-trained organ segmentation model or probabilistic segmentation model is used to predict the class probability value for each voxel, generating an initial liver region mask. Three-dimensional connected component analysis is performed to remove isolated regions that are too small and far from the typical location of the liver, and the boundaries are corrected based on the prior of liver morphology. Local threshold refinement is also performed based on the higher contrast between the hepatic veins and the main portal vein in the portal venous phase data.

[0012] Furthermore, the process of localizing the candidate hepatic artery region also includes:

[0013] The hepatic artery candidate region localization process includes extracting the liver surface contour based on the liver region mask and estimating the approximate location of the hepatic hilum in three-dimensional space. A certain thickness spatial range around the hepatic hilum is defined as the key analysis region. Within the key analysis region, the tubular structure response value of each system is calculated to evaluate the local gray-scale distribution directionality and curvature characteristics. Threshold segmentation is performed to obtain an initial set of hepatic artery candidate voxels. Multi-phase enhancement feature screening is performed to mark candidate voxels with significant arterial phase enhancement and relatively weak portal venous phase enhancement as hepatic artery features and to remove candidate voxels that do not conform to the vascular enhancement pattern.

[0014] Furthermore, the process of three-dimensional blood vessel segmentation also includes:

[0015] The three-dimensional vessel segmentation process involves cropping standardized multi-phase enhanced image data within the spatial range defined by the mask structure of the hepatic artery candidate region to obtain local three-dimensional volume data. Vessel enhancement filtering is performed to enhance the response of slender tubular structures and suppress the response of plate-like or block-like structures by analyzing the local gray-level gradient direction and principal curvature characteristics. A set of intensity thresholds and connectivity conditions are selected on the vessel response intensity field to perform three-dimensional threshold segmentation and region growing. Voxels that meet the intensity conditions and are spatially connected are aggregated into vessel candidate regions. Isolated small clumps or structurally abnormal short branches are screened through connected domain analysis and volume statistics to remove clumps that do not conform to the structural characteristics of the hepatic artery tree, generating a three-dimensional vessel segmentation voxel structure.

[0016] Furthermore, the centerline tracking process also includes:

[0017] The centerline tracing process involves performing morphological contraction operations on the segmented voxel structure of blood vessels on a 3D mesh to generate a skeleton structure with a voxel width, obtaining a set of candidate center voxels. An undirected graph structure is constructed according to spatial connectivity. Seed nodes are selected from near the origin region of the hepatic artery. A centerline tracing strategy based on shortest path search is used to gradually expand from the seed nodes along the direction with smaller edge weights until the terminal region is reached. When a node with a degree greater than two is encountered, it is marked as a blood vessel bifurcation point, and the start point, end point, length, and average radius information of each path are recorded simultaneously. Paths with a length significantly shorter than a preset threshold and endpoints close to the segmentation edge are marked as suspicious pseudo-branches. The hepatic artery tree centerline graph structure is then constructed.

[0018] Furthermore, the process of anatomy prior typing annotation also includes:

[0019] The anatomical prior classification and labeling process includes performing topology cleaning and trunk identification on the hepatic artery tree centerline map structure. Based on the attributes of vessel radius, path length, and branch angle, short branches and abnormally acute-angled branches are screened and deleted or merged. Branches that do not conform to the continuous course characteristics of the hepatic artery tree are deleted or merged. Starting from the node near the hepatic hilum, the main hepatic artery, proper hepatic artery, and the main segments extending to the left and right lobes are traced along a path with a large radius and stable direction to determine the main trunk path set. Secondary branches are initially grouped according to the location and course direction of the branches. Based on the preset set of hepatic artery anatomical prior rules, the trunk path set and secondary branch set in the hepatic artery tree topology description structure are matched with the anatomical prior rules. The hepatic artery variation type is determined by comparing the relative position of the branch origin node in three-dimensional space, the number of branches, and the course direction. Clinically significant anatomical names are assigned to each edge and node in the topology map, generating a hepatic artery anatomical labeling structure with variation labels and key branch annotations.

[0020] Furthermore, the process of identifying the blood supply branches to the lesion also includes:

[0021] The lesion blood supply branch identification and processing involves extracting the coordinates of the lesion center point and several sampling points representing the lesion surface from the lesion spatial location data for each lesion instance. In a unified three-dimensional coordinate system, the coordinates are projected onto the set of centerlines in the hepatic artery anatomical annotation structure. By finding the nearest centerline point, the corresponding vascular branch identifier and hierarchical information are recorded. The matching frequency and spatial distance distribution between all sampling points in the lesion and each branch are statistically weighted. Branches with more matching frequency and smaller average distance are marked as high-confidence blood supply branches. With reference to the liver segment attribution information, lesions that are clearly located in a certain liver segment are preferentially matched with the corresponding liver segment artery to generate a set of lesion blood supply branches.

[0022] Furthermore, the target encoding process also includes:

[0023] The target encoding process involves extracting the centerline point sequence and radius information of the corresponding vascular branch from the hepatic artery anatomical annotation structure for each candidate blood supply branch record in the lesion blood supply branch set. The centerline point with the smallest distance from the lesion center point is found along the centerline of the vascular branch as the main target point location. For cases where the lesion volume is large or extends to multiple sub-regions, multiple candidate target points are detected at the intersection of the lesion volume mask and the vascular branch centerline. A unique target point number is generated for each target point, along with local vascular radius, local curvature features, and distance attributes from the geometric center of the lesion, thus generating a target point encoding structure.

[0024] Furthermore, the process of candidate path search and geometric feature calculation also includes:

[0025] The candidate path search process involves extracting the vessel centerline map from the anatomy annotation structure of the hepatic artery to construct a structural path model. Each centerline node is treated as a node in the graph, and length and radius attributes are added to the edges. Based on the entry point definition in the candidate interventional entry point set, the mapping position of each interventional entry point is found in the vessel graph structure, and the corresponding entry point node is added as a connection edge between the existing vessel node and the existing vessel node. Each target point in the target point encoding structure is treated as a target node. A path search task is constructed for each candidate interventional entry point and each target point combination. The shortest path or minimum cost path search strategy based on the graph structure is executed. The search cost considers the vessel segment length, vessel radius, and local branch angle, and a penalty term is added for vessels with too small a radius or too large a bend. High-risk paths and unreachable combinations are marked, and a candidate path set is generated.

[0026] The geometric feature calculation process includes accumulating the length of the vessel segment segment by segment according to the path node sequence for each candidate path to obtain the total length index, calculating the turning angle of all turning points in the path to obtain the curvature index and the maximum turning angle index, recording the minimum radius value traversed by the path as the narrowest diameter index of the path, and calculating the stenosis distribution index and the branch complexity index to generate an interventional path candidate set structure containing multiple geometric indices.

[0027] The key innovations of this invention include:

[0028] (1) Based on the anatomical annotation structure of the hepatic artery, the spatial location data of the lesion and the set of candidate interventional entry points, the identification of the blood supply branches of the lesion, the target encoding and candidate path search and geometric feature calculation are performed to generate an interventional path candidate set structure containing multiple geometric indicators.

[0029] (2) Based on the candidate set structure of intervention path, perform path multi-indicator normalization and weighted scoring sorting processing on the total path length, path curvature, maximum turning angle, path narrowest pipe diameter, narrow section distribution information, branch complexity index and risk labeling fields to generate intervention path sorting result set.

[0030] (3) Based on the candidate set structure of intervention path and the anatomy annotation structure of hepatic artery, perform three-dimensional model overlay rendering operation, overlay the candidate path trajectory with the hepatic artery tree centerline and key branch names to generate an intervention planning output structure containing intervention point and termination point information.

[0031] The following are its main beneficial effects:

[0032] (1) The above-mentioned technical solution for constructing an interventional path candidate set structure based on the anatomical annotation structure of the hepatic artery, the spatial location data of the lesion, and the candidate interventional entry set organizes the identification of the blood supply branch of the lesion, the target encoding, and the candidate path search results into a path record containing multiple geometric indicators under a unified spatial coordinate system. This allows each candidate path to be associated with both the spatial location data of the lesion and the candidate interventional entry set. Compared with the existing solutions that rely on manual interpretation of angiographic images and empirical path selection, this solution can form an interventional path candidate set structure with a complete structure and clear fields in the case of hepatic artery variations and complex lesion distribution. This facilitates subsequent automated evaluation and reproduction of the candidate path set for the same lesion.

[0033] (2) The above-mentioned technical solution of normalizing and weighted scoring of path multi-indicator ranking for total path length, path curvature, maximum turning angle, narrowest diameter of path, distribution information of stenotic segments, branch complexity index and risk labeling field maps geometric and semantic indicators with different dimensions and different value ranges to a unified scoring scale, and forms an adjustable comprehensive score through indicator weight configuration table and risk labeling field. In the case that the existing scheme of ranking only by total path length or single indicator is difficult to reflect the differences in the distribution of stenotic segments and branch complexity of path, the set of intervention path ranking results can simultaneously reflect the differences in path under multidimensional constraints, which is convenient for selecting candidate paths that are more in line with operational needs in the scenario of hepatic artery variation anatomy.

[0034] (3) The above-mentioned technical solution of performing three-dimensional model overlay rendering and generating interventional planning output structure based on interventional path candidate set structure and hepatic artery anatomical annotation structure jointly displays the hepatic artery tree centerline, vessel radius attribute and key branch name with candidate path trajectory, intervention point and termination point with high comprehensive score in the same three-dimensional model, and organizes the recommended path record into interventional planning output structure containing intervention point and termination point information. Compared with the existing solution that only provides static three-dimensional display or text description, it helps to directly call the anatomical background and path information consistent with the hepatic artery variant type label during preoperative discussion and intraoperative navigation, and reduce the misunderstanding caused by the inconsistency of interventional path expression methods between different systems. Attached Figure Description

[0035] Figure 1 This is a flowchart illustrating a method for rapid reconstruction of hepatic artery variant anatomy and recommendation of interventional site points, provided in an embodiment of this application. Detailed Implementation

[0036] Example 1: Refer to Figure 1 This is a flowchart illustrating a method for rapid reconstruction of hepatic artery anatomy and recommendation of interventional site points according to an embodiment of the present invention. The flowchart may include at least steps S100-S400:

[0037] S100: Acquire multi-phase enhanced hepatic artery phase images and patient basic information, perform format-unified registration, noise suppression, liver organ segmentation and hepatic artery candidate region localization processing, and generate hepatic artery candidate region mask structure.

[0038] S200: Based on the candidate region mask structure of the hepatic artery, perform three-dimensional vessel segmentation, centerline tracing and anatomical prior classification annotation processing to generate a hepatic artery anatomical annotation structure with variant labels and key branch annotations.

[0039] S300, based on the hepatic artery anatomical annotation structure with variant labels and key branch annotations, performs lesion blood supply branch identification, target encoding and candidate path search and geometric feature calculation and processing to generate an interventional path candidate set structure containing multiple geometric indicators;

[0040] S400: Based on the candidate set structure of intervention paths, perform multi-index normalization of paths, weighted scoring and sorting, and three-dimensional model overlay rendering operations to generate an intervention planning output structure containing information on intervention points and termination points.

[0041] Specifically, in S100, multi-phase enhanced hepatic artery phase images and patient basic information are acquired, and format-unified registration, noise suppression, liver organ segmentation and hepatic artery candidate region localization processing are performed to generate a hepatic artery candidate region mask structure.

[0042] The multi-phase enhanced hepatic arterial phase imaging specifically includes a set of three-dimensional grayscale volumetric data of arterial and portal venous phase sequences acquired at different time points, such as the arterial and portal venous phases, after intravenous injection of contrast agent. The data dimensionality is determined by the scanning device model, slice thickness, and voxel spacing parameters. The patient's basic information specifically includes patient identification, gender, age, weight, previous liver-related surgical records, preliminary clinical diagnosis information, and records of scanning device model, slice thickness, voxel spacing, contrast agent dosage, and injection time to support image data standardization and post-processing. The system retrieves the aforementioned image data and text information from the image archiving system and examination information system based on the examination number through the task management unit. The image data is loaded as raw three-dimensional volumetric data, and the patient's basic information is loaded as structured text data. In the processing session established in memory, both are used as unified inputs for subsequent format registration, noise suppression, liver organ segmentation, and hepatic artery candidate region localization.

[0043] Specifically, in this embodiment, the input sources for the multi-phase enhanced image standardization and hepatic artery candidate region localization steps are the multi-phase enhanced hepatic artery phase images generated in the previous imaging examination and the patient's basic information corresponding to this examination. The multi-phase enhanced hepatic artery phase images refer to the three-dimensional grayscale volumetric data set of arterial and portal venous phase sequences acquired at different time points after intravenous injection of contrast agent. The patient's basic information refers to the patient's identification, gender, age, weight, previous liver-related surgical records, preliminary clinical diagnosis information, and records of scanning equipment model, slice thickness, voxel spacing, contrast agent dosage, and injection time related to this imaging examination. When the system initiates this step, the task management unit retrieves the corresponding multi-phase enhanced hepatic artery phase images and patient basic information from the image archiving system and examination information system according to the examination number. The image data is loaded as raw three-dimensional volumetric data, and the patient's basic information is loaded as structured text data. A processing session for the current patient's examination is established in memory, and the above data serves as the unified input entry for subsequent format unification registration, noise suppression, liver organ segmentation, and hepatic artery candidate region localization processing.

[0044] Specifically, the standardized registration process first normalizes the differences in matrix size, voxel spacing, and coordinate origin between images from different phases. The system reads the row and column count, layer count, voxel spacing, and scan direction description recorded in the arterial and portal venous phase sequences, maps these parameters to the internal spatial description structure, and establishes a unified reference coordinate system. Based on this, the system performs 3D resampling on each phase image according to the preset standard voxel spacing and matrix size, interpolating the original volume data onto a unified voxel grid to obtain multi-phase 3D volume data with consistent scale. Subsequently, the system uses the coordinate system of the arterial phase sequence as the registration reference and maps the portal venous phase sequence to the reference coordinate system through rigid or non-rigid spatial transformation. The transformation parameters are obtained through iterative optimization based on indicators such as gray-level correlation, edge overlap, or anatomical landmark overlap. During the optimization process, the system detects anomalies. When a sharp jump or significant misalignment in a local area is detected during the convergence process, the current examination number and sequence number are recorded, and the optimization is terminated. This situation is written to the processing log for manual review. After the above processing is completed, standardized multi-phase enhanced image data with a unified spatial scale and coordinate system is obtained. This data is registered in the internal structure as a basic image resource that can be shared by the subsequent vascular analysis module.

[0045] After obtaining standardized multi-phase contrast-enhanced imaging data, this step performs noise suppression processing on the data. The purpose of noise suppression is to reduce random noise and streak artifacts without damaging the liver and vascular boundary structures. The system first searches its knowledge base for common noise types corresponding to the scanning equipment model and parameter settings in the patient's basic information, broadly classifying the noise into categories such as high-frequency random noise, low-frequency streak artifacts, and quantization noise. Subsequently, the system performs a block-by-block traversal of the standardized multi-phase contrast-enhanced imaging data, calculating the local gray-level mean, gray-level variance, and gradient intensity distribution for each three-dimensional sub-region, thereby determining whether the region is closer to the liver parenchyma, a densely vascularized area, or the background. For sub-blocks identified as liver parenchyma regions, the system employs 3D smoothing filtering or correlation filtering based on similar neighborhoods to reduce high-frequency noise and maintain the overall grayscale distribution trend without significant drift. For sub-blocks identified as densely vascularized regions, a combination of anisotropic filtering and edge-preserving filtering is used to maintain structural continuity along the direction of blood vessel course and suppress noise diffusion and blur expansion in directions perpendicular to the course of blood vessels. For background regions, a strong smoothing strategy is used to suppress background noise without affecting the interpretation of regions of interest. After processing, denoised multi-phase enhanced image data is obtained. The system establishes an index relationship between this data and the original standardized multi-phase enhanced image data, and records the noise suppression strategy type, main parameters, and processing time in the metadata for subsequent backtracking and algorithm adjustment. The denoised multi-phase enhanced image data serves as the direct input for liver organ segmentation and hepatic artery candidate region localization.

[0046] After noise suppression, this step proceeds to the liver organ segmentation stage. Liver organ segmentation aims to extract the overall contour of the liver from the denoised multi-phase enhanced image data, providing boundary constraints for subsequent localization of candidate hepatic artery regions within the liver. Based on the patient's gender and body weight information, combined with prior knowledge of the arterial and portal venous phase enhancement modes, the system infers the approximate grayscale range of the liver and the abdominal spatial range where the liver is located from the denoised multi-phase enhanced image data. Subsequently, the system loads a pre-trained organ segmentation model or probabilistic segmentation model within this spatial range, predicting the probability value for each voxel belonging to the liver, lower border of the heart, lower border of the lung, stomach, spleen, or other tissue categories. For each voxel, the system selects the category with the highest probability value as the initial label, obtaining a preliminary liver region mask. To reduce spurious segmentation, the system performs 3D connected component analysis on the preliminary liver region mask, eliminating isolated regions that are too small and far from the typical location of the liver, and correcting regions with obvious depressions or local defects in the boundary based on the liver morphology and the prior knowledge of the scaphoid contour. Taking advantage of the higher contrast between the hepatic veins and the main portal vein in the portal venous phase data, the system further refines the local threshold within the liver region, performing local intensity statistics on areas near the hepatic hilum and upper border to correct for liver edge truncation caused by global threshold deviation. After these steps, a liver region mask conforming to the anatomical structure is obtained. The system records this mask as a liver mask field bound to the denoised multi-phase enhanced image data, defining the spatial range for subsequent hepatic artery candidate regions.

[0047] After obtaining the liver region mask, this step continues with the localization of the hepatic artery candidate region. The goal of hepatic artery candidate region localization is to extract the spatial region corresponding to the main trunk and major branches of the hepatic artery from the liver and its adjacent structures. The system first extracts the liver surface contour based on the liver region mask and estimates the approximate location of the hepatic hilum in three-dimensional space based on anatomical knowledge, defining a spatial range of a certain thickness around the hepatic hilum as the key analysis region. This key analysis region typically covers the origin of the proper hepatic artery, the main trunk of the portal vein, and the common bile duct, and is the area where the main trunk of the hepatic artery and the initial segments of the left and right hepatic arteries are concentrated. Within the key analysis region, the system calculates the tubular structure response value of each voxel from the denoised multi-phase enhanced image data. The tubular structure response value measures whether the neighborhood of the voxel contains a slender and continuous tubular structure by evaluating the directionality and curvature characteristics of the local gray-scale distribution, which is used to enhance vascular-like structures and weaken the response of plate-like or blocky tissues. Based on the magnitude of the tubular structure response value, the system performs threshold segmentation on the key analysis region to obtain an initial set of hepatic artery candidate voxels composed of high-response voxels.

[0048] Since the portal vein and bile ducts exist near the porta hepatis in addition to the hepatic artery, this step also performs multi-phase enhancement feature screening on the initial hepatic artery candidate voxel set. The system calculates the average gray value and gray value change amplitude of each candidate voxel and its neighborhood in the arterial and portal venous phase images, respectively. Candidate voxels with more pronounced enhancement in the arterial phase and relatively weaker enhancement in the portal venous phase are marked as more consistent with hepatic artery characteristics, while candidate voxels with significant enhancement in the portal venous phase and relatively slow enhancement in the arterial phase are marked as more consistent with portal vein characteristics. Candidate voxels exhibiting bile duct-like low density or density changes that do not conform to the vascular enhancement pattern are eliminated or downgraded from the candidate set. Through the above temporal enhancement pattern screening, a screened hepatic artery candidate voxel set is obtained, in which the voxels more concentratedly reflect the spatial location of the main hepatic artery and its larger branches.

[0049] After obtaining the selected set of candidate hepatic arteries voxels, this step further constructs candidate connected regions based on spatial connectivity. The system employs a 3D connectivity analysis algorithm to merge candidate systems sharing common surfaces or edges into the same connected region. For each connected region, the system calculates its 3D spatial geometric features, including the coordinates of its center location, its total volume, its principal axis direction vector, its distance from the estimated location of the hepatic hilum, and its length along the principal axis. Combining these geometric features with prior anatomical knowledge, the system marks connected regions that are close to the hepatic hilum, have a larger volume, and whose principal axis direction is similar to the typical course of the hepatic artery as high-confidence hepatic artery candidate regions, and marks connected regions that are far from the hepatic hilum or have a smaller volume as low-confidence regions or regions to be verified. Low-confidence regions are not included in the final hepatic artery candidate region mask structure, but their spatial location and feature values ​​are recorded in the system's internal log for subsequent interactive correction or parameter adjustment.

[0050] After screening high-confidence connected components, the system generates a final hepatic artery candidate region mask structure based on all high-confidence hepatic artery candidate connected components. During the generation process, the system creates a corresponding 3D binary mask at the location of each high-confidence connected component, marking all voxels within the connected component as candidate hepatic artery regions. To reduce the impact of boundary position errors in subsequent 3D vessel segmentation, the system adds one or more voxel extensions around each marked voxel, forming a candidate region with a small buffer zone, ensuring that the actual hepatic artery lumen remains within the candidate region even under slight registration or interpolation errors. The final hepatic artery candidate region mask structure, as the output of this step, is written into the structured record of the current patient examination along with standardized multi-phase enhanced image data. Among them, the hepatic artery candidate region mask structure is explicitly marked as the input field for the three-dimensional vessel segmentation and centerline tracing sub-stage in the subsequent integrated reconstruction of the hepatic artery tree and anatomical prior classification step. The standardized multi-phase enhanced image data is marked as the input field of the same step, realizing continuous data transfer from the multi-phase enhanced image standardization and hepatic artery candidate region localization step to the subsequent vessel reconstruction step.

[0051] The technical effect of this step can be summarized as follows: by performing standardized registration, noise suppression, liver organ segmentation, and hepatic artery candidate region localization on multi-phase enhanced hepatic artery images and patient basic information, a clear hepatic artery candidate region mask structure is obtained. This structure is then output in conjunction with standardized multi-phase enhanced image data, providing a stable spatial constraint and image basis for subsequent integrated reconstruction of the hepatic artery tree and anatomical prior classification annotation steps.

[0052] S200: Based on the candidate region mask structure of the hepatic artery, perform three-dimensional vessel segmentation, centerline tracing and anatomical prior classification annotation processing to generate a hepatic artery anatomical annotation structure with variant labels and key branch annotations.

[0053] In this embodiment, the input to the integrated reconstruction and anatomical prior classification annotation step of the hepatic artery tree comes from the hepatic artery candidate region mask structure and standardized multi-phase enhanced image data output from the aforementioned multi-phase enhanced image standardization and hepatic artery candidate region localization steps. The hepatic artery candidate region mask structure is three-dimensional mask data representing the spatial range of the suspected main trunk and major branches of the hepatic artery under a unified spatial coordinate system, using voxel-level binary labeling to record the candidate region positions. The standardized multi-phase enhanced image data is multi-phase enhanced three-dimensional volume data after resampling and spatial registration, containing grayscale information for time points such as the arterial phase and portal venous phase. When the system starts executing this step, it retrieves the corresponding hepatic artery candidate region mask structure and standardized multi-phase enhanced image data under the same examination number from the data management module, establishes the coordinate correspondence between the two in the working memory, and treats them as the joint input for the three-dimensional vessel segmentation and centerline tracing processing in this step.

[0054] Specifically, the 3D vessel segmentation process first crops standardized multi-phase enhanced image data within the spatial range defined by the hepatic artery candidate region mask structure, obtaining local 3D volume data containing only the candidate hepatic artery region. To reduce interference from background tissue on vessel analysis, the system performs mask cropping on the original voxels according to the hepatic artery candidate region mask structure, uniformly marking voxels outside the mask as background. Subsequently, the system performs vessel enhancement filtering on the cropped local 3D volume data. This vessel enhancement filtering enhances the response of slender tubular structures and suppresses the response of plate-like or block-like structures by analyzing the local gray-level gradient direction and principal curvature characteristics. The filtered output forms a vessel response intensity field. The system selects a set of intensity thresholds and connectivity conditions on this intensity field, performs 3D threshold segmentation and region growing, and aggregates voxels that meet the intensity conditions and are spatially connected into vessel candidate regions. For isolated small clusters or structurally abnormal short branches in the vessel candidate region, the system screens them through connected component analysis and volume statistics. Clusters that clearly do not conform to the structural characteristics of the hepatic artery tree are recorded as suspected artifacts and removed from the current vessel candidate results. At the same time, their spatial location and volume characteristics are recorded in the log for reference by the subsequent quality control module. After the above processing, a three-dimensional vessel segmentation architecture is formed. This structure is recorded in the current processing session as the basic input for subsequent centerline tracing and topology construction, and is labeled as the vessel segmentation voxel structure field in the data description.

[0055] After obtaining the blood vessel segmentation voxel structure, this step performs centerline tracing to extract the geometric skeleton of the hepatic artery tree and establish topological relationships. The system performs morphological shrinkage on the blood vessel segmentation voxel structure on a 3D mesh, transforming the large blood vessel bodies into a single-width skeleton structure, resulting in a set of candidate center voxels. For this set, the system constructs an undirected graph structure based on spatial connectivity, treating each center voxel as a node and the connectivity between adjacent center voxels as edges, recording the corresponding spatial distance on each edge. Subsequently, the system selects several seed nodes near the inferred hepatic artery origin region and employs a centerline tracing strategy based on shortest path search, gradually expanding from the seed nodes along the direction with the smaller edge weight until reaching the terminal region of the blood vessel segmentation system structure. During the tracing process, when the system encounters a node with a degree greater than two, it marks the node as a blood vessel bifurcation point and continues tracing paths on two or more branches, simultaneously recording the start point, end point, length, and average radius information of each path. For paths whose length is significantly shorter than a preset threshold and whose endpoints are close to the segmentation edge, the system marks them as suspicious pseudo-branches and reduces their weight or removes them during subsequent topology cleaning. Through centerline tracing, a preliminary set of hepatic artery centerlines is obtained. Based on this, the system constructs a hepatic artery tree centerline graph structure, which consists of a set of nodes and a set of edges. Nodes correspond to vessel bifurcation points or terminal points, and edges correspond to vessel segments between adjacent nodes. Attributes such as length, radius, and spatial direction are attached to nodes and edges to provide input for anatomical prior classification and key branch identification.

[0056] Furthermore, to construct a hepatic artery tree topological description that conforms to clinical anatomical logic, the system performs topology cleaning and trunk identification processing on the hepatic artery tree centerline diagram structure. During topology cleaning, the system screens short branches and abnormally acute-angled branches in the centerline diagram structure based on attributes such as vessel radius, path length, and branch angle. Branches that clearly do not conform to the continuous course characteristics of the hepatic artery tree are deleted or merged to avoid excessive interference from false branches during subsequent anatomical prior matching. During trunk identification, the system starts from a node near the hepatic hilum region and traces along a path with a large radius and stable direction to identify the common hepatic artery, the proper hepatic artery, and the main segments extending to the left and right hepatic lobes, recording these segments as the trunk path set. For secondary branches connected to the trunk path set, the system performs preliminary grouping according to the branch's location and course direction, preparing a candidate list for subsequent segmental artery and lobar segment labeling. After topology cleaning and trunk identification processing, the system obtains a structurally regular hepatic artery tree topological description structure, which is registered as a hepatic artery tree topological description field in the data management module.

[0057] After constructing the hepatic artery tree topology description structure, this step proceeds to the anatomical prior classification annotation process. This process relies on a pre-defined set of hepatic artery anatomical prior rules. These rules are derived from a summary of common clinical hepatic artery anatomical types, including normal structures, variations originating from the right hepatic artery, variations originating from the left hepatic artery, and multiple blood supply structures. For each pattern, the system provides the typical number of branches, branch origin locations, and course ranges for the main trunk segment, left hepatic segment, right hepatic segment, and segmental branches. The system matches the main trunk path set and secondary branch set in the hepatic artery tree topology description structure with the anatomical prior rule set. By comparing the relative positions of branch origin nodes in three-dimensional space, the number of branches, and course directions with the descriptions in the rules, the system determines the hepatic artery variation type corresponding to this examination. If multiple rule models have similar matching scores, the system selects the type that best fits the overall topology structure according to a pre-defined priority strategy. Simultaneously, the system records the matching scores and ambiguous branch numbers in a log for later review by the manual review module. Through the above matching process, the system generates a variant label field for the current hepatic artery tree. The variant label field records the classification number and text description of the hepatic artery anatomy type in this case at the structural level, which is used for subsequent steps to identify the blood supply branches of the lesion and to generate the preoperative report.

[0058] After determining the overall variant type, this step continues with critical branch annotation processing on the hepatic artery tree topology description structure. Critical branch annotation processing, constrained by the determined variant label fields and the set of anatomical prior rules, assigns clinically significant anatomical names to each edge and node in the topology graph. The system first identifies the common hepatic artery, proper hepatic artery, and main hepatic artery segments in the trunk path set, and adds name labels to the corresponding edges and nodes in the data structure. Subsequently, based on the relationship between the variant type and the spatial quadrant of the node, the system distinguishes between the left and right hepatic artery trunks. The upstream and downstream branches connected to the left and right hepatic artery trunks are matched one-to-one with the segmental artery names listed in the anatomical prior rules according to the branching order. Branches that do not meet the length threshold or have excessive spatial deviation are recorded as candidate abnormal branches, and a "to be confirmed" mark is added to the label field. For duplicated blood supply branches and common trunk branches in certain special variant types, the system refers to specific patterns in the rule set and assigns appropriate names to these branches by simultaneously observing the number of branches, their origin location, and their course angle. Multiple blood supply area labels are then associated with these branches in the structure, facilitating the subsequent lesion blood supply analysis module's identification of multi-source blood supply. Through key branch annotation processing, each important edge and node in the topology graph obtains a clear anatomical name, forming a hepatic artery tree structure with rich semantic labeling.

[0059] To facilitate direct use of the results in subsequent steps, this embodiment integrates the hepatic artery tree topology description structure, variant label field, and key branch name label to form a unified hepatic artery anatomical annotation structure. At the data level, the hepatic artery anatomical annotation structure is a multi-component dataset, including a three-dimensional centerline coordinate sequence recording the vessel geometry, numerical fields recording vessel radius and length attributes, topological fields recording the hierarchical relationships between vessels, variant label fields recording the overall anatomical type, and key branch annotation fields recording the names of each branch segment and the range of its corresponding liver segment. This structure is registered as the hepatic artery anatomical annotation structure output field in the patient data record corresponding to the current examination, serving as one of the inputs for subsequent steps of lesion blood supply branch identification and multi-entry interventional path candidate generation. Specifically, in the subsequent steps of lesion blood supply branch identification and multi-entry interventional path candidate generation, the system reads the hepatic artery anatomical annotation structure output from the patient data record, and together with the lesion spatial location data and candidate interventional entry set introduced in subsequent steps, participates in lesion blood supply branch search and target encoding processing; at the same time, in the step of multi-index weighted scoring and interventional point recommendation output structure generation, the system also overlays and renders the hepatic artery anatomical annotation structure and the interventional path candidate set structure to display the spatial relationship between the path and the anatomical structure in the three-dimensional view.

[0060] The technical effect of this step can be summarized as follows: by performing three-dimensional vessel segmentation, centerline tracing, and anatomically prior-driven typing and key branch annotation on the hepatic artery candidate region mask structure and standardized multi-phase enhanced image data from the previous steps, a complete hepatic artery anatomical annotation structure is obtained, providing a stable topological foundation and semantic support for subsequent steps to carry out lesion blood supply branch identification and interventional path planning.

[0061] S300, based on the hepatic artery anatomical annotation structure with variant labels and key branch annotations, performs lesion blood supply branch identification, target encoding and candidate path search and geometric feature calculation and processing to generate an interventional path candidate set structure containing multiple geometric indicators;

[0062] In this embodiment, the identification of lesion blood supply branches, target encoding, candidate path search, and geometric feature calculation are jointly driven by the hepatic artery anatomical annotation structure generated in the preceding steps, the spatial location data of the lesion provided by the image analysis module or an external system, and the set of candidate interventional entry points given by the preoperative configuration module. The hepatic artery anatomical annotation structure is a comprehensive data set describing the hepatic artery tree centerline, vessel radius, topological relationships, and anatomical names in a unified three-dimensional coordinate system. It includes fields labeling the common hepatic artery, proper hepatic artery, left hepatic artery, right hepatic artery, and the names of each segment and the information fields of the hepatic segments to which each segment belongs. The lesion spatial location data is three-dimensional spatial description data generated by the lesion segmentation module based on multi-phase enhanced images or other imaging examination results. It includes at least the coordinates of the lesion center point, the coordinates of several sampling points within the lesion volume, and the lesion volume range mask field. By sharing the same coordinate system and voxel grid with the aforementioned standardized multi-phase enhanced image data, it can be directly aligned with the hepatic artery anatomical annotation structure. The set of candidate interventional entry points is provided by the interventional physician or the system pre-configuration module, including various entry options such as femoral artery approach, radial artery approach, brachial artery approach, or other local artery approach. Each entry option corresponds to an entry node identifier in the vascular tree topology, a set of three-dimensional coordinates, and local orientation information related to the direction of the interventional device. When the system enters this step, it first retrieves the hepatic artery anatomical annotation structure corresponding to the current examination number from the patient data record. Then, it reads the lesion spatial location data associated with the same examination number from the lesion management module or external system, and then obtains the set of candidate interventional entry points from the preoperative planning configuration. The three are linked together in the same processing session as a unified input for subsequent lesion blood supply branch identification and path search calculations.

[0063] Specifically, the lesion blood supply branch identification process is based on the spatial relationship between the hepatic artery anatomical annotation structure and the lesion spatial location data. For each lesion instance, the system first extracts the coordinates of the lesion's center point and several sampling points representing the lesion surface from the spatial location data. These coordinates are then projected onto the set of centerlines in the hepatic artery anatomical annotation structure within a unified three-dimensional coordinate system. The projection process involves finding the centerline point closest to the lesion sampling point among all vessel centerline segments. For each lesion sampling point, the system records its corresponding nearest centerline point, the corresponding vessel branch identifier, and the branch's hierarchy information in the tree structure. Subsequently, the system statistically weights vessels that may be blood supply branches according to the number of matches and spatial distance distribution between all sampling points within the lesion and each branch. Branches with a high number of matches and a small average distance are marked as high-confidence blood supply branches, while branches with fewer matches or only a few matches at the distal end are marked as low-confidence branches. For lesions that may have multiple blood supply sources, the system allows recording several high-confidence blood supply branches for the same lesion and attaching a blood supply weight estimation field to the data. This is used by the subsequent path search module to identify the primary and secondary blood supply paths. During the lesion blood supply branch identification process, the system also refers to liver segment attribution information. By comparing the spatial location of the lesion's center point with the liver segment information marked in the hepatic artery anatomical annotation structure, lesions clearly located within a certain liver segment are preferentially matched with the corresponding segmental artery, thereby improving the ranking of that segmental artery in the blood supply branch candidate list. After the above processing, the system forms a lesion blood supply branch set for each lesion. Each record in the set contains a lesion identifier, a blood supply branch identifier, matching statistics, and a blood supply weight estimation value. This set is registered as the intermediate output field "lesion blood supply branch set" in the data structure, providing a foundation for subsequent target encoding and candidate path search.

[0064] After obtaining the set of lesion blood supply branches, this step performs target encoding processing, converting the specific target locations on the blood supply branches into target descriptions that can be directly referenced by the path search module. Specifically, for each candidate blood supply branch record in the lesion blood supply branch set, the system extracts the centerline point sequence and radius information of the corresponding vascular branch from the hepatic artery anatomical annotation structure. Then, the system searches for the centerline point with the smallest distance to the lesion center point along the centerline of the vascular branch, and regards this centerline point as the main target location of the lesion on that blood supply branch. At the same time, to deal with cases where the lesion volume is large or extends to multiple sub-regions, the system can also detect multiple candidate target points at the intersection of the lesion volume mask and the vascular branch centerline, for example, selecting several intersection points close to the centerline near the lesion boundary and recording their relative position index on the centerline. During the target encoding process, the system generates a unique target number for each target point, which contains a combined description of the lesion identifier, blood supply branch identifier, and centerline position index, and also includes attributes such as local vascular radius, local curvature features, and distance from the geometric center of the lesion, for subsequent path geometric feature calculation. All target records are organized into a target encoding list and registered as the intermediate output field, the target encoding structure. In the subsequent candidate path search process, the system treats each target in the target encoding structure as a target node and each entry node in the candidate intervention entry set as a starting node, thus constructing a multi-starting-point, multi-endpoint path search task set.

[0065] After target encoding is completed, this step proceeds to candidate path search processing. Candidate path search processing constructs a computationally achievable graph-structured path model based on the vascular topology information in the hepatic artery anatomical annotation structure and the entry node information in the candidate interventional entry point set. The system first extracts the vascular centerline map from the hepatic artery anatomical annotation structure, treating each centerline node as a node in the graph and the vascular segments between adjacent centerline nodes as edges, attaching length and radius attributes to the edges. Simultaneously, based on the entry point definitions in the candidate interventional entry point set, the system finds a mapping location in the vascular graph structure for each interventional entry point. For example, the femoral artery approach is mapped to an entry node in a segment of the abdominal aorta, and the radial artery approach is mapped to a relevant entry node in the subclavian or brachial artery. New connecting edges are added between the corresponding entry node and existing vascular nodes in the graph structure, with length and radius attributes set based on clinical experience. After completing the entry point mapping, the system encodes and records the corresponding centerline node identifier for each target point in the graph structure, using it as the target node set for path search.

[0066] During the specific path search process, the system constructs a path search task for each candidate interventional entry point and each target point combination. It sequentially executes a graph-based shortest path or minimum-cost path search strategy from the candidate interventional entry point node to the target point node. The search cost considers not only vessel segment length but also vessel radius, local branch angles, and other factors. By adding a penalty term to the cost of each edge for excessively small vessel radii or excessively curved paths, the search process is more biased towards selecting paths with geometric features more conducive to instrument passage. During the search, when the system detects that a candidate path passes through a vessel radius continuously below a set threshold or exhibits multiple extreme acute-angle bifurcations, it marks the path as a high-risk path and adds a risk marker field to the path record. If a connected path from the candidate interventional entry point node to the target point node cannot be found in the given graph structure, the system records the entry point and target point combination as an unreachable combination and writes the task status to the log for subsequent manual review. For all feasible combinations, the system stores at least one candidate path for each entry point and target point combination. If multiple paths with similar costs exist, several alternative paths with costs within a preset range are retained. After the above processing, the system forms a set containing a large number of candidate paths. Each candidate path carries an entry number, a target number, and a node sequence description generated by the path search strategy. This set exists as a candidate path set field in the data structure.

[0067] After the candidate path set is generated, this step continues with geometric feature calculation processing to extract various geometric indicators reflecting path accessibility and operational difficulty from the candidate path set. Geometric feature calculation processing first, for each candidate path, accumulates the length of each vessel segment based on the spatial distance between the path node sequence and corresponding nodes, obtaining the total length indicator of the path. Then, based on the spatial direction changes between three consecutive points in the path node sequence, the system calculates the turning angle at each turning point, and statistically analyzes all turning angles throughout the entire path to obtain the curvature indicator and the maximum turning angle indicator describing the overall curvature of the path. Furthermore, the system retrieves the vessel radius attribute corresponding to each centerline node from the hepatic artery anatomical annotation structure, records the minimum radius value traversed by the path node sequence while iterating through the path node sequence, and uses this as the narrowest diameter indicator of the path, reflecting the most restrictive parts of the path for instrument passage. For paths with multiple radii close to the threshold, the system also calculates the distribution location and number of consecutive segments of these narrowing points on the path, recording them as the narrowing segment distribution indicator. In addition, the system can calculate a branch complexity index based on the number of branches and branch levels traversed by each path, which is used to express the number of branch nodes and the level depth that need to be traversed in the path. For abnormal situations that occur during the geometric feature calculation process, such as incomplete path node sequences or missing radius information for some nodes, the system marks the path as incomplete and records the reason in the log to avoid misusing the path in subsequent scoring steps.

[0068] After completing the aforementioned lesion blood supply branch identification, target encoding, candidate path search, and geometric feature calculation, this step organizes all candidate paths and their corresponding geometric features into a unified data structure, forming the output of this step. Specifically, the system generates a record for each candidate path, including the intervention entry number, target number, path node sequence identifier, total path length, tortuosity index, maximum turning angle, narrowest path diameter, stenosis distribution index, branch complexity index, and optional risk labeling fields. All path records are aggregated into an intervention path candidate set structure, which is registered as an output field in the patient data record and explicitly labeled as one of the inputs for the subsequent multi-index weighted scoring and intervention point recommendation output structure generation steps. When performing the subsequent multi-index weighted scoring and intervention point recommendation output structure generation process, the system reads the intervention path candidate set structure output from this step from the patient data record. Together with the hepatic artery anatomical annotation structure output from the previous main step in the same record, this forms the basic data source for the scoring and 3D model overlay rendering process, thus forming a continuous data flow across main steps.

[0069] The technical effect of this step can be summarized as follows: by jointly processing the hepatic artery anatomical annotation structure, lesion spatial location data and candidate interventional entry point set, an interventional path candidate set structure covering lesion blood supply branches, target points and multiple entry point candidate paths is obtained. In addition, various geometric indicators such as path length, tortuosity and narrowest diameter are added to this structure, providing a complete data foundation and clear geometric description for subsequent path multi-indicator normalization and weighted scoring ranking.

[0070] S400: Based on the candidate set structure of intervention paths, perform path multi-index normalization, weighted scoring sorting and three-dimensional model overlay rendering operations to generate an intervention planning output structure containing information on intervention points and termination points.

[0071] In this embodiment, the inputs for path multi-index normalization, weighted scoring ranking, and 3D model overlay rendering operations mainly come from the interventional path candidate set structure and hepatic artery anatomical annotation structure generated in the aforementioned steps. The interventional path candidate set structure is a comprehensive dataset describing each candidate interventional path in a unified spatial coordinate system. Each path record includes at least the interventional entry number, corresponding target point number, path node sequence identifier, total path length, path curvature, maximum turning angle, narrowest path diameter, stenosis distribution information, branch complexity index, and risk marker fields, among other geometric and semantic attributes. The hepatic artery anatomical annotation structure is the output of the aforementioned integrated reconstruction and anatomical prior classification annotation steps of the hepatic artery tree, including the hepatic artery tree centerline coordinate sequence, vessel radius attribute, topological relationship field, variation type label, and key branch names and their corresponding hepatic segments. When the system enters this step, it retrieves the two structures mentioned above from the data management unit using the current patient examination number, establishes an index relationship between the two in the working memory, uses the interventional path candidate set structure as the main data source for path scoring and ranking, and uses the hepatic artery anatomical annotation structure as the anatomical background input for path interpretation and 3D model overlay rendering, thus constructing a unified processing session environment for this step.

[0072] Specifically, the path multi-index normalization process first performs centralized statistics and range analysis on various geometric indices in the candidate path structure. The system traverses all candidate path records, summarizing the value ranges of indices such as total path length, path curvature, maximum turning angle, narrowest pipe diameter, and branch complexity, and calculates the minimum, maximum, and distribution characteristics of each type of index on the candidate path set. Based on this statistical information, the system constructs an index normalization configuration table, mapping the original indices of different dimensions and scales to a unified dimensionless scoring scale. For indices like total path length, the normalization process sets the mapping relationship according to the principle that smaller lengths result in higher scores; for indices like the narrowest pipe diameter, the mapping relationship is set according to the principle that larger diameters result in higher scores; and for indices reflecting the degree of bending, such as curvature and maximum turning angle, the conversion is performed according to the principle that smaller values ​​result in higher scores. For paths with missing or obviously anomalous values, the system records the corresponding path number before normalization and replaces it with a fixed penalty score or reference median to avoid excessive interference from a single anomalous path on the overall score distribution. Simultaneously, the system logs the processing status of the path. Through this process, the system converts the original geometric indices of each candidate path into normalized indices under a unified dimension, generating an intervention path scoring input structure. Each record in the intervention path scoring input structure corresponds one-to-one with the path records in the original intervention path candidate set structure, adding fields such as normalized length, normalized pipe diameter, and normalized bending.

[0073] After obtaining the access path scoring input structure, this step performs weighted scoring and sorting. The weighted scoring process is based on a preset indicator weight configuration table, combining various normalized indicators into a comprehensive score for each path. The weight configuration table is typically determined by both the system's default configuration and the interventional physician's personalized configuration adjusted according to actual preferences. It records the weight ratio of each type of normalized indicator in the comprehensive score, such as the total path length, path curvature, narrowest tube diameter, branch complexity, and the weight factor corresponding to risk markers. For each path record in the interventional path scoring input structure, the system reads each normalized indicator value of the path according to the weight configuration table, multiplies it by the corresponding weight factor, sums the results, and registers the result as the basic comprehensive score field for that path. For path records marked with high risk, the system applies an additional penalty to the basic comprehensive score, deducting a certain score according to the risk level to reflect risk factors such as too many narrow sections, excessively large extreme turning angles, or too many branch levels crossed. In scenarios with multiple lesions or multiple targets, the system also adds appropriate bonus points to paths leading to the main blood supply branches based on the lesion's blood supply weight, thus reflecting the path's priority coverage of key lesions in the score. After completing the above calculations, the system obtains a set of access path ranking results, where each record includes a path identifier, entry number, target number, comprehensive score value, and detailed score composition. During the ranking process, the system groups the access path ranking results set according to the entry and target dimensions. Within each entry and target group, it sorts the paths from high to low based on the comprehensive score value, while recording the ranking order within the group and whether the preset threshold for the number of retained paths has been reached. Paths with a comprehensive score that is too low and exceeds the threshold for the number of retained paths are marked as low-priority paths and can be selectively hidden or collapsed in subsequent rendering and output stages.

[0074] Furthermore, to correlate the scoring results with the anatomical information in the hepatic artery anatomical annotation structure, this step performs path annotation expansion processing after the weighted scoring and ranking are completed. The system traverses each high-priority path in the interventional path ranking result set, and searches for the corresponding centerline node set and branch name set in the hepatic artery anatomical annotation structure through the path node sequence identifier. The names of the key branches traversed by the path, the information of the liver segment to which it belongs, and the variant type label of the hepatic artery to which it belongs are appended to the path record, forming a path ranking record with anatomical annotations. For paths that cross multiple segmental arteries or involve multiple blood supply areas, the system estimates the coverage ratio of the path in different liver segments according to the proportion of the length of the path node in the liver segment space, and adds a simplified liver segment coverage description field to the path record for displaying the correspondence between the path and the target area in subsequent structured reports. Thus, the interventional path ranking result set not only contains comprehensive scoring and ranking information in a purely geometric sense, but also embeds anatomical semantic content consistent with the hepatic artery anatomical annotation structure, providing a complete foundation for 3D model overlay rendering and textual expression of planning results.

[0075] After completing the scoring and annotation expansion, this step proceeds to the 3D model overlay and rendering stage. The 3D model overlay and rendering operation reconstructs the hepatic artery tree into a visualized 3D model based on the centerline coordinates and vessel radius attributes in the hepatic artery anatomical annotation structure. Simultaneously, selected high-priority interventional paths are overlaid onto this 3D model as lines or tubular trajectories. Specifically, the system first generates a 3D skeleton of the hepatic artery according to the centerline sequence information in the hepatic artery anatomical annotation structure. Then, based on the vessel radius information of each centerline node, it constructs a tubular surface model of the hepatic artery. Different thicknesses and contrasts of light and dark are used to visually distinguish branches at different levels, thereby visually differentiating the main trunk, lobar segments, and terminal branches. Subsequently, based on the path node sequence recorded in the interventional path ranking result set, the system generates corresponding path trajectories in 3D space. Path trajectories with higher overall scores are drawn with a more prominent style, while paths with lower overall scores but still within the retention range are drawn with an auxiliary style, facilitating users to observe multiple candidate paths in the same view. For each path trajectory, the system adds intervention point and termination point markers at its start and end positions. The positions of these markers are derived from the corresponding entry node coordinates and target node coordinates in the path record. The markers are clearly presented in three-dimensional space, allowing users to directly identify the entry point of the interventional device into the vascular tree and the termination point at the blood supply branch. During the three-dimensional model overlay rendering process, the system also supports multi-view observation and display on different clipping planes, synchronously displaying the hepatic artery tree model and the original multi-phase enhanced images on the cross-section, and displaying the projection trajectory of the path on the original cross-section when needed, helping to understand the path's course in real anatomical space.

[0076] Based on the completed 3D model overlay rendering, this step constructs the intervention planning output structure, organizing the aforementioned scoring and ranking results and visualization rendering configurations into standardized output products. The intervention planning output structure includes recommended path records for each candidate intervention entry point and each lesion target point combination. Each recommended path record includes at least the following fields: intervention entry point number, 3D coordinates of the intervention point, 3D coordinates of the termination point, path node sequence summary, comprehensive score value, total path length, major bend location indicators, narrowest tube diameter and its location within the path, a list of key branch names, and corresponding trunk segment coverage description. For low-priority paths not recommended, only the path identifier and key indicators are retained as alternative information. The intervention planning output structure also records the view parameter information used in this 3D model overlay rendering, including viewing direction, zoom level, and clipping plane settings, for restoring the same viewing scene in subsequent sessions. After constructing the interventional planning output structure, the system writes this structure into the structured data record corresponding to the current patient's examination and links it to the interventional planning extended field under the patient's basic information entry. This allows the system to retrieve existing planning results from the patient's basic information extended registration field and compare them with new imaging data or use them as a follow-up basis when performing multi-phase enhanced imaging standardization and hepatic artery candidate region localization steps again. Simultaneously, the interventional planning output structure is also marked as a callable input field for external intraoperative navigation and report generation systems, facilitating a complete data loop within the same system, from preoperative image processing, anatomical reconstruction, and path planning to intraoperative navigation and postoperative recording.

[0077] The technical effect of this step can be summarized as follows: by performing normalization and weighted scoring and ranking on various geometric indicators in the interventional pathway candidate set structure, and overlaying the scoring results with the hepatic artery anatomical annotation structure to render a three-dimensional visual model, an interventional planning output structure containing intervention point and termination point information is generated. At the data level, a complete planning result from pathway candidate to comprehensive recommendation is formed, providing a clear and consistent structured foundation for data integration and clinical use between the preceding and following steps.

Claims

1. A method for rapid anatomical reconstruction and interventional point recommendation of hepatic artery variations, characterized in that, include: We acquire multi-phase enhanced hepatic artery phase images and patient basic information, perform format-unified registration, noise suppression, liver organ segmentation and hepatic artery candidate region localization processing, and generate hepatic artery candidate region mask structure. Based on the candidate region mask structure of the hepatic artery, three-dimensional vessel segmentation, centerline tracing and anatomical prior classification annotation are performed to generate a hepatic artery anatomical annotation structure with variant labels and key branch annotations. The anatomical prior classification and labeling process includes performing topology cleaning and trunk identification on the hepatic artery tree centerline diagram structure. Short branches and abnormally acute-angled branches are screened based on vessel radius, path length, and branch angle attributes. Branches that do not conform to the continuous course characteristics of the hepatic artery tree are deleted or merged. Starting from a node near the hepatic hilum, the main hepatic artery, proper hepatic artery, and major segments extending to the left and right hepatic lobes are traced along a path with a large radius and stable direction to determine the trunk path set. Secondary branches are initially grouped according to their location and course direction. Based on a preset set of hepatic artery anatomical prior rules, the trunk path set and secondary branch set in the hepatic artery tree topology description structure are matched with the anatomical prior rules. The rule set is derived from a summary of common clinical hepatic artery anatomical types, including normal structures, right hepatic artery origin variation structures, left hepatic artery origin variation structures, and multiple blood supply structures. The hepatic artery variation type is determined by comparing the relative position, number of branches and direction of travel of the branch origin node in three-dimensional space, and clinically significant anatomical names are assigned to each edge and node in the topology graph, generating a hepatic artery anatomical annotation structure with variation labels and key branch annotations. Based on the hepatic artery anatomical annotation structure with variant labels and key branch annotations, the system performs lesion blood supply branch identification, target encoding and candidate path search and geometric feature calculation and processing to generate an interventional path candidate set structure containing multiple geometric indicators. The lesion blood supply branch identification process involves extracting the coordinates of the lesion center point and several sampling points representing the lesion surface from the lesion spatial location data for each lesion instance. The coordinates are then projected onto the set of center lines in the hepatic artery anatomical annotation structure in a unified three-dimensional coordinate system. The nearest center line point is found to record the corresponding vascular branch identifier and hierarchical information. The matching frequency and spatial distance distribution between all sampling points within the lesion and each branch are statistically weighted. Branches with a high number of matching frequencies and a small average distance are marked as high-confidence blood supply branches. Furthermore, lesions clearly located within a certain hepatic segment are preferentially matched with the corresponding hepatic segment artery based on the hepatic segment attribution information to generate a set of lesion blood supply branches. The target coding involves extracting the centerline point sequence and radius information of the corresponding vascular branch from the hepatic artery anatomical annotation structure for each candidate blood supply branch record in the lesion blood supply branch set, finding the centerline point with the smallest distance from the lesion center point along the vascular branch centerline as the main target point location, and detecting multiple candidate target points at the intersection of the lesion volume mask and the vascular branch centerline for cases where the lesion volume is large or extends to multiple sub-regions. A unique target point number is generated for each target point, along with local vascular radius, local curvature features and distance attributes from the geometric center of the lesion, thus generating a target point coding structure. The candidate path search involves extracting the vascular centerline map from the anatomy annotation structure of the hepatic artery to construct a structural path model. Each centerline node is regarded as a node in the graph, and length and radius attributes are attached to the edges. According to the entry point definition in the candidate interventional entry point set, the mapping position of each interventional entry point is found in the vascular graph structure, and the corresponding entry point node is added to the connection edge between the existing vascular node and the existing vascular node. Each target point in the target point encoding structure is regarded as a target node. A path search task is constructed for each candidate interventional entry point and each target point combination. The shortest path or minimum cost path search strategy based on the graph structure is executed. The search cost considers the vascular segment length, vascular radius, and local branch angle, and a penalty term is added for vascular radii that are too small or bends that are too large. High-risk paths and unreachable combinations are marked, and a candidate path set is generated. The geometric feature calculation process includes, for each candidate path, accumulating the length of the vessel segment segment by segment according to the path node sequence to obtain the total length index, calculating the turning angle of all turning points in the path to obtain the curvature index and the maximum turning angle index, recording the minimum radius value traversed by the path as the narrowest diameter index of the path, and calculating the stenosis segment distribution index and the branch complexity index to generate an interventional path candidate set structure containing multiple geometric indices. Based on the candidate set structure of intervention paths, the system performs multi-index normalization of paths, weighted scoring and sorting, and three-dimensional model overlay rendering to generate an intervention planning output structure containing information on intervention points and termination points. The weighted scoring and ranking process includes: combining the normalized indicators into a comprehensive score value; recording the total path length, path curvature, narrowest path diameter, branch complexity, and weight factors corresponding to risk markers; deducting scores from paths with high-risk markers according to their risk level; adding bonus points to paths leading to the main blood supply branches based on the lesion blood supply weight in the case of multiple lesions; and sorting the paths from high to low according to the comprehensive score value after grouping by entry and target dimensions.

2. The method according to claim 1, characterized in that, The process of standardizing and registering formats also includes: The format-unified registration process includes normalizing the differences between the arterial phase sequence and the portal venous phase sequence in terms of matrix size, voxel spacing, and coordinate origin; performing three-dimensional resampling to interpolate the original volume data onto a unified voxel grid to obtain multi-phase three-dimensional volume data with consistent scale; and mapping the portal venous phase sequence to the reference coordinate system through rigid or non-rigid spatial transformation, using the coordinate system of the arterial phase sequence as the registration reference.

3. The method according to claim 1, characterized in that, The process of liver organ division also includes: The liver organ segmentation process involves inferring the range of liver grayscale values ​​and abdominal spatial range based on the patient's gender and body weight information in the basic patient information, combined with the prior of the enhanced patterns in the arterial and portal venous phases. A pre-trained organ segmentation model or probabilistic segmentation model is used to predict the class probability value for each voxel, generating an initial liver region mask. Three-dimensional connected component analysis is performed to remove isolated regions that are too small and far from the typical location of the liver, and the boundaries are corrected based on the prior of liver morphology. Local threshold refinement is also performed based on the higher contrast between the hepatic veins and the main portal vein in the portal venous phase data.

4. The method according to claim 1, characterized in that, The process of localizing the candidate region of the hepatic artery also includes: The hepatic artery candidate region localization process includes extracting the liver surface contour based on the liver region mask and estimating the approximate location of the hepatic hilum in three-dimensional space. A certain thickness spatial range around the hepatic hilum is defined as the key analysis region. Within the key analysis region, the tubular structure response value of each system is calculated to evaluate the local gray-scale distribution directionality and curvature characteristics. Threshold segmentation is performed to obtain an initial set of hepatic artery candidate voxels. Multi-phase enhancement feature screening is performed to mark candidate voxels with significant arterial phase enhancement and relatively weak portal venous phase enhancement as hepatic artery features and to remove candidate voxels that do not conform to the vascular enhancement pattern.

5. The method according to claim 1, characterized in that, The process of three-dimensional blood vessel segmentation also includes: The three-dimensional vessel segmentation process involves cropping standardized multi-phase enhanced image data within the spatial range defined by the mask structure of the hepatic artery candidate region to obtain local three-dimensional volume data. Vessel enhancement filtering is performed to enhance the response of slender tubular structures and suppress the response of plate-like or block-like structures by analyzing the local gray-level gradient direction and principal curvature characteristics. A set of intensity thresholds and connectivity conditions are selected on the vessel response intensity field to perform three-dimensional threshold segmentation and region growing. Voxels that meet the intensity conditions and are spatially connected are aggregated into vessel candidate regions. Isolated small clumps or structurally abnormal short branches are screened through connected domain analysis and volume statistics to remove clumps that do not conform to the structural characteristics of the hepatic artery tree, generating a three-dimensional vessel segmentation voxel structure.

6. The method according to claim 1, characterized in that, The centerline tracking process also includes: The centerline tracing process involves performing morphological contraction operations on the segmented voxel structure of blood vessels on a 3D mesh to generate a skeleton structure with a voxel width, obtaining a set of candidate center voxels. An undirected graph structure is constructed according to spatial connectivity. Seed nodes are selected from near the origin region of the hepatic artery. A centerline tracing strategy based on shortest path search is used to gradually expand from the seed nodes along the direction with smaller edge weights until the terminal region is reached. When a node with a degree greater than two is encountered, it is marked as a blood vessel bifurcation point, and the start point, end point, length, and average radius information of each path are recorded simultaneously. Paths with a length significantly shorter than a preset threshold and endpoints close to the segmentation edge are marked as suspicious pseudo-branches. The hepatic artery tree centerline graph structure is then constructed.

Images