Vessel topology-based functional liver volume simulation method and system for medical surgery

Abstract

The application relates to the technical field of medical image processing, in particular to a medical operation functional liver volume simulation method and system based on a blood vessel topology, and aims to solve the problems of inaccurate FRLV evaluation and neglecting the secondary ischemic effect in the prior art. The method comprises the following steps: inputting pretreated image data of an abdominal two-dimensional medical image sequence into a pre-trained deep learning segmentation network to obtain a plurality of binary segmentation masks of target tissues; extracting a blood vessel skeleton according to the binary segmentation masks and converting the blood vessel skeleton into a blood vessel topology directed graph; discretizing liver parenchyma voxels into a set of catchment cell sets belonging to different blood vessel branches based on the blood vessel topology directed graph; identifying a fracture edge cut by a virtual operation and a set of affected blood vessel branches downstream of the fracture edge to obtain a corresponding inactivated catchment cell set, calculating a corresponding inactivated volume, and calculating FRLV from the inactivated volume. The method and system provided in the application quantitatively calculate the real and effective FRLV, and improve the preoperative planning accuracy and operation safety of complex liver resection operations.

Description

Technical Field

[0001] This application relates to the field of medical image processing technology, and in particular to a method and system for simulating functional liver volume in medical surgery based on vascular topology. Background Technology

[0002] Currently, in the clinical diagnosis process, doctors mainly rely on medical imaging technologies such as computed tomography (CT) or magnetic resonance imaging (MRI) to obtain a series of discrete two-dimensional tomographic image sequences of the patient's lesion and surrounding tissues, and then carry out preoperative assessment and surgical planning based on these sequences.

[0003] However, the traditional two-dimensional image reading method has significant limitations: this method relies heavily on the doctor's spatial imagination. By reconstructing the spatial psychology of discrete two-dimensional slices, the positional relationship between the lesion and key blood vessels can be determined. This method is not only highly subjective, but also difficult to accurately identify the complex encirclement and invasion of key blood vessels such as the portal vein and hepatic vein by the tumor. It is easy for human judgment bias to lead to inaccurate preoperative planning, which may pose a hidden danger to the safety of the operation.

[0004] With the continuous development of computer-aided surgery technology, some 3D visualization workstations have appeared on the market to assist doctors in preoperative planning. However, these workstations generally suffer from the technical deficiency of emphasizing geometric display while neglecting physiological function assessment. Existing 3D visualization technologies can mostly only display static 3D models of tissues such as the liver, blood vessels, and lesions, or provide basic geometric cutting functions, which cannot meet the precise planning needs of complex liver resection surgeries.

[0005] In the planning of complex liver resection surgeries, whether the remaining liver tissue has sufficient functional blood supply is a key assessment indicator for preventing postoperative liver failure and ensuring patient recovery. Among these, the accurate assessment of Future Remnant Liver Volume (FRLV) is particularly important. Although current three-dimensional reconstruction technology can achieve three-dimensional visualization of medical images and perform virtual cutting operations, there is a fatal blind spot in the accurate assessment of FRLV volume calculation—namely, the neglect of secondary ischemic effects.

[0006] Current virtual surgical systems, when simulating liver resection, only geometrically calculate the physical volume of the liver preserved on one side of the cutting plane and use this volume as the basis for evaluating FRLV (Functionally Retained Liver Volume). This fails to consider the physiological characteristics of the liver as a solid organ highly dependent on vascular perfusion. In actual clinical anatomical procedures, if the cutting plane blocks the main portal vein or its branches supplying a region of the remaining liver, the downstream liver region, even if not physically and geometrically removed, will suffer extensive ischemia, inactivation, or even necrosis due to loss of blood supply. This region cannot perform normal liver physiological functions and should not be included in the effective FRLV.

[0007] Traditional 3D surgical planning systems lack real-time interactive simulation capabilities based on vascular anatomy and topology, making it impossible to capture the downstream ischemic effect triggered by vascular rupture. Their FRLV calculation method, which only performs geometric subtraction and does not consider the logic of vascular blood supply, can lead to FRLV assessment results obtained by doctors that are significantly higher than the actual effective functional residual liver volume. This can easily cause serious clinical misjudgments, resulting in unreasonable surgical plans, increasing the risk of serious complications such as postoperative liver failure, and posing a huge threat to the patient's life.

[0008] In summary, the field of preoperative planning in hepatobiliary surgery urgently needs a surgical simulation method and system that can combine artificial intelligence segmentation technology with vascular underlying topological logic to achieve high-precision three-dimensional reconstruction of the liver, blood vessels, and lesions, while simultaneously performing functional ischemia deduction and precise quantitative assessment of FRLV. This would address the technical pain points of existing technologies, such as inaccurate FRLV assessment, neglect of secondary ischemic effects, and high risk of clinical misjudgment, thereby improving the accuracy of preoperative planning and surgical safety in complex liver resection surgeries. Summary of the Invention

[0009] To address the aforementioned technical problems in the prior art, namely the inaccurate assessment of FRLV and the neglect of secondary ischemic effects, this application provides a method and system for simulating functional liver volume in medical surgery based on vascular topology.

[0010] In a first aspect of this application, a method for simulating functional liver volume in medical surgery based on vascular topology is provided, comprising:

[0011] A two-dimensional medical image sequence of the abdomen of a target tissue is acquired. The target tissue includes the target organ parenchyma and the internal vascular system of the target organ parenchyma. The two-dimensional medical image sequence of the abdomen is preprocessed to obtain preprocessed image data. The two-dimensional medical image sequence of the abdomen is then parsed to obtain the row pixel spacing, column pixel spacing and slice thickness.

[0012] The preprocessed image data is input into a pre-trained deep learning segmentation network, which outputs a probability distribution map of multiple target tissues, including background, liver, tumor, portal vein, and hepatic vein. The probability distribution maps of the multiple target tissues are then thresholded to obtain a binarized segmentation mask for the multiple target tissues.

[0013] Based on the binarized segmentation mask, an initial three-dimensional mesh model of each target tissue is constructed using an isosurface extraction algorithm, and the initial three-dimensional mesh model is optimized using a Laplacian smoothing algorithm and rendered to obtain a visualization model of each target tissue.

[0014] The vascular midline skeleton is extracted based on the binary segmentation mask of the blood vessel. The vascular midline skeleton includes the blood vessel bifurcation points and the direction of the blood vessel branches. The global topological connectivity of the blood vessel is represented by an adjacency list or adjacency matrix of nodes and edges. The adjacency list or adjacency matrix records the spatial connection relationship between each blood vessel bifurcation point.

[0015] Identify the blood vessel endpoint closest to the preset anatomical region, and using the blood vessel endpoint as the topological root node, convert the blood vessel midline skeleton into a directed topological graph starting from the topological root node. , where nodes This is the bifurcation point of the blood vessel, side For blood vessel branches, a graph traversal algorithm is used to label edges from top to bottom based on the direction of blood vessel inflow into the liver. The direction;

[0016] Based on the aforementioned vascular midline skeleton, the liver parenchyma voxels are discretized into a set of watershed micro-elements belonging to different vascular branches. These liver parenchyma voxels are obtained based on a binary segmentation mask of the liver, and an edge-based system is established. A hash table mapping to the corresponding set of watershed micro-elements;

[0017] A virtual cutting surface is generated in response to the input virtual surgical cutting trajectory. When the virtual cutting surface spatially intersects with the edge of the directed graph of the vascular topology, the broken edge being cut is identified. The fracture edge Let E be the edge in the directed graph of the blood vessel topology that intersects with the virtual cutting surface;

[0018] From the fracture edge Begin, along the edge The calibration direction is searched on the directed graph of the blood vessel topology to identify all vessels located at the fracture edge. downstream affected vascular branches ;

[0019] The mapping hash table is queried to retrieve the set of affected blood vessel branches. The corresponding set of inactivated watershed micro-elements, wherein the set of inactivated watershed micro-elements is the set of affected vascular branches. The set of watershed micro-elements innervated by all vascular branches in the watershed, and the set of inactivated watershed micro-elements is visualized and rendered based on the visualization model;

[0020] The functional residual liver volume (FRLV) after virtual resection of the virtual cutting surface is calculated as: original liver volume - cutting volume - inactivation volume. The cutting volume is the sum of the volumes of the liver parenchyma voxels that are determined to be physically removed on one side of the virtual cutting surface. The inactivation volume is the sum of the volumes of all voxels in the inactivation domain micro-element set. The volume of each voxel is equal to the product of the row pixel spacing, column pixel spacing, and layer thickness of the voxel.

[0021] In a second aspect of this application, a medical surgical functional liver volume simulation system based on vascular topology is provided. A preprocessing module is used to acquire a two-dimensional medical image sequence of the abdomen of a target tissue, the target tissue including the target organ parenchyma and the internal vascular system of the target organ parenchyma. The two-dimensional medical image sequence of the abdomen is preprocessed to obtain preprocessed image data, and the two-dimensional medical image sequence of the abdomen is parsed to obtain row pixel spacing, column pixel spacing and layer thickness.

[0022] The deep learning segmentation network module is used to input the preprocessed image data into a pre-trained deep learning segmentation network, output a probability distribution map of multiple target tissues, including background, liver, tumor, portal vein and hepatic vein, and perform threshold processing on the probability distribution map of multiple target tissues to obtain a binarized segmentation mask of multiple target tissues.

[0023] The 3D model construction and rendering module is used to construct an initial 3D mesh model of each target tissue based on the binarized segmentation mask using an isosurface extraction algorithm, optimize the initial 3D mesh model using a Laplacian smoothing algorithm, and render the initial mesh model to obtain a visualization model of each target tissue.

[0024] The vascular midline skeleton extraction module is used to extract the vascular midline skeleton based on the binary segmentation mask of the blood vessel. The vascular midline skeleton includes the blood vessel bifurcation points and the direction of the blood vessel branches. The global topological connectivity of the blood vessel is represented by an adjacency list or adjacency matrix of nodes and edges. The adjacency list or adjacency matrix records the spatial connection relationship between each blood vessel bifurcation point.

[0025] The topology module is used to identify the blood vessel endpoints closest to a preset anatomical region, and using these endpoints as the topological root nodes, converts the blood vessel midline skeleton into a directed topological graph. , where nodes This is the bifurcation point of the blood vessel, side For blood vessel branches, a graph traversal algorithm is used to label edges from top to bottom based on the direction of blood vessel inflow into the liver. The direction;

[0026] The watershed micro-element set partitioning module is used to discretize liver parenchyma voxels into watershed micro-element sets belonging to different vascular branches based on the vascular midline skeleton. Wherein, if the liver parenchyma voxels... With vascular branches The distance is smaller than the liver parenchyma voxel The distance from any other blood vessel branch will be used to measure the hepatic parenchyma. Belongs to vascular branches The dominant watershed micro-element set; if the liver parenchyma voxel If the minimum distances to two or more vascular branches are equal, then the average diameter or topological level of the two or more vascular branches is compared, and the liver parenchyma voxel is selected. The set of watershed micro-elements belonging to blood vessel branches with larger average diameters or higher topological levels is obtained based on the binary segmentation mask of the liver parenchyma voxels, and an edge is established. A hash table mapping to the corresponding set of watershed micro-elements;

[0027] The collision detection module generates a virtual cutting surface in response to the input virtual surgical cutting trajectory. When the virtual cutting surface spatially intersects with the edge of the directed graph of the blood vessel topology, it identifies the broken edge that has been cut. The fracture edge Let E be the edge in the directed graph of the blood vessel topology that intersects with the virtual cutting surface;

[0028] Affected vessel branch identification module, used to identify the affected vessel branch from the rupture edge. Begin, along the edge The calibration direction is searched on the directed graph of the blood vessel topology to identify all vessels located at the fracture edge. downstream affected vascular branches ;

[0029] The inactivation detection and rendering module is used to query the mapping hash table and retrieve the set of affected blood vessel branches. The corresponding set of inactivated watershed micro-elements, wherein the set of inactivated watershed micro-elements is the set of affected vascular branches. The set of watershed micro-elements innervated by all vascular branches in the watershed, and the set of inactivated watershed micro-elements is visualized and rendered based on the visualization model;

[0030] The FRLV monitoring module is used to calculate the functional remaining liver volume (FRLV) after virtual resection of the virtual cutting surface. FRLV = original liver volume - cutting volume - inactivation volume, where the cutting volume is the sum of the volumes of the liver parenchyma voxels that are determined to be physically removed on one side of the virtual cutting surface, and the inactivation volume is the sum of the volumes of all voxels in the inactivation domain micro-element set. The volume of each voxel is equal to the product of the row pixel spacing, column pixel spacing and layer thickness of the voxel.

[0031] The functional liver volume simulation method and system based on vascular topology provided in this application effectively overcomes the technical limitations of traditional virtual surgery systems that can only perform mesh geometric subtraction. It innovatively introduces vascular topology directed graph construction and watershed micro-element partitioning technology, deeply integrating the liver's anatomical structure with the logic of vascular blood supply. This enables physiological-level functional extrapolation for organs with high blood supply such as the liver, changing the current situation where existing technologies emphasize geometry while neglecting function. It elevates preoperative planning from a purely geometric level to a physiological functional level, making it more in line with actual clinical surgical needs. It significantly improves the real-time computational efficiency of preoperative planning and interactive operation. Through precise mapping of rendering parameters, it achieves spatial hierarchical display of internal vessels and lesions through a semi-transparent liver parenchyma model, allowing doctors to intuitively observe the tumor's encirclement of key blood vessels. By reducing the computational complexity of traditional whole-liver voxel traversal calculations to a topological search of vascular topology directed graph nodes, the computational load is significantly reduced. This effectively solves the technical challenge of computational lag during interactive cutting and ischemia simulation in complex liver resection surgical planning, enabling rapid response to surgeon's instructions and providing technical support for efficient preoperative planning. It also precisely addresses the critical flaw in existing techniques that neglects secondary ischemic effects in FRLV assessment, accurately capturing the downstream liver tissue ischemia effect caused by vascular rupture. Three-dimensional rendering technology displays ischemia warning areas and quantifies the true and effective functional residual liver volume (FRLV), effectively avoiding the risk of unreasonable surgical plans and serious postoperative liver failure due to overestimation of FRLV. This greatly improves the accuracy of preoperative planning and surgical safety in complex liver resection surgery, and has significant clinical implications for ensuring postoperative recovery and reducing surgical risks. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1A flowchart illustrating the functional liver volume simulation method for medical surgery based on vascular topology provided in this application embodiment;

[0034] Figure 2 This is a schematic diagram of the structure of the deep learning segmentation network (U-Net++ architecture) in the embodiments of this application;

[0035] Figure 3 A schematic diagram illustrating the interactive surgical simulation, topological fracture detection, and downstream ischemia risk extrapolation functions provided in the embodiments of this application;

[0036] Figure 4 A schematic diagram of a three-dimensional visualization interface and quantitative analysis of functional residual liver volume (FRLV) provided for embodiments of this application;

[0037] Figure 5 The structural block diagram of the functional liver volume simulation system for medical surgery based on vascular topology provided in the embodiments of this application is shown. Detailed Implementation

[0038] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0039] The following will describe in detail, with reference to the accompanying drawings, a method for simulating functional liver volume in medical surgery based on vascular topology according to an embodiment of this application. Figure 1 This is a flowchart illustrating the functional liver volume simulation method for medical surgery based on vascular topology provided in this application embodiment. (Combined with...) Figure 1 As shown in the embodiments of this application, the method for simulating functional liver volume in medical surgery based on vascular topology includes:

[0040] Step S101: Obtain the abdominal two-dimensional medical image sequence of the target tissue, the target tissue including the target organ parenchyma and the internal vascular system of the target organ parenchyma; preprocess the abdominal two-dimensional medical image sequence to obtain preprocessed image data; and parse the abdominal two-dimensional medical image sequence to obtain the row pixel spacing, column pixel spacing and slice thickness.

[0041] Step S102: Input the preprocessed image data into a pre-trained deep learning segmentation network to output a probability distribution map of multiple target tissues, including background, liver, tumor, portal vein and hepatic vein. Then, perform threshold processing on the probability distribution map of the multiple target tissues to obtain a binarized segmentation mask of the multiple target tissues.

[0042] Step S103: Based on the binarized segmentation mask, an initial three-dimensional mesh model of each target tissue is constructed using an isosurface extraction algorithm, and the initial three-dimensional mesh model is optimized using a Laplacian smoothing algorithm, and the initial mesh model is rendered to obtain a visualization model of each target tissue.

[0043] Step S104: Extract the vascular central axis skeleton based on the binary segmentation mask of the blood vessel. The vascular central axis skeleton includes the blood vessel bifurcation point and the direction of the blood vessel branches. The global topological connectivity of the blood vessel is represented by an adjacency list or adjacency matrix of nodes and edges. The adjacency list or adjacency matrix records the spatial connection relationship between each blood vessel bifurcation point.

[0044] Step S105: Identify the blood vessel endpoint closest to the preset anatomical region, and using the blood vessel endpoint as the topological root node, convert the blood vessel midline skeleton into a directed topological graph starting from the topological root node. , where nodes This is the bifurcation point of the blood vessel, side For blood vessel branches, a graph traversal algorithm is used to label edges from top to bottom based on the direction of blood vessel inflow into the liver. The direction;

[0045] Step S106: Based on the vascular central axis skeleton, the liver parenchyma voxels are discretized into a set of watershed micro-elements belonging to different vascular branches. The liver parenchyma voxels are obtained based on the binary segmentation mask of the liver, and an edge... A hash table mapping to the corresponding set of watershed micro-elements;

[0046] Step S107: In response to the input virtual surgical cutting trajectory, a virtual cutting surface is generated. When the virtual cutting surface spatially intersects with the edge of the directed graph of the vascular topology, the broken edge being cut is identified. The fracture edge Let E be the edge in the directed graph of the blood vessel topology that intersects with the virtual cutting surface;

[0047] Step S108, from the fracture edge Begin, along the edge The calibration direction is searched on the directed graph of the blood vessel topology to identify all vessels located at the fracture edge. downstream affected vascular branches ;

[0048] Step S109: Query the mapping hash table to retrieve the set of affected blood vessel branches. The corresponding set of inactivated watershed micro-elements, wherein the set of inactivated watershed micro-elements is the set of affected vascular branches. The set of watershed micro-elements innervated by all vascular branches in the watershed, and the set of inactivated watershed micro-elements is visualized and rendered based on the visualization model;

[0049] Step S110: Calculate the functional remaining liver volume FRLV after virtual resection of the virtual cutting surface = original liver volume - cutting volume - inactivation volume, where the cutting volume is the sum of the volumes of the liver parenchyma voxels that are determined to be physically removed on one side of the virtual cutting surface, the inactivation volume is the sum of the volumes of all voxels in the inactivation domain micro-element set, and the volume of the voxel is equal to the product of the row pixel spacing, column pixel spacing and layer thickness of the voxel.

[0050] Specifically, in step S101, the target tissue includes the target organ parenchyma and the internal vascular system of the target organ parenchyma, such as liver parenchyma and related hepatic vascular system. The abdominal two-dimensional medical image sequence of the target tissue can be in DICOM format. Preprocessing of the abdominal two-dimensional medical image sequence can include: using a bilateral filtering algorithm to remove noise from the abdominal two-dimensional medical image sequence while preserving the edge features of the abdominal two-dimensional medical image sequence. The bilateral filtering algorithm for noise removal can refer to conventional techniques, so it will not be elaborated here. The abdominal two-dimensional medical image sequence corresponding to each target tissue is truncated according to the HU value range corresponding to each target tissue, and the abdominal two-dimensional medical image sequence corresponding to each target tissue is linearly mapped to the [0, 1] interval, thereby completing the window width and window level adjustment and normalization. When the abdominal two-dimensional medical image sequence is used to train a deep learning segmentation network, the abdominal two-dimensional medical image sequence can also be preprocessed with online data augmentation before being used as the training dataset for the deep learning segmentation network to obtain a better deep learning segmentation network. The data augmentation can include random rotation, elastic deformation, horizontal flipping, and Gaussian noise addition, etc., to make the training data more diverse. Simultaneously, in step S101, the abdominal two-dimensional medical image sequence is parsed, for example, the header file information of the abdominal two-dimensional medical image sequence in DICOM format is parsed to obtain the row pixel spacing, column pixel spacing and slice thickness in the abdominal two-dimensional medical image sequence.

[0051] Specifically, in step S102, the structure of the deep learning segmentation network can be as follows: Figure 2 As shown, based on the U-Net++ architecture, the encoder uses ResNet as the backbone network to extract deep semantic features, and the decoder uses transposed convolutions to progressively restore spatial resolution. Nested dense skip connections are introduced between the encoder and decoder. This structure effectively reduces the semantic gap and preserves fine-grained features, thereby achieving high-precision binarization segmentation of multiple target tissues such as background, liver, tumor, portal vein, and hepatic vein. The deep learning segmentation network is pre-trained, and the optimization objective of training the deep learning segmentation network is to minimize an objective function, which can be... ,in, The loss function is the Dice coefficient. Let cross-entropy be the loss function. The Dice coefficient loss function and cross-entropy loss function are preset weight coefficients and can be referenced from conventional techniques, so they will not be elaborated here. Preprocessed image data obtained from preprocessing the abdominal two-dimensional medical image sequence is input into a deep learning segmentation network. Through multi-scale feature fusion, it automatically outputs probability distribution maps of multiple target tissues, including the background, liver, tumor, portal vein, and hepatic vein. Then, thresholding is applied to the output probability distribution maps of multiple target tissues to obtain binary segmentation masks for the multiple target tissues. Specifically, thresholding can be performed in the channel dimension, generating a binary segmentation mask for the corresponding target tissue based on the channel index with the highest probability value for each pixel in the probability distribution map. Thresholding is a common method in medical image processing, so it will not be elaborated here.

[0052] Specifically, in step S103, the isosurface extraction algorithm can be a moving cube algorithm, with the isosurface threshold set to 0.5. The moving cube algorithm traverses the voxel cubes in the three-dimensional scalar field, searches the triangular facet topology table based on the vertex state index (e.g., an index constructed based on the mask values ​​of the eight vertices of the voxel), and calculates the precise coordinates of the triangular facet vertices on the cube edges using linear interpolation. The Laplacian smoothing algorithm can be used to perform 10 iterations of Laplacian smoothing with a relaxation factor of 0.5 on the stereolithography model generated by the isosurface extraction algorithm to eliminate image jagged edges. The initial mesh model can be rendered using the Phong lighting model and a semi-transparent blending algorithm, and a three-dimensional volume can be drawn using a one-dimensional discrete mapping function as the transfer function. The transfer function takes the segmentation category label of each target tissue as the input independent variable, and outputs a four-dimensional vector corresponding to each target tissue by querying a preset optical attribute lookup table. The four-dimensional vector contains red, green, and blue color channel values ​​and alpha opacity values ​​to map the preset colors and preset alpha opacity values ​​to different target tissues in the initial mesh model, thereby realizing the spatial hierarchical display of the internal vascular system and lesion tissue through the semi-transparent liver parenchyma model.

[0053] Specifically, in step S104, the vascular midline skeleton can be extracted based on the binary segmentation mask of the blood vessels using an improved Lee-Kashyap-Chu 3D parallel topology-preserving thinning algorithm. The vascular midline skeleton includes the bifurcation points and branch directions of the blood vessels, and a single-pixel vascular midline skeleton is output. The improved Lee-Kashyap-Chu 3D parallel topology-preserving thinning algorithm includes: dividing the 3D binary segmentation mask into multiple data blocks stored contiguously in memory according to the currently available number of threads; and in each thinning iteration, using multi-threaded parallelism to independently extract the data from each block. Within a data block, removable simple points on the six directional boundaries are searched. This search can refer to the existing Lee-Kashyap-Chu 3D parallel topology-preserving refinement algorithm. After the parallel search is completed, the simple points found in each data block are merged into a global deletion list according to the data block division order, and the simple points in the global deletion list are deleted. Further, the bifurcation points and terminal points of the blood vessel branches on the vascular midline skeleton are extracted. Short blood vessel branches connecting to the terminal points and with a physical length less than a preset connectivity threshold are designated as pseudo-spicules. These pseudo-spicules are removed, and a smooth, single-pixel vascular midline skeleton that maintains global topological connectivity is output. The global topological connectivity of the blood vessel is represented by an adjacency list or adjacency matrix of nodes and edges, and the adjacency list or adjacency matrix records the spatial connection relationships between each blood vessel bifurcation point.

[0054] Specifically, to improve the accuracy of extracting the vascular midline skeleton, before extracting the vascular midline skeleton based on the vascular binarized segmentation mask, the functional liver volume simulation method for medical surgery based on vascular topology provided in this application embodiment may further include: using a 26-neighborhood connected component labeling algorithm to perform three-dimensional connected component analysis on the binarized segmentation mask of each target tissue; using a breadth-first search algorithm or a depth-first search algorithm to traverse the foreground voxels in the binarized segmentation mask; assigning the same independent connected component label to foreground voxels that are adjacent to each other in the 26-neighborhood space; counting the total number of voxels with the same independent connected component label; calculating the volume of the connected component (i.e., the volume of each voxel multiplied by the total number) based on the total number of voxels with the same independent connected component label; and for each target tissue's binarized segmentation mask, retaining the connected component with the largest corresponding volume as the main structure, and removing discrete noise regions where the total number of voxels is less than a preset volume threshold, further reducing the noise influence in the binarized segmentation mask.

[0055] Specifically, in step S105, the vascular midline skeleton is converted into a vascular topological directed graph. The process includes: extracting all vascular endpoints on the vascular midline skeleton; using the hepatic hilum anatomical region as a preset anatomical region; calculating the spatial distance from each vascular endpoint to the preset anatomical region; identifying the vascular endpoint closest to the preset anatomical region; identifying the vascular endpoint closest to the preset anatomical region as the main input point of the portal vein or hepatic vein; and using the main input point (i.e., the vascular endpoint closest to the preset anatomical region) as the topological root node; and using a graph traversal algorithm to traverse the vascular midline skeleton starting from the topological root node, defining the skeleton lines in the vascular midline skeleton as edges during the traversal. Define the bifurcation point of the blood vessel as a node. The bifurcation point of the blood vessel is the intersection of the lines connecting adjacent skeletal structures. A graph traversal algorithm is used from top to bottom to mark the direction of edge E according to the direction of blood vessel inflow into the liver, that is, to define the edge. The direction is from the parent node to the child node. This completes the construction of the directed graph of the blood vessel topology.

[0056] Specifically, in step S106, based on the vascular central axis skeleton, each independent vascular branch on the vascular central axis skeleton is used as a spatial generator of the Voronoi graphic algorithm. The shortest three-dimensional spatial distance from each voxel in the liver parenchyma voxel to each spatial generator is calculated using a three-dimensional Euclidean distance transformation on a voxel-by-voxel basis, thereby generating a three-dimensional distance field covering the entire liver parenchyma. Based on the spatial expansion and allocation mechanism of the three-dimensional distance field, the liver parenchyma voxels are discretized into a set of watershed micro-elements belonging to different vascular branches. The Voronoi graphic algorithm and the three-dimensional Euclidean distance transformation for generating the three-dimensional distance field can be referenced from existing technologies and will not be elaborated further. The spatial expansion and allocation mechanism is as follows: based on the liver parenchyma voxels... Distance minimization function to blood vessel branches Determining the relationship with liver parenchyma somatoforms nearest blood vessel branch Then the liver parenchyma body element Belongs to liver parenchyma somatic elements nearest blood vessel branch The dominant set of watershed micro-elements, Belongs to the liver parenchyma sox collection , The number of liver parenchyma voxels. Belongs to the group of vascular branches , This refers to the number of vascular branches, i.e., the number of liver parenchyma voxels. With vascular branches The distance is smaller than that of liver parenchyma voxels The distance from any other blood vessel branch will be used to measure the hepatic parenchyma. Belongs to vascular branches The dominant watershed micro-element set; if the liver parenchyma voxel If the minimum distances to two or more vascular branches are equal, then the average diameter or topological level of the two or more vascular branches is compared, and the liver parenchyma voxel is selected. The set of watershed micro-elements belongs to the vascular branch with the largest average diameter or the highest topological level. The Lee-Kashyap-Chu 3D parallel topology-preserving refinement algorithm records the average diameter of vascular segments (vascular branches, i.e., edges E) simultaneously when extracting the vascular central axis skeleton. The improved Lee-Kashyap-Chu 3D parallel topology-preserving refinement algorithm in this application does not reduce or change this function and can also simultaneously record the average diameter of vascular branches. Therefore, the average diameter can be used for partitioning the watershed micro-elements set. The topological level of the vascular branch can be calculated by traversing the directed graph of the vascular topology. Specifically, the vascular branch trunk associated with the topological root node is taken as the starting level (denoted as level 1). During the top-down graph traversal, each node... (i.e., the bifurcation point of the blood vessel), its downstream edge The topological level of a child vascular branch is incremented by one from the level of its parent vascular branch, and so on, recursively traversing the entire vascular topology tree to record the topological level of each vascular branch in the directed vascular topology graph. After obtaining the set of watershed elements governed by the vascular branches, the edge can be constructed. A hash table mapping (i.e., vascular branches) and the set of watershed micro-elements they govern.

[0057] Specifically, in step S107, the virtual surgery human-computer interaction device receives the two-dimensional cutting trajectory drawn by the user in screen space. Using a ray projection algorithm based on viewpoint matrix transformation, the discrete pixels on the two-dimensional cutting trajectory are spatially projected along the current viewing direction. This obtains a set of spatial intersection points between the projected ray and the visualization model of each target tissue, or a set of spatial intersection points between the projected ray and the depth buffer generated during the rendering of the visualization model. The depth buffer records the depth information of each pixel from the viewpoint. A surface fitting algorithm is then used to construct a freeform surface in three-dimensional space from the set of spatial intersection points. This freeform surface is used as the virtual cutting surface. The construction of the virtual cutting surface can refer to the construction of virtual cutting surfaces in existing virtual surgeries, so it will not be elaborated further. Then, the virtual cutting surface is connected to the edge of the directed graph of the vascular topology. Perform spatial intersection operation, where edges For a three-dimensional line segment connecting two adjacent nodes, if an intersection point exists, and the calculated three-dimensional spatial coordinates of the intersection point are located on the edge... The coordinates of the corresponding two nodes are used to determine the edge between the virtual cutting plane and the directed graph of the blood vessel topology. Spatial intersection occurs when the edges in the directed graph of the blood vessel topology intersect with the virtual cutting plane. Identified as a cut fracture edge .

[0058] Specifically, in step S108, a depth-first search algorithm or a breadth-first search algorithm is used on the directed graph of the vascular topology to identify fracture edges. The process traverses in a one-way manner, starting from the downstream node of the break point, where the break point is the virtual cutting surface and the break edge. The intersection point will be determined by traversing all the edges associated with the nodes on the path. Identified as a set of affected vascular branches downstream of the rupture edge .

[0059] Specifically, in step S109, querying the mapping hash table retrieves the set of inactivated watershed micro-elements corresponding to the affected vascular branch set E_downstream. This set of inactivated watershed micro-elements is the set of watershed micro-elements dominated by all vascular branches in the affected vascular branch set E_downstream. Then, the set of inactivated watershed micro-elements is visualized and rendered based on the visualization model. This can specifically include: extracting the initial material color value corresponding to the set of inactivated watershed micro-elements during the visualization rendering process; the initial material color value is preset; using a linear interpolation algorithm, the initial material color value is updated frame-by-frame to a preset ischemia warning color value within a preset number of rendering frames, to achieve a smooth transition from the material of the inactivated watershed micro-elements to the ischemia warning color. The linear interpolation algorithm used for rendering can refer to conventional techniques and will not be elaborated further. Figure 3 As shown, when the virtual cutting surface ( Figure 3 When the dashed line (as shown in the middle) intersects with a blood vessel branch and causes a collision and circuit break ( Figure 3 At the "X" mark, the fracture edge is identified, and the affected blood vessel branch set is retrieved by unidirectional traversal of the directed graph of blood vessel topology. Then, the set of inactivated watershed micro-elements dominated by the blood vessel branches in the affected blood vessel branch set can be retrieved by mapping hash table, and their material is smoothly transitioned to the ischemia warning color on the 3D model. Figure 3 (As shown in the midpoint shadow and grid shadow areas), this can break through the limitations of traditional geometric Boolean subtraction, show the dynamic deduction process of downstream ischemia based on vascular topological fracture, and further dynamically subtract the inactivated volume of this part, so as to intuitively show doctors the potential surgical risks caused by the "secondary ischemia effect".

[0060] Specifically, in step S1010, the functional residual liver volume (FRLV) after virtual resection using the virtual cutting surface is calculated. FRLV = original liver volume - cutting volume - inactivation volume. The cutting volume is the sum of the volumes of the liver parenchyma voxels on one side of the virtual cutting surface that are determined to be physically removed, i.e., the portion to be removed on one side of the virtual surgical cutting trajectory. The inactivation volume is the sum of the volumes of all voxels in the inactivation domain micro-element set. The volume corresponding to each voxel is equal to the product of the row pixel spacing, column pixel spacing, and slice thickness. The row pixel spacing, column pixel spacing, and slice thickness are obtained by analyzing the abdominal two-dimensional medical image sequence. Figure 4 As shown, the functional liver volume simulation method based on vascular topology provided in this application can intuitively display visualization results and quantitative analysis results on the human-computer interaction interface. The left main view shows the rendered three-dimensional spatial topological relationship of the liver, portal vein, hepatic vein and tumor, allowing doctors to intuitively observe the tumor's encirclement of key blood vessels. The right data panel can provide real-time feedback on current quantitative indicators, including the percentage of functional residual liver volume (FRLV), providing accurate data support for preoperative planning. It can also include data that needs to be observed, such as the physical volume of the lesion and the distance between the tumor edge and key blood vessels.

[0061] The functional liver volume simulation method for medical surgery based on vascular topology provided in this application constitutes a complete closed loop from image input to functional deduction. It breaks through the limitation of traditional virtual surgery systems that can only perform mesh geometric subtraction. It introduces vascular topology directed graph construction and watershed micro-element partitioning technology, integrates liver anatomical structure and vascular blood supply logic, and realizes physiological-level functional deduction of organs with high blood supply such as the liver. It changes the current situation of prioritizing geometry and neglecting function in existing technologies, and elevates preoperative planning from the geometric level to the physiological function level. It improves the real-time computational efficiency of preoperative planning and interactive operation, reduces the traditional whole liver voxel traversal calculation to topological search of vascular topology directed graph nodes, reduces the amount of computation, and solves the problem of computational lag during interactive cutting and ischemia deduction, so as to achieve rapid response to the doctor's operation instructions. It solves the defect of existing technologies that ignore the secondary ischemic effect in FRLV assessment, captures the downstream ischemic effect caused by vascular rupture, displays the ischemic warning area and quantifies the real and effective FRLV, avoids the surgical risk caused by overestimating FRLV, and effectively improves the planning accuracy and safety of complex liver resection surgery.

[0062] A second aspect of this application provides a medical surgical functional liver volume simulation system based on vascular topology. Figure 5 This is a structural block diagram of a functional liver volume simulation system for medical surgery based on vascular topology, provided in an embodiment of this application. (Combined with...) Figure 5As shown in the embodiments of this application, the functional liver volume simulation system for medical surgery based on vascular topology includes:

[0063] The preprocessing module 501 is used to acquire a two-dimensional medical image sequence of the abdomen of a target tissue, the target tissue including the target organ parenchyma and the internal vascular system of the target organ parenchyma, preprocess the two-dimensional medical image sequence of the abdomen to obtain preprocessed image data, and parse the two-dimensional medical image sequence of the abdomen to obtain the row pixel spacing, column pixel spacing and slice thickness.

[0064] The deep learning segmentation network module 502 is used to input the preprocessed image data into a pre-trained deep learning segmentation network, output a probability distribution map of multiple target tissues, including background, liver, tumor, portal vein and hepatic vein, and perform threshold processing on the probability distribution map of multiple target tissues to obtain a binarized segmentation mask of multiple target tissues.

[0065] The 3D model construction and rendering module 503 is used to construct an initial 3D mesh model of each target tissue based on the binarized segmentation mask using an isosurface extraction algorithm, optimize the initial 3D mesh model using a Laplacian smoothing algorithm, and render the initial mesh model to obtain a visualization model of each target tissue.

[0066] The vascular midline skeleton extraction module 504 is used to extract the vascular midline skeleton based on the binary segmentation mask of the blood vessel. The vascular midline skeleton includes the blood vessel bifurcation point and the direction of the blood vessel branches. The global topological connectivity of the blood vessel is represented by an adjacency list or adjacency matrix of nodes and edges. The adjacency list or adjacency matrix records the spatial connection relationship between each blood vessel bifurcation point.

[0067] Topology module 505 is used to identify the blood vessel endpoint closest to a preset anatomical region, and using the blood vessel endpoint as the topological root node, convert the blood vessel midline skeleton into a directed blood vessel topology graph starting from the topological root node. , where nodes This is the bifurcation point of the blood vessel, side For blood vessel branches, a graph traversal algorithm is used to label edges from top to bottom based on the direction of blood vessel inflow into the liver. The direction;

[0068] The watershed micro-element set partitioning module 506 is used to discretize the liver parenchyma voxels into watershed micro-element sets belonging to different vascular branches based on the vascular midline skeleton, wherein, if the liver parenchyma voxels With vascular branches The distance is smaller than the liver parenchyma voxel The distance from any other blood vessel branch will be used to measure the hepatic parenchyma. Belongs to vascular branches The dominant watershed micro-element set; if the liver parenchyma voxel If the minimum distances to two or more vascular branches are equal, then the average diameter or topological level of the two or more vascular branches is compared, and the liver parenchyma voxel is selected. The set of watershed micro-elements belonging to blood vessel branches with larger average diameters or higher topological levels is obtained based on the binary segmentation mask of the liver parenchyma voxels, and an edge is established. A hash table mapping to the corresponding set of watershed micro-elements;

[0069] Collision detection module 507 is used to generate a virtual cutting surface in response to the input virtual surgical cutting trajectory, and to identify the broken edge being cut when the virtual cutting surface spatially intersects with the edge of the directed graph of the blood vessel topology. The fracture edge Let E be the edge in the directed graph of the blood vessel topology that intersects with the virtual cutting surface;

[0070] Affected vessel branch identification module 508, used to identify the affected vessel branch from the rupture edge Begin, along the edge The calibration direction is searched on the directed graph of the blood vessel topology to identify all vessels located at the fracture edge. downstream affected vascular branches ;

[0071] The inactivation detection and rendering module 509 is used to query the mapping hash table and retrieve the set of affected blood vessel branches. The corresponding set of inactivated watershed micro-elements, wherein the set of inactivated watershed micro-elements is the set of affected vascular branches. The set of watershed micro-elements innervated by all vascular branches in the watershed, and the set of inactivated watershed micro-elements is visualized and rendered based on the visualization model;

[0072] The FRLV monitoring module 5010 is used to calculate the functional remaining liver volume (FRLV) after virtual resection of the virtual cutting surface: FRLV = original liver volume - cutting volume - inactivation volume. The cutting volume is the sum of the volumes of the liver parenchyma voxels that are determined to be physically removed on one side of the virtual cutting surface. The inactivation volume is the sum of the volumes of all voxels in the inactivation domain micro-element set. The volume of each voxel is equal to the product of the row pixel spacing, column pixel spacing, and layer thickness of the voxel.

[0073] The preprocessing module 501, when preprocessing the abdominal two-dimensional medical image sequence, specifically performs the following: using a bilateral filtering algorithm to remove noise from the abdominal two-dimensional medical image sequence while preserving its edge features; truncating the abdominal two-dimensional medical image sequence corresponding to each target tissue according to the HU value range corresponding to each target tissue, and linearly mapping the abdominal two-dimensional medical image sequence corresponding to each target tissue to the [0, 1] interval; when the abdominal two-dimensional medical image sequence is used to train the deep learning segmentation network, performing online data augmentation on the abdominal two-dimensional medical image sequence and using it as the training dataset for the deep learning segmentation network, wherein the data augmentation includes random rotation, elastic deformation, horizontal flipping, and Gaussian noise addition.

[0074] Specifically, the deep learning segmentation network module 502 is based on the U-Net++ architecture, where the encoder uses ResNet as the backbone network to extract deep semantic features, and the decoder uses transposed convolutions. Nested dense skip connections are introduced between the encoder and decoder. The training of the deep learning segmentation network can be optimized by minimizing an objective function, which is: ,in, The loss function is the Dice coefficient. Let cross-entropy be the loss function. These are preset weight coefficients. The deep learning segmentation network module 502 performs thresholding on the probability distribution maps of the multiple target organizations, specifically: in the channel dimension, it generates a binarized segmentation mask for the corresponding target organization based on the channel index of the pixel with the highest probability value in the probability distribution map.

[0075] Specifically, the 3D model construction and rendering module 503 is used for: the isosurface extraction algorithm being a moving cube algorithm with an isosurface threshold of 0.5; the moving cube algorithm traversing voxel cubes in the 3D scalar field, searching the triangular facet topology table based on vertex state indices, and calculating the precise coordinates of the triangular facet vertices on the cube edges using linear interpolation; the Laplacian smoothing algorithm performing 10 iterations and a relaxation factor of 0.5 on the stereolithography model generated by the isosurface extraction algorithm; and using the Phong lighting model and a semi-transparent blending algorithm to smooth the initial... The mesh model is rendered, and a three-dimensional volume is drawn using a one-dimensional discrete mapping function as the transfer function. The transfer function takes the segmentation category label of each target tissue as the input independent variable, and outputs a four-dimensional vector corresponding to each target tissue by querying a preset optical attribute lookup table. The four-dimensional vector contains red, green, and blue color channel values ​​and alpha opacity values, which are used to map the preset colors and preset alpha opacity values ​​corresponding to different target tissues in the initial mesh model, so as to realize the spatial hierarchical display of the internal vascular system and lesion tissue through the semi-transparent liver parenchyma model.

[0076] Specifically, the vessel midline skeleton extraction module 504 is used to: extract the vessel midline skeleton based on the binary segmentation mask of the vessel using an improved Lee-Kashyap-Chu 3D parallel topology-preserving refinement algorithm, and output a single-pixel vessel midline skeleton. The improved Lee-Kashyap-Chu 3D parallel topology-preserving refinement algorithm includes: dividing the 3D binary segmentation mask into multiple data blocks stored contiguously in memory according to the number of available threads; in each refinement iteration, using multi-threaded parallel independent searching for removable simple points on the six directional boundaries within each data block; after the parallel search is completed, merging the simple points found in each data block into a global deletion list according to the data block division order, and deleting the simple points in the global deletion list; extracting the vessel bifurcation points and terminal points of the vessel branches on the vessel midline skeleton; using short vessel branches connecting the terminal points and with a physical length less than a preset connectivity threshold as pseudo-spicules; removing the pseudo-spicules; and outputting a smooth single-pixel vessel midline skeleton that maintains global topological connectivity.

[0077] In one possible implementation, the functional liver volume simulation system based on vascular topology provided in this application can also include a connected component analysis and denoising module. Specifically, before extracting the vascular midline skeleton from the vascular-based binarized segmentation mask, the connected component analysis and denoising module performs the following steps: Using a 26-neighborhood connected component labeling algorithm, three-dimensional connected component analysis is performed on the binarized segmentation mask for each target tissue. A breadth-first search algorithm or a depth-first search algorithm is used to traverse the foreground voxels in the binarized segmentation mask. The same independent connected component label is assigned to foreground voxels adjacent to each other in the 26-neighborhood space, and the total number of voxels with the same independent connected component label is counted. The volume of the connected component is calculated based on the total number of voxels with the same independent connected component label. The connected component is composed of voxels with the same independent connected component label. For the binarized segmentation mask of each target tissue, the connected component with the largest corresponding volume is retained as the main structure, and discrete noise regions with a total number of voxels less than a preset volume threshold are removed.

[0078] Specifically, the topology module 505 is used to: extract all vascular endpoints on the vascular central axis skeleton, take the hepatic hilum anatomical region as a preset anatomical region, calculate the spatial distance from each of the vascular endpoints to the preset anatomical region, identify the vascular endpoint closest to the preset anatomical region as the main input point of the portal vein or hepatic vein, and use the main input point as the topology root node; starting from the topology root node, traverse the vascular central axis skeleton using a graph traversal algorithm, and during the traversal, define the skeleton lines in the vascular central axis skeleton as edges. Define the bifurcation point of the blood vessel as a node. The bifurcation point of the blood vessel is the intersection of the lines connecting adjacent skeletons, and the edges are defined. The direction is from the parent node to the child node.

[0079] Specifically, the watershed micro-element set partitioning module 506 is used to: based on the vascular central axis skeleton, treat each independent vascular branch on the vascular central axis skeleton as a spatial generator of the Voronoi graphic algorithm, use three-dimensional Euclidean distance transformation to calculate the shortest three-dimensional spatial distance from each voxel in the liver parenchyma voxel to each of the spatial generators, so as to generate a three-dimensional distance field covering the entire liver parenchyma, and discretize the liver parenchyma voxels into a watershed micro-element set belonging to different vascular branches based on the spatial expansion allocation mechanism of the three-dimensional distance field. The spatial expansion allocation mechanism is as follows: based on the liver parenchyma voxels... Distance minimization function to blood vessel branches Determining the relationship with liver parenchyma somatoforms nearest blood vessel branch Then the liver parenchyma body element Belongs to liver parenchyma somatic elements nearest blood vessel branch The dominant set of watershed micro-elements, Belongs to the liver parenchyma sox collection , The number of liver parenchyma voxels. Belongs to the group of vascular branches , The number of vascular branches; if the liver parenchyma voxel If the minimum distances to two or more vascular branches are equal, then the average diameter or topological level of the two or more vascular branches is compared, and the liver parenchyma voxel is selected. The set of watershed micro-elements belonging to the blood vessel branches with the largest average diameter or the highest topological level.

[0080] Specifically, the collision detection module 507 is used to: receive a two-dimensional cutting trajectory drawn by the user in screen space through a virtual surgical human-computer interaction device; use a ray casting algorithm based on viewpoint matrix transformation to spatially project discrete pixels on the two-dimensional cutting trajectory along the current viewing direction; obtain a set of spatial intersection points between the projected ray and the visualization model of each target tissue, or a set of spatial intersection points between the projected ray and the depth buffer generated during the rendering of the visualization model; and use a surface fitting algorithm to construct a free-form surface in three-dimensional space from the set of spatial intersection points, using the free-form surface as the virtual cutting surface; and connect the virtual cutting surface to the edge of the directed graph of the blood vessel topology. Perform spatial intersection operation, where the edges For a three-dimensional line segment connecting two adjacent nodes, if an intersection point exists, and the calculated three-dimensional spatial coordinates of the intersection point are located on the edge... The coordinates of the corresponding two nodes are used to determine the edge between the virtual cutting surface and the directed graph of the blood vessel topology. Spatial intersection occurs when the edges in the directed graph of the blood vessel topology intersect with the virtual cutting surface. Identified as a cut fracture edge .

[0081] Specifically, the affected blood vessel branch identification module 508 is used to: use a depth-first search algorithm or a breadth-first search algorithm on the directed graph of the blood vessel topology to identify the ruptured edge. The process traverses in a one-way direction, starting from the downstream node of the break point, where the break point is the intersection of the virtual cutting surface and the break edge. The intersection point will be determined by traversing all the edges associated with the nodes on the path. The group of affected vascular branches identified as being downstream of the rupture edge .

[0082] Specifically, the deactivation detection and rendering module 509, during the visualization rendering process, is used to: extract the preset initial material color value corresponding to the deactivated watershed micro-element set; and use a linear interpolation algorithm to update the initial material color value to the preset ischemia warning color value frame by frame within a preset number of rendering frames.

[0083] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the device described above can be referred to the corresponding process in the foregoing method embodiments, and therefore will not be repeated here.

[0084] The functional liver volume simulation system based on vascular topology provided in this application overcomes the limitations of traditional virtual surgical systems that can only perform mesh geometric subtraction. It introduces vascular topology directed graph construction and watershed micro-element partitioning technology, integrates liver anatomy and vascular blood supply logic, and realizes physiological-level functional extrapolation of organs with high blood supply such as the liver. This changes the current situation of prioritizing geometry over function, elevating preoperative planning from the geometric level to the physiological functional level. At the same time, it improves the real-time computational efficiency of preoperative planning and interactive operation, reducing the traditional whole liver voxel traversal calculation to vascular topology directed graph node search, which can significantly reduce the amount of computation, solve the problem of computational lag in interactive cutting and ischemia extrapolation, and quickly respond to the doctor's operation instructions. In addition, it solves the defect of existing technology that ignores the secondary ischemic effect in FRLV assessment, captures the downstream ischemic effect caused by vascular rupture, displays ischemic warning areas and quantifies the real effective FRLV, avoids overestimating the surgical risks brought by FRLV, and improves the planning accuracy and safety of complex liver resection surgery.

[0085] It should be noted that the functional liver volume simulation system based on vascular topology for medical surgery provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of this application can be further decomposed or combined. For example, the modules in the above embodiments can be merged into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of this application are only for distinguishing the various modules or steps and are not considered as an improper limitation of this application.

[0086] In practical applications, the functional liver volume simulation system based on vascular topology provided in this application may further include related hardware and software, such as: a data interface layer configured to communicate with a PACS system via the DICOM network protocol; a computing core layer containing a GPU acceleration unit for parallel execution of deep learning segmentation network operations and 3D graphics rendering pipeline operations; a logic processing layer including a segmentation module, a reconstruction module, a quantization analysis module, and a virtual surgical simulation engine; and an interactive presentation layer including a high-resolution display and human-computer interaction devices, supporting multi-viewport split-screen display and 3D view linkage. The surgical simulation engine supports "undo / redo" stack operations, allowing users to save multiple resection plans and perform multi-window side-by-side comparisons.

[0087] Those skilled in the art will recognize that the modules and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. The programs corresponding to the software modules and method steps can be placed in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art. To clearly illustrate the interchangeability of electronic hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0088] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of devices, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0089] The terms “first,” “second,” etc., are used to distinguish similar objects, not to describe or indicate a specific order or sequence. The term “comprising,” or any other similar term, is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to those processes, methods, articles, or apparatus / devices.

[0090] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.

[0091] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for simulating functional liver volume in medical surgery based on vascular topology, characterized in that, include: A two-dimensional medical image sequence of the abdomen of a target tissue is acquired. The target tissue includes the target organ parenchyma and the internal vascular system of the target organ parenchyma. The two-dimensional medical image sequence of the abdomen is preprocessed to obtain preprocessed image data. The two-dimensional medical image sequence of the abdomen is then parsed to obtain the row pixel spacing, column pixel spacing and slice thickness. The preprocessed image data is input into a pre-trained deep learning segmentation network, which outputs a probability distribution map of multiple target tissues, including background, liver, tumor, portal vein, and hepatic vein. The probability distribution maps of the multiple target tissues are then thresholded to obtain a binarized segmentation mask for the multiple target tissues. Based on the binarized segmentation mask, an initial three-dimensional mesh model of each target tissue is constructed using an isosurface extraction algorithm, and the initial three-dimensional mesh model is optimized using a Laplacian smoothing algorithm and rendered to obtain a visualization model of each target tissue. The vascular midline skeleton is extracted based on the binary segmentation mask of the blood vessel. The vascular midline skeleton includes the blood vessel bifurcation points and the direction of the blood vessel branches. The global topological connectivity of the blood vessel is represented by an adjacency list or adjacency matrix of nodes and edges. The adjacency list or adjacency matrix records the spatial connection relationship between each blood vessel bifurcation point. Identify the blood vessel endpoint closest to the preset anatomical region, and using the blood vessel endpoint as the topological root node, convert the blood vessel midline skeleton into a directed topological graph starting from the topological root node. , where nodes This is the bifurcation point of the blood vessel, side For blood vessel branches, a graph traversal algorithm is used to label edges from top to bottom based on the direction of blood vessel inflow into the liver. The direction; Based on the aforementioned vascular midline skeleton, the liver parenchyma voxels are discretized into a set of watershed micro-elements belonging to different vascular branches. These liver parenchyma voxels are obtained based on a binary segmentation mask of the liver, and an edge-based system is established. A hash table mapping to the corresponding set of watershed micro-elements; A virtual cutting surface is generated in response to the input virtual surgical cutting trajectory. When the virtual cutting surface spatially intersects with the edge of the directed graph of the vascular topology, the broken edge being cut is identified. The fracture edge Let E be the edge in the directed graph of the blood vessel topology that intersects with the virtual cutting surface; From the fracture edge Begin, along the edge The calibration direction is searched on the directed graph of the blood vessel topology to identify all vessels located at the fracture edge. Downstream affected vascular branches ; The mapping hash table is queried to retrieve the set of affected blood vessel branches. The corresponding set of inactivated watershed micro-elements, wherein the set of inactivated watershed micro-elements is the set of affected vascular branches. The set of watershed micro-elements innervated by all vascular branches in the watershed, and the set of inactivated watershed micro-elements is visualized and rendered based on the visualization model; The functional residual liver volume (FRLV) after virtual resection of the virtual cutting surface is calculated as: original liver volume - cutting volume - inactivation volume. The cutting volume is the sum of the volumes of the liver parenchyma voxels that are determined to be physically removed on one side of the virtual cutting surface. The inactivation volume is the sum of the volumes of all voxels in the inactivation domain micro-element set. The volume of each voxel is equal to the product of the row pixel spacing, column pixel spacing, and layer thickness of the voxel.

2. The method for simulating functional liver volume in medical surgery based on vascular topology according to claim 1, characterized in that, The preprocessing of the abdominal two-dimensional medical image sequence includes: The bilateral filtering algorithm is used to remove noise from the abdominal two-dimensional medical image sequence while preserving the edge features of the abdominal two-dimensional medical image sequence; The abdominal two-dimensional medical image sequences corresponding to each target tissue are truncated according to the HU value range corresponding to each target tissue, and the abdominal two-dimensional medical image sequences corresponding to each target tissue are linearly mapped to the [0, 1] interval. When the abdominal two-dimensional medical image sequence is used to train the deep learning segmentation network, the abdominal two-dimensional medical image sequence is used as the training dataset of the deep learning segmentation network after online data augmentation. The data augmentation includes random rotation, elastic deformation, horizontal flipping, and Gaussian noise addition.

3. The method for simulating functional liver volume in medical surgery based on vascular topology according to claim 1, characterized in that, The deep learning segmentation network is based on the U-Net++ architecture, in which the encoder uses ResNet as the backbone network to extract deep semantic features, and the decoder uses transposed convolution, introducing nested dense skip connections between the encoder and decoder. The optimization objective of training the deep learning segmentation network is to minimize the objective function, which is: ,in, The loss function is the Dice coefficient. Let cross-entropy be the loss function. These are preset weighting coefficients; The thresholding process for the probability distribution maps of the multiple target organizations includes: in the channel dimension, generating a binarized segmentation mask for the corresponding target organization based on the channel index of the probability distribution map where each pixel has the highest probability value.

4. The method for simulating functional liver volume in medical surgery based on vascular topology according to claim 1, characterized in that, The process of constructing an initial 3D mesh model for each target tissue based on the binarized segmentation mask using an isosurface extraction algorithm, optimizing the initial 3D mesh model using a Laplacian smoothing algorithm, and rendering the initial mesh model includes: The isosurface extraction algorithm is a moving cube algorithm with an isosurface threshold of 0.

5. The moving cube algorithm traverses the voxel cubes in the three-dimensional scalar field, searches the triangular facet topology table based on the vertex state index, and uses linear interpolation to calculate the precise coordinates of the triangular facet vertices on the cube edges. The Laplacian smoothing algorithm performs Laplacian smoothing on the stereolithography model generated by the isosurface extraction algorithm with 10 iterations and a relaxation factor of 0.

5. The initial mesh model is rendered using the Phong lighting model and a semi-transparent blending algorithm. A one-dimensional discrete mapping function is used as the transfer function for three-dimensional volume rendering. This transfer function takes the segmentation category labels of each target tissue as input variables and outputs a four-dimensional vector corresponding to each target tissue by querying a preset optical attribute lookup table. The four-dimensional vector contains red, green, and blue color channel values ​​and an alpha opacity value, which are used to map preset colors and preset alpha opacity values ​​to different target tissues in the initial mesh model.

5. The method for simulating functional liver volume in medical surgery based on vascular topology according to claim 1, characterized in that, The extraction of the vascular midline skeleton based on the vascular binary segmentation mask includes: The blood vessel midline skeleton is extracted using a binarized segmentation mask based on blood vessels and an improved Lee-Kashyap-Chu 3D parallel topology-preserving thinning algorithm, outputting a single-pixel blood vessel midline skeleton. The improved Lee-Kashyap-Chu 3D parallel topology-preserving thinning algorithm includes: The three-dimensional binarized segmentation mask is divided into multiple data blocks that are stored contiguously in memory based on the number of available threads. In each refinement iteration, deletable simple points on the six-directional boundaries are searched independently and in parallel using multi-threaded parallelism within each data block. After the parallel search is completed, the simple points found in each data block are merged into the global deletion list according to the data block division order, and the simple points in the global deletion list are deleted. Extract the bifurcation points and terminal points of the blood vessels on the central axis skeleton of the blood vessels. Short blood vessel branches that connect to the terminal points and have a physical length less than a preset connectivity threshold are taken as pseudo-burrs. Remove the pseudo-burrs and output a smooth single-pixel central axis skeleton of the blood vessels that maintains global topological connectivity.

6. The method for simulating functional liver volume in medical surgery based on vascular topology according to claim 1, characterized in that, Before extracting the vascular midline skeleton based on the vascular binarization segmentation mask, the following steps are also included: A 26-neighborhood-based connected component labeling algorithm is used to perform three-dimensional connected component analysis on the binarized segmentation mask of each target organization. A breadth-first search algorithm or a depth-first search algorithm is used to traverse the foreground voxels in the binarized segmentation mask. Foreground voxels that are adjacent to each other in the 26-neighborhood space are assigned the same independent connected component label, and the total number of voxels with the same independent connected component label is counted. The volume of the connected component is calculated based on the total number of voxels with the same independent connected component label. The connected component is composed of voxels with the same independent connected component label. For each target organization, the binarized segmentation mask retains the connected component with the largest corresponding volume as the main structure, and removes discrete noise regions with a total number of voxels less than a preset volume threshold.

7. The method for simulating functional liver volume in medical surgery based on vascular topology according to claim 5, characterized in that, The process of converting the vascular midline skeleton into a vascular topological directed graph. ,include: Extract all vascular endpoints on the vascular central axis skeleton, take the hepatic hilum anatomical region as the preset anatomical region, calculate the spatial distance from each of the vascular endpoints to the preset anatomical region, identify the vascular endpoint closest to the preset anatomical region as the main input point of the portal vein or hepatic vein, and use the main input point as the topological root node. Starting from the root node of the topology, a graph traversal algorithm is used to traverse the vascular midline skeleton. During the traversal, the skeleton lines connecting the vascular midline skeleton are defined as edges. Define the bifurcation point of the blood vessel as a node. The bifurcation point of the blood vessel is the intersection of the lines connecting adjacent skeletons, and the edges are defined. The direction is from the parent node to the child node.

8. The method for simulating functional liver volume in medical surgery based on vascular topology according to claim 7, characterized in that, Based on the vascular midline framework, the liver parenchyma voxels are discretized into a set of watershed micro-elements belonging to different vascular branches, including: Based on the vascular midline skeleton, each independent vascular branch on the vascular midline skeleton is used as a spatial generator of the Voronoi graphic algorithm. The shortest three-dimensional spatial distance from each voxel in the liver parenchyma voxel to each of the spatial generators is calculated using three-dimensional Euclidean distance transformation on a voxel-by-voxel basis to generate a three-dimensional distance field covering the entire liver parenchyma. Based on the spatial expansion and allocation mechanism of the three-dimensional distance field, the liver parenchyma voxels are discretized into a set of watershed micro-elements belonging to different vascular branches. The spatial expansion allocation mechanism is as follows: based on hepatocytes... Distance minimization function to blood vessel branches Determining the relationship with liver parenchyma somatoforms nearest blood vessel branch Then the liver parenchyma body element Belongs to liver parenchyma somatic elements nearest blood vessel branch The dominant set of watershed micro-elements, Belongs to the liver parenchyma somatoform set , The number of liver parenchyma voxels. Belongs to the group of vascular branches , The number of vascular branches; if the liver parenchyma voxel If the minimum distances to two or more vascular branches are equal, then the average diameter or topological level of the two or more vascular branches is compared, and the liver parenchyma voxel is selected. The set of watershed micro-elements belonging to the blood vessel branches with the largest average diameter or the highest topological level.

9. The method for simulating functional liver volume in medical surgery based on vascular topology according to any one of claims 1-8, characterized in that, The virtual surgical cutting trajectory is generated in response to the input, forming a virtual cutting surface. When the virtual cutting surface spatially intersects with the edge of the directed graph of the vascular topology, the broken edge being cut is identified. ,include: The virtual surgical human-computer interaction device receives the two-dimensional cutting trajectory drawn by the user in the screen space. Using a ray projection algorithm based on viewpoint matrix transformation, the discrete pixels on the two-dimensional cutting trajectory are spatially projected along the current viewing direction. The set of spatial intersection points between the projected ray and the visualization model of each target tissue is obtained, or the set of spatial intersection points between the projected ray and the depth buffer generated when the visualization model is rendered is obtained. The set of spatial intersection points is then constructed into a free surface in three-dimensional space using a surface fitting algorithm. The free surface is used as the virtual cutting surface. Connect the virtual cutting surface to the edges of the directed graph of the blood vessel topology. Perform spatial intersection operation, where the edges For a three-dimensional line segment connecting two adjacent nodes, if an intersection point exists, and the calculated three-dimensional spatial coordinates of the intersection point are located on the edge... The coordinates of the corresponding two nodes are used to determine the edge between the virtual cutting surface and the directed graph of the blood vessel topology. Spatial intersection occurs when the edges in the directed graph of the blood vessel topology intersect with the virtual cutting surface. Identified as a cut fracture edge .

10. The method for simulating functional liver volume in medical surgery based on vascular topology according to claim 9, characterized in that, The from the fracture edge Initially, a search is performed on the directed graph of the blood vessel topology to identify all vessels located at the fracture edges. Downstream affected vascular branches ,include: A depth-first search algorithm or a breadth-first search algorithm is used on the directed graph of the blood vessel topology, with the fracture edges as the starting point. The process traverses in a one-way direction, starting from the downstream node of the break point, where the break point is the intersection of the virtual cutting surface and the break edge. The intersection point will be determined by traversing all the edges associated with the nodes on the path. The group of affected vascular branches identified as being downstream of the rupture edge .

11. The method for simulating functional liver volume in medical surgery based on vascular topology according to claim 10, characterized in that, The visualization rendering of the deactivated watershed micro-element set based on the visualization model includes: extracting the preset initial material color value corresponding to the deactivated watershed micro-element set during the visualization rendering process; and using a linear interpolation algorithm to update the initial material color value to the preset ischemia warning color value frame by frame within a preset number of rendering frames.

12. A functional liver volume simulation system for medical surgery based on vascular topology, characterized in that, include: The preprocessing module is used to acquire a two-dimensional medical image sequence of the abdomen of the target tissue, which includes the target organ parenchyma and the internal vascular system of the target organ parenchyma. The preprocessing module performs preprocessing on the two-dimensional medical image sequence of the abdomen to obtain preprocessed image data, and parses the two-dimensional medical image sequence of the abdomen to obtain the row pixel spacing, column pixel spacing and slice thickness. The deep learning segmentation network module is used to input the preprocessed image data into a pre-trained deep learning segmentation network, output a probability distribution map of multiple target tissues, including background, liver, tumor, portal vein and hepatic vein, and perform threshold processing on the probability distribution map of multiple target tissues to obtain a binarized segmentation mask of multiple target tissues. The 3D model construction and rendering module is used to construct an initial 3D mesh model of each target tissue based on the binarized segmentation mask using an isosurface extraction algorithm, optimize the initial 3D mesh model using a Laplacian smoothing algorithm, and render the initial mesh model to obtain a visualization model of each target tissue. The vascular midline skeleton extraction module is used to extract the vascular midline skeleton based on the binary segmentation mask of the blood vessel. The vascular midline skeleton includes the blood vessel bifurcation points and the direction of the blood vessel branches. The global topological connectivity of the blood vessel is represented by an adjacency list or adjacency matrix of nodes and edges. The adjacency list or adjacency matrix records the spatial connection relationship between each blood vessel bifurcation point. The topology module is used to identify the blood vessel endpoints closest to a preset anatomical region, and using these endpoints as the topological root nodes, converts the blood vessel midline skeleton into a directed topological graph. , where nodes This is the bifurcation point of the blood vessel, side For blood vessel branches, a graph traversal algorithm is used to label edges from top to bottom based on the direction of blood vessel inflow into the liver. The direction; The watershed micro-element set partitioning module is used to discretize liver parenchyma voxels into watershed micro-element sets belonging to different vascular branches based on the vascular midline skeleton. The liver parenchyma voxels are obtained based on the binary segmentation mask of the liver, and an edge is established. A hash table mapping to the corresponding set of watershed micro-elements; The collision detection module generates a virtual cutting surface in response to the input virtual surgical cutting trajectory. When the virtual cutting surface spatially intersects with the edge of the directed graph of the blood vessel topology, it identifies the broken edge that has been cut. The fracture edge Let E be the edge in the directed graph of the blood vessel topology that intersects with the virtual cutting surface; Affected vessel branch identification module, used to identify the affected vessel branch from the rupture edge. Begin, along the edge The calibration direction is searched on the directed graph of the blood vessel topology to identify all vessels located at the fracture edge. Downstream affected vascular branches ; The inactivation detection and rendering module is used to query the mapping hash table and retrieve the set of affected blood vessel branches. The corresponding set of inactivated watershed micro-elements, wherein the set of inactivated watershed micro-elements is the set of affected vascular branches. The set of watershed micro-elements innervated by all vascular branches in the watershed, and the set of inactivated watershed micro-elements is visualized and rendered based on the visualization model; The FRLV monitoring module is used to calculate the functional remaining liver volume (FRLV) after virtual resection of the virtual cutting surface. FRLV = original liver volume - cutting volume - inactivation volume, where the cutting volume is the sum of the volumes of the liver parenchyma voxels that are determined to be physically removed on one side of the virtual cutting surface, and the inactivation volume is the sum of the volumes of all voxels in the inactivation domain micro-element set. The volume of each voxel is equal to the product of the row pixel spacing, column pixel spacing and layer thickness of the voxel.

13. The functional liver volume simulation system for medical surgery based on vascular topology according to claim 12, characterized in that, The watershed micro-element set partitioning module is specifically used for: Based on the aforementioned vascular midline framework, each independent vascular branch on the framework is used as a spatial generator of the Voronoi diagram algorithm. A three-dimensional Euclidean distance transformation is used to calculate the shortest three-dimensional spatial distance from each voxel in the liver parenchyma to each of the aforementioned spatial generators, thereby generating a three-dimensional distance field covering the entire liver parenchyma. Based on the spatial expansion and allocation mechanism of the three-dimensional distance field, the liver parenchyma voxels are discretized into a set of watershed micro-elements belonging to different vascular branches. The spatial expansion allocation mechanism is as follows: based on hepatocytes... Distance minimization function to blood vessel branches Determining the relationship with liver parenchyma somatoforms nearest blood vessel branch Then the liver parenchyma body element Belongs to liver parenchyma somatic elements nearest blood vessel branch The dominant set of watershed micro-elements, Belongs to the liver parenchyma somatoform set , The number of liver parenchyma voxels. Belongs to the group of vascular branches , The number of vascular branches; if the liver parenchyma voxel If the minimum distances to two or more vascular branches are equal, then the average diameter or topological level of the two or more vascular branches is compared, and the liver parenchyma voxel is selected. The set of watershed micro-elements belonging to the blood vessel branches with the largest average diameter or the highest topological level.

Images