Liver cancer interventional surgery path dynamic planning method based on multi-modal image fusion

Abstract

The application relates to the technical field of computer-aided interventional therapy, and discloses a liver cancer interventional surgery path dynamic planning method based on multi-modal image fusion. In view of the problem that the prior art relies on preoperative static image planning and cannot adapt to the dynamic change of organs in operation, leading to precision reduction, the application proposes the following: a fine three-dimensional model is constructed by fusing computer tomography, magnetic resonance imaging and other multi-modal images; the model is dynamically aligned with real-time ultrasound or digital subtraction angiography through a fast non-rigid registration algorithm during operation, a navigation view reflecting the anatomic change in real time is generated; and based on the dynamic view and combined with instrument constraints, a path search algorithm is used to plan and display an optimal and collision-free puncture path in real time. The application realizes the upgrade of the surgery path from static presetting to dynamic self-adaption, provides accurate and intuitive real-time decision support for the operator, and significantly improves the targeting accuracy, safety and efficiency of the liver cancer interventional surgery.

Description

Technical Field

[0001] This invention relates to the field of computer-assisted interventional therapy technology, particularly to interventional surgery for liver cancer. Background Technology

[0002] This invention relates to the field of computer-aided interventional therapy technology, and in particular to a dynamic planning method for interventional surgical pathways in liver cancer based on multimodal image fusion.

[0003] Interventional surgery for liver cancer, especially minimally invasive therapies such as percutaneous ablation, is an important clinical treatment method. The key to its success lies in precisely guiding the treatment device to the tumor target while avoiding the complex vascular system. To this end, current techniques typically involve constructing a static three-dimensional model of the patient's liver preoperatively based on high-resolution images such as computed tomography (CT) or magnetic resonance imaging (MRI), and pre-planning a theoretically optimal puncture path on this model as a guiding reference during the procedure. This planning model based on preoperative static data forms the basis of current image-guided interventional surgery.

[0004] However, the aforementioned technical approaches relying on static models for planning have fundamental flaws. The liver, as a soft and highly mobile organ, undergoes significant and complex non-rigid deformations during surgery due to various factors such as the patient's breathing and the interaction of surgical instruments. This leads to severe spatial discrepancies between the preoperative planned path and the real-time anatomical structure, greatly diminishing the guiding value of the static model. Although real-time two-dimensional imaging, such as ultrasound, can be used for intraoperative monitoring, its limited field of view and inability to visually represent complete three-dimensional spatial relationships make it difficult for surgeons to accurately determine the relative positions of instruments to deep, complex structures (especially blood vessels). Surgical precision and safety heavily depend on the surgeon's personal experience. More critically, existing path planning methods are almost entirely limited to geometric obstacle avoidance—finding the shortest path that avoids collisions with major blood vessels—completely ignoring the biomechanical damage to liver tissue caused by the puncture process itself. This invisible damage is closely related to the risk of postoperative complications. Therefore, how to compensate for the dynamic deformation of the organ in real-time and accurately, and how to establish a dynamic path planning method that can comprehensively assess geometric paths, anatomical safety, and the risk of biomechanical damage, are current technical challenges to be solved in this field. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] Therefore, this invention provides a dynamic planning method for interventional surgery path in liver cancer based on multimodal image fusion. This solves the comprehensive technical problems of existing technologies, which rely on static image planning and cannot compensate for the dynamic non-rigid deformation of organs during surgery in real time, resulting in a serious decrease in navigation accuracy; path planning is limited to geometric obstacle avoidance and lacks assessment of biomechanical damage; at the same time, the surgeon has to bear an excessively high cognitive load and lacks intuitive three-dimensional decision support.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] This invention provides a dynamic planning method for interventional surgical pathways in liver cancer based on multimodal image fusion, which includes the following steps:

[0009] Step 1: In the preoperative stage, multimodal imaging data such as the patient's computed tomography scan and magnetic resonance imaging are integrated to construct a personalized four-dimensional biomechanical model that integrates the anatomical geometry and biophysical characteristics of tumors, vascular systems and other structures.

[0010] Step 2: During the operation, acquire two-dimensional intraoperative images of the patient in real time, identify and continuously track anatomical feature points from them, and at the same time acquire the real-time three-dimensional spatial position of the surgical instrument tip through a spatial positioning system.

[0011] Step 3: The two-dimensional real-time position of anatomical feature points and the three-dimensional real-time position of surgical instruments are used as multi-source dynamic constraints to jointly drive the personalized four-dimensional biomechanical model to perform physical simulation calculations, so as to calculate the three-dimensional non-rigid deformation field of the liver caused by respiratory movement and instrument interaction in real time.

[0012] Step 4: Use the three-dimensional non-rigid deformation field of the liver to reconstruct the preoperative model in three dimensions, and generate a dynamic physical navigation model that accurately reflects the true shape, position and internal stress distribution of the current anatomical structure.

[0013] Step 5: On the dynamic physical navigation model, a multidimensional dynamic cost function is defined, which weights and fuses the path geometric length, the safety margin of key vessels, and the tissue strain energy accumulated along the path; and a stochastic path optimization method is used to iteratively optimize under the guidance of the multidimensional dynamic cost function to plan an optimal surgical path from the body surface to the tumor target in real time that can minimize the risk of tissue damage caused by puncture, and the optimal surgical path is visualized in augmented reality.

[0014] As a preferred embodiment of the dynamic planning method for interventional surgery path of liver cancer based on multimodal image fusion described in this invention, the step of constructing a personalized four-dimensional biomechanical model integrating anatomical geometry and biophysical characteristics in step 1 is further specified as follows: spatial registration and fusion of multimodal image data acquired by the patient before surgery are performed to generate a fused image dataset in a unified coordinate system. Subsequently, an automated image segmentation method is used to simultaneously identify and reconstruct multiple targets such as liver parenchyma, tumor lesions and internal multi-level vascular system in the fused image dataset, thereby generating a high-precision geometric model representing the anatomical morphology.

[0015] Within the liver volume defined by the high-precision geometric model, a finite element volume mesh with adaptive spatial resolution is generated. The element density of the finite element volume mesh is locally adjusted according to the complexity of the anatomical structure. At the same time, biomechanical property parameters representing the spatial variation of liver tissue heterogeneity are extracted and calculated from the registered image data. The biomechanical property parameters are assigned to the finite element volume mesh, thereby constructing a static biomechanical model that integrates precise geometric morphology and heterogeneous physical properties.

[0016] To endow the static biomechanical model with dynamic characteristics, a four-dimensional image sequence covering the patient's complete respiratory cycle is used to capture and quantify the continuous three-dimensional deformation of the liver during respiration. By statistically modeling this continuous deformation, its main motion patterns are extracted and a dynamic model that can characterize individualized respiratory motion patterns is constructed. This dynamic model is coupled with the static biomechanical model to finally form the personalized four-dimensional biomechanical model.

[0017] As a preferred embodiment of the dynamic planning method for interventional surgery path of liver cancer based on multimodal image fusion described in this invention, the step of identifying and tracking anatomical feature points and simultaneously acquiring the spatial position of surgical instruments during the operation is further specified as follows: after acquiring the patient's two-dimensional intraoperative image stream in real time, the anatomical landmarks predefined in the personalized four-dimensional biomechanical model are mapped to the two-dimensional image space to dynamically define the corresponding prediction search area for the anatomical landmarks. Then, within the prediction search area, the anatomical landmarks are matched and identified according to their pre-stored visual appearance features to complete the localization. For continuous image frames, a temporal motion model is established to predict the subsequent position of the feature points, so as to use the predicted subsequent position to correct and maintain the tracking process.

[0018] Meanwhile, the step of obtaining the real-time three-dimensional spatial position of the surgical instrument end via the spatial positioning system in step 2 specifically involves continuously calculating the real-time spatial pose of the instrument in the navigation coordinate system. In order to ensure the spatiotemporal consistency between image data and instrument pose data, real-time information from different data sources is processed for time synchronization and correlation. Finally, these spatiotemporally aligned heterogeneous data are integrated into a multi-source dynamic constraint set for driving physical simulation.

[0019] As a preferred embodiment of the dynamic planning method for interventional surgery path of liver cancer based on multimodal image fusion described in this invention, the step of jointly driving the personalized four-dimensional biomechanical model to perform physical simulation calculation in step 3 to calculate the three-dimensional non-rigid deformation field in real time is further specified as follows: the multi-source dynamic constraint set is analyzed, and based on this, boundary conditions and internal constraints are set on the personalized four-dimensional biomechanical model to limit its current state.

[0020] The physical response of the biomechanical model under the combined action of the boundary conditions and internal constraints is calculated using a numerical solver. The calculation follows a preset biological tissue constitutive relation. The calculation process aims to find the unique physical equilibrium state that simultaneously satisfies the boundary conditions, internal constraints, and biological tissue constitutive relation, thereby determining the three-dimensional displacement vector of the discrete unit nodes in the model, thus forming the three-dimensional non-rigid deformation field.

[0021] As a preferred embodiment of the dynamic planning method for interventional surgery path of liver cancer based on multimodal image fusion described in this invention, the step 4, which involves reconstructing the preoperative model in three dimensions based on the real-time calculated non-rigid deformation field to generate a dynamic physical navigation model, is further specified as follows: the three-dimensional non-rigid deformation field of the liver is used as a spatial transformation function and applied to the reference geometry of the personalized four-dimensional biomechanical model. This application process updates the spatial coordinates of all discrete unit nodes by applying their corresponding three-dimensional displacement vectors in the deformation field to the discrete unit nodes in the model, thereby reconstructing a transient geometric configuration that can accurately reflect the current macroscopic shape and position of the liver.

[0022] Based on the transient geometric configuration, the internal physical state changes caused by tissue deformation are calculated by further utilizing the three-dimensional non-rigid deformation field of the liver and the pre-set constitutive relation in the personalized four-dimensional biomechanical model. This calculation is performed for each discrete element in the model to solve the stress tensor and strain tensor of the model under the current deformation state.

[0023] The spatial coordinate information of the transient geometric configuration is fused with the stress tensor and strain tensor calculated for all discrete elements in the model. This fusion process integrates the spatial coordinate information and the stress tensor and strain tensor into a unified data entity, thereby forming the dynamic physical navigation model.

[0024] As a preferred embodiment of the dynamic planning method for interventional surgery path of liver cancer based on multimodal image fusion described in this invention, the step of planning and visualizing the path on the dynamic physical navigation model in step 5 is further specified as including: constructing a multidimensional dynamic cost function for any candidate path in the dynamic physical navigation model. The construction process includes quantifying the geometric properties of the path, the safety relationship with key anatomical structures, and the biomechanical risk of tissue damage, and integrating these quantified risk dimensions into a comprehensive cost value according to a preset fusion rule.

[0025] Using the aforementioned random path optimization method, iterative path exploration is performed within the space defined by the dynamic physical navigation model. The exploration process is characterized by being guided by the aforementioned comprehensive cost value to prioritize expansion in low-risk areas. Furthermore, the exploration strategy of this method is adaptive, capable of dynamically adjusting its exploration behavior according to the local risk level of the area to balance global search efficiency and local obstacle avoidance accuracy.

[0026] Furthermore, the optimal surgical path obtained through iterative path exploration is used to generate perceptible safety guidance information for augmented reality visualization. The perceptible attributes of the safety guidance information are dynamically modulated according to the comprehensive cost value corresponding to each point along the path to warn of changes in risk in the path. Finally, the safety guidance information is spatially registered with the navigation coordinate system and fused and overlaid with the surgeon's perceived scene for display.

[0027] As a preferred embodiment of the dynamic planning method for interventional surgery path of liver cancer based on multimodal image fusion described in this invention, the step of integrating according to preset fusion rules is further specified as follows: the quantified geometric attributes, the safety relationship with key anatomical structures, and the tissue damage risk based on biomechanics are jointly defined as a risk dimension set.

[0028] The fusion rule is a dynamic weighting mechanism used to assign weight coefficients to each dimension in the risk dimension set. The determination of these weight coefficients is based on contextual analysis of the local area where the candidate path is located in the dynamic physical navigation model. This contextual analysis aims to assess the risk level of the anatomical environment surrounding the candidate path, and adaptively adjusts the weight coefficients assigned to each dimension in the risk dimension set based on the assessed risk level, so that the relative importance of each risk dimension can change dynamically in different anatomical environments.

[0029] As a preferred embodiment of the dynamic planning method for interventional surgery path of liver cancer based on multimodal image fusion described in this invention, the step of calculation by a numerical solver further includes an event-driven mechanism, which specifically includes: the system continuously monitors and analyzes the real-time spatial pose of the surgical instruments obtained by the spatial positioning system; when the analysis determines that the change characteristics of the spatial pose conform to the preset event model representing the surgeon's active intervention, the system triggers a recalculation of the physical simulation;

[0030] The recalculation process involves using the spatial pose of the surgical instruments at the end of the active intervention event as the boundary condition, and resolving and updating the three-dimensional non-rigid deformation field of the liver according to the calculation process described in claim 4. The updated three-dimensional non-rigid deformation field of the liver is then used to update the dynamic physical navigation model and the optimal surgical path planned on the model.

[0031] As a preferred embodiment of the dynamic planning method for interventional surgical pathways for liver cancer based on multimodal image fusion described in this invention, the augmented reality visualization process of the optimal surgical pathway further includes:

[0032] From the dynamic physical navigation model, a dynamically updated three-dimensional risk field model is constructed, representing the distribution of the comprehensive risk value in the three-dimensional space around the optimal surgical path. The three-dimensional risk field model is presented in the form of a three-dimensional visualization object. The perceptible attributes of the visualization object correspond to the risk level in space, and the display range and level of detail of the three-dimensional visualization object can be interactively adjusted by the surgeon. The guidance information of the optimal surgical path is combined with the adjusted three-dimensional visualization object, and the combined information is fused and superimposed with the surgeon's perceived scene for display.

[0033] As a preferred embodiment of the dynamic planning method for interventional surgery path of liver cancer based on multimodal image fusion described in this invention, the method for constructing a personalized four-dimensional biomechanical model further includes a step of online adaptive calibration of the model during the operation. This step specifically involves: during the operation, continuously acquiring the image positions of anatomical feature points predicted by the personalized four-dimensional biomechanical model and the corresponding image positions of anatomical feature points observed by real-time intraoperative images, and comparing these two sets of position data to generate a deviation signal;

[0034] When the deviation signal meets the preset triggering conditions, the system initiates a reverse optimization of the model parameters. This optimization process aims to adjust the spatially varying biomechanical property parameters in the static biomechanical model to reduce the inconsistency between the model's predicted location and the actual observed location. The output of this optimization process is used to update the biomechanical property parameters in the personalized four-dimensional biomechanical model.

[0035] The beneficial effects of this invention are as follows: This invention significantly improves the targeting accuracy of punctures by precisely compensating for the real-time non-rigid deformation of the liver during surgery caused by respiratory movements and instrument interactions through a dynamically driven four-dimensional biomechanical model. Furthermore, this invention achieves a qualitative leap in path planning. By incorporating biomechanical damage risk into the core optimization dimension, the planned path not only safely avoids blood vessels but also minimizes tissue trauma from a physical perspective, which is of great value in reducing the risk of postoperative complications. Finally, this invention uses augmented reality technology to intuitively present the dynamic optimal path and three-dimensional risk field to the surgeon, significantly reducing the difficulty of surgical procedures and cognitive load, thereby effectively improving surgical efficiency and safety. Attached Figure Description

[0036] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a schematic diagram of the overall process of a dynamic planning method for interventional surgical pathways in liver cancer based on multimodal image fusion.

[0038] Figure 2 This is a detailed flowchart illustrating the steps involved in constructing a personalized four-dimensional biomechanical model in a dynamic planning method for interventional surgical pathways in liver cancer based on multimodal image fusion.

[0039] Figure 3This is a detailed flowchart illustrating the dynamic path planning and visualization steps in a dynamic path planning method for interventional surgery of liver cancer based on multimodal image fusion. Detailed Implementation

[0040] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0041] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0042] Secondly, the term "one embodiment" or "example" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the invention. The appearance of an embodiment in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments.

[0043] Example 1

[0044] Reference Figure 1 This is an embodiment of the present invention, illustrating the overall process of the invention. The present invention proposes a dynamic planning method for interventional surgical pathways in liver cancer based on multimodal image fusion, characterized by comprising the following steps:

[0045] Step 1: In the preoperative stage, multimodal imaging data such as the patient's computed tomography scan and magnetic resonance imaging are integrated to construct a personalized four-dimensional biomechanical model that integrates the anatomical geometry and biophysical characteristics of tumors, vascular systems and other structures.

[0046] Step 2: During the operation, acquire two-dimensional intraoperative images of the patient in real time, identify and continuously track anatomical feature points from them, and at the same time acquire the real-time three-dimensional spatial position of the surgical instrument tip through a spatial positioning system.

[0047] Step 3: The two-dimensional real-time position of anatomical feature points and the three-dimensional real-time position of surgical instruments are used as multi-source dynamic constraints to jointly drive the personalized four-dimensional biomechanical model to perform physical simulation calculations, so as to calculate the three-dimensional non-rigid deformation field of the liver caused by respiratory movement and instrument interaction in real time.

[0048] Step 4: Use the three-dimensional non-rigid deformation field of the liver to reconstruct the preoperative model in three dimensions, and generate a dynamic physical navigation model that accurately reflects the true shape, position and internal stress distribution of the current anatomical structure.

[0049] Step 5: On the dynamic physical navigation model, a multidimensional dynamic cost function is defined, which weights and fuses the path geometric length, the safety margin of key vessels, and the tissue strain energy accumulated along the path; and a stochastic path optimization method is used to iteratively optimize under the guidance of the multidimensional dynamic cost function, and to plan in real time an optimal surgical path from the body surface to the tumor target that can minimize the risk of tissue damage caused by puncture, and to visualize the optimal surgical path using augmented reality.

[0050] The procedure of this invention begins in the preoperative preparation stage. First, in order to construct a comprehensive digital patient model, a set of multimodal imaging data needs to be acquired, specifically including: high-resolution enhanced computed tomography (CT) images for accurately depicting the geometry of the vascular tree, high soft tissue contrast magnetic resonance imaging (MRI) images for clearly defining tumor boundaries, magnetic resonance elastography (MRE) data for quantitatively measuring tissue stiffness, and four-dimensional dynamic CT (4D-CT) images for capturing respiratory dynamics.

[0051] Subsequently, the core step of model construction begins. First, the system rigidly registers the aforementioned multimodal image data, unifying them to the same coordinate system. Next, using a deep learning segmentation network based on the U-Net architecture, the fused images are automatically processed, synchronously and with high precision, to reconstruct a high-fidelity 3D geometric model of the liver parenchyma, internal vascular tree, and tumor lesions. Based on this geometric model, the system further generates a spatially resolution-adaptive tetrahedral finite element volume mesh to decompose the continuous liver geometry into a digital model composed of countless tiny units suitable for physical simulation calculations. Simultaneously, shear modulus distribution maps characterizing spatial differences in tissue stiffness are extracted from the registered MRE data, and these heterogeneous biophysical property parameters are precisely assigned to each element of the finite element mesh in a node-to-node manner, thereby constructing a static biomechanical model integrating precise geometry and heterogeneous physical properties. Finally, to endow the model with dynamic characteristics, the system analyzes 4D-CT image sequences and uses statistical methods such as principal component analysis (PCA) to extract key dynamic patterns that describe the patient's individualized respiratory movement patterns, and constructs a dynamic motion model based on these patterns. By mathematically coupling this dynamic motion model with the aforementioned static biomechanical model, a personalized four-dimensional biomechanical model capable of high-fidelity physical simulation during surgery is finally constructed. The "four dimensions" in the personalized four-dimensional biomechanical model include spatial dimensions (X, Y, Z) and temporal dimensions (T). The spatial dimension includes high-fidelity three-dimensional geometry (including the precise shape, size, and spatial position of the liver, tumor, and internal vascular system) acquired through CT and MRI, accurately reflecting the patient's individual anatomical characteristics, and integrates heterogeneous biophysical properties (such as spatially varying tissue stiffness) acquired through functional imaging such as MRE, reflecting tissue differences. The temporal dimension does not refer to arbitrary time changes, but specifically to the dynamic process captured and modeled through time-series images such as 4D-CT, accurately characterizing the patient's individualized respiratory movement patterns.

[0052] During the surgical procedure, intraoperative imaging equipment (such as digital subtraction angiography or ultrasound) begins to capture a two-dimensional image stream in real time (i.e., a real-time data sequence composed of continuous two-dimensional image frames). Simultaneously, a spatial positioning system connected to the surgical instruments continuously acquires the three-dimensional spatial position of their ends. The system uses image recognition algorithms to stably track predefined anatomical feature points (such as vascular bifurcation points) in the real-time two-dimensional images. The positions of these real-time tracked two-dimensional feature points and the three-dimensional instrument positions, after strict time synchronization processing, together constitute multi-source dynamic constraints.

[0053] Next, these dynamic constraints were used to drive the physical simulation of a personalized four-dimensional biomechanical model constructed preoperatively. A numerical solver, based on these real-time constraints and the constitutive relations of biological tissues embedded in the model, quickly calculated the overall non-rigid deformation of the liver caused by the interaction between respiration and instruments. The result of this calculation is a three-dimensional non-rigid deformation field containing displacement information of all nodes in the model.

[0054] Once the real-time deformation field is obtained, the system immediately uses it to drive the deformation and reconstruction of the preoperative model, generating a dynamic physical navigation model that is completely consistent with the current real anatomical morphology. This navigation model not only updates the spatial position of all anatomical structures in real time, but more importantly, it also synchronously calculates and attaches the stress and strain distribution information caused by deformation inside the model.

[0055] Finally, dynamic path planning is performed on this information-rich dynamic physical navigation model. The system defines a multidimensional cost function that comprehensively evaluates the length of candidate paths, the safe distance from blood vessels, and the strain energy accumulated along the path, representing the potential risk of tissue damage. Guided by this cost function, a stochastic path optimization algorithm quickly finds and plans an optimal surgical path with the lowest overall risk in real time. This path is transformed into a three-dimensional, visualized safety corridor, whose color or thickness dynamically changes according to the risk level of each path segment. Finally, using augmented reality (AR) equipment, this dynamic safety corridor image is overlaid onto the surgeon's field of vision, providing intuitive, accurate, and risk-alert dynamic navigation. The entire process is continuously looped during the operation to achieve real-time response to dynamic changes.

[0056] Example 2

[0057] Reference Figure 2 This embodiment, based on the overall process described in Embodiment 1, elaborates in more detail the construction process of the core personalized four-dimensional biomechanical model, which serves as the foundation for all subsequent calculations, and the online adaptive calibration mechanism of this model during surgery. The key feature is that the step of constructing a personalized four-dimensional biomechanical model integrating anatomical geometry and biophysical properties in step 1 is further specified as follows: spatial registration and fusion of multimodal image data acquired before surgery are performed to generate a fused image dataset in a unified coordinate system. Subsequently, an automated image segmentation method is used to simultaneously identify and reconstruct multiple targets, such as liver parenchyma, tumor lesions, and internal multi-level vascular systems, within the fused image dataset, thereby generating a high-precision geometric model representing the anatomical morphology.

[0058] Within the liver volume defined by a high-precision geometric model, a finite element volumetric mesh with adaptive spatial resolution is generated. The element density of the finite element volumetric mesh is locally adjusted according to the complexity of the anatomical structure. At the same time, biomechanical property parameters representing the spatial variation of liver tissue heterogeneity are extracted and calculated from the registered image data. These biomechanical property parameters are then assigned to the finite element volumetric mesh, thereby constructing a static biomechanical model that integrates precise geometric morphology and heterogeneous physical properties.

[0059] To endow the static biomechanical model with dynamic characteristics, a four-dimensional image sequence covering the patient's complete respiratory cycle was used to capture and quantify the continuous three-dimensional deformation of the liver during respiration. By statistically modeling this continuous deformation, its main motion patterns were extracted and a dynamic model that can characterize individualized respiratory motion patterns was constructed. This dynamic model was then coupled with the static biomechanical model to finally form a personalized four-dimensional biomechanical model.

[0060] The method for constructing a personalized four-dimensional biomechanical model further includes a step of online adaptive calibration of the model during surgery. Specifically, during surgery, the image positions of anatomical feature points predicted by the personalized four-dimensional biomechanical model and the corresponding image positions of anatomical feature points observed by real-time intraoperative images are continuously acquired, and the two sets of position data are compared to generate a deviation signal.

[0061] When the deviation signal meets the preset triggering conditions, the system initiates a reverse optimization of the model parameters. This optimization process aims to adjust the spatially varying biomechanical property parameters in the static biomechanical model to reduce the inconsistency between the model's predicted location and the actual observed location. The output of this optimization process is used to update the biomechanical property parameters in the personalized four-dimensional biomechanical model.

[0062] The process begins with offline model construction during preoperative preparation. This begins with the precise fusion and segmentation of multimodal images. Specifically, the patient's high-resolution multi-temporal enhanced CT sequences and high soft tissue contrast T2-weighted magnetic resonance imaging (T2WI-MRI) sequences are aligned into a unified coordinate system using a rigid registration algorithm based on mutual information. The core advantage of the mutual information-based registration method is that it does not rely on the consistency of specific gray values ​​between the images, but rather achieves precise alignment by maximizing the statistical dependence of the gray value distributions of the two images. Therefore, it is particularly suitable for processing multimodal images with significant differences in imaging principles and appearance.

[0063] Subsequently, a multi-stage 3D U-Net deep learning segmentation network was used to process the fused image dataset. The multi-stage network architecture, by assigning tasks such as global coarse localization and local fine segmentation to different processing stages, employs a coarse-to-fine strategy to improve the accuracy, stability, and adaptability of segmenting complex anatomical structures in large-scale 3D images. Through this network, the complete 3D geometry of the liver parenchyma, tumor lesions, and the intrahepatic artery, portal vein, and hepatic vein vascular trees can be segmented simultaneously with extremely high accuracy. Simultaneously, from the magnetic resonance elastography (MRE) data obtained in the same examination, a specialized inversion algorithm was used to calculate a 3D shear modulus distribution map that can finely characterize tissue stiffness differences. This shear modulus map was precisely registered with the segmented geometry and used as a heterogeneous biophysical property, assigned node-to-node to the subsequently generated finite element model. Then, an advanced mesh generation technique was used within the model to generate an adaptive tetrahedral finite element mesh. This mesh was automatically refined in areas such as the vessel walls and tumor edges based on the complexity of the geometry to ensure the computational accuracy of subsequent physical simulations. To give the model dynamic characteristics, a set of four-dimensional dynamic CT (4D-CT) images covering at least one complete respiratory cycle was acquired before the operation of the patient. By performing non-rigid registration on each frame of the 4D-CT sequence and using the statistical method of principal component analysis (PCA), several major deformation patterns that can characterize the patient's individualized respiratory motion were extracted.

[0064] Finally, this dynamic motion model is mathematically coupled with the previously constructed static biomechanical model rich in heterogeneous physical properties, forming a high-fidelity, personalized four-dimensional biomechanical model that can accurately simulate the complex morphology, non-uniform physical response, and unique respiratory dynamics of the patient's liver in the preoperative stage. Furthermore, to support adaptive online calibration, the system defines an objective function that aims to minimize the root mean square error between the model-predicted feature point image locations and the actual observed locations. When the error exceeds a preset threshold, the system initiates a gradient descent-based optimization algorithm to inversely adjust the biomechanical property parameters in the model to minimize the objective function.

[0065] Example 3

[0066] Reference Figure 1Steps 2-4 in the above are an embodiment of the present invention. This embodiment is based on the personalized four-dimensional biomechanical model constructed and calibrated in embodiment 2, and elaborates in detail the core dynamic feedback loop of embodiment 1, which realizes real-time data acquisition, physical simulation calculation, navigation model generation and interactive response during the operation. The key feature is that the step of identifying and tracking anatomical feature points and simultaneously acquiring the spatial position of surgical instruments during the operation is further specified as follows: After acquiring the patient's two-dimensional intraoperative image stream in real time, the predefined anatomical landmarks in the personalized four-dimensional biomechanical model are mapped to the two-dimensional image space to dynamically define the corresponding prediction search area for the anatomical landmarks. Then, within the prediction search area, matching and identification are performed based on the pre-stored visual appearance features of the anatomical landmarks to complete the positioning. For continuous image frames, a temporal motion model is established to predict the subsequent position of the feature points, so as to use the predicted subsequent position to correct and maintain the tracking process.

[0067] Meanwhile, in step 2, the step of obtaining the real-time three-dimensional spatial position of the surgical instrument end via a spatial positioning system specifically involves continuously calculating the real-time spatial pose of the instrument in the navigation coordinate system. In order to ensure the spatiotemporal consistency between image data and instrument pose data, real-time information from different data sources is processed for time synchronization and correlation. Finally, these spatiotemporally aligned heterogeneous data are integrated into a multi-source dynamic constraint set for driving physical simulation.

[0068] Step 3, which involves jointly driving the personalized four-dimensional biomechanical model to perform physical simulation and calculation in order to calculate the three-dimensional non-rigid deformation field in real time, is further specified as follows: the multi-source dynamic constraint set is analyzed, and based on this, boundary conditions and internal constraints are set on the personalized four-dimensional biomechanical model to limit its current state.

[0069] The numerical solver is used to calculate the physical response of the biomechanical model under the combined action of boundary conditions and internal constraints. The calculation follows the preset constitutive relation of biological tissue. The calculation process aims to find the unique physical equilibrium state that simultaneously satisfies the boundary conditions, internal constraints, and constitutive relation of biological tissue, and thereby determine the three-dimensional displacement vector of the discrete unit node in the model, thus forming a three-dimensional non-rigid deformation field.

[0070] Step 4, which involves reconstructing the preoperative model in three dimensions based on the real-time calculated non-rigid deformation field to generate a dynamic physical navigation model, is further specified as follows: The three-dimensional non-rigid deformation field of the liver is used as a spatial transformation function and applied to the reference geometry of the personalized four-dimensional biomechanical model. This application process updates the spatial coordinates of all discrete unit nodes by applying their corresponding three-dimensional displacement vectors in the deformation field to the discrete unit nodes in the model, thereby reconstructing a transient geometric configuration that can accurately reflect the current macroscopic shape and position of the liver.

[0071] Based on the transient geometric configuration, the internal physical state changes caused by tissue deformation are calculated by further utilizing the three-dimensional non-rigid deformation field of the liver and the pre-set constitutive relations in the personalized four-dimensional biomechanical model. This calculation is performed for each discrete element in the model to solve the stress tensor and strain tensor of the model under the current deformation state.

[0072] The spatial coordinate information of the transient geometric configuration is fused with the stress tensor and strain tensor calculated for all discrete elements in the model. This fusion process integrates the spatial coordinate information with the stress tensor and strain tensor into a unified data entity, thereby forming a dynamic physical navigation model.

[0073] The calculation process using a numerical solver further includes an event-driven mechanism, which specifically involves the system continuously monitoring and analyzing the real-time spatial pose of the surgical instruments obtained by the spatial positioning system. When the analysis determines that the change characteristics of the spatial pose conform to a preset event model that represents the surgeon's active intervention, the system triggers a recalculation of the physical simulation.

[0074] The recalculation process involves using the spatial pose of the surgical instruments at the end of the active intervention event as the boundary condition, and resolving and updating the three-dimensional non-rigid deformation field of the liver according to the calculation process of claim 4. The updated three-dimensional non-rigid deformation field of the liver is then used to update the dynamic physical navigation model and the optimal surgical path planned on the model.

[0075] Once the surgery begins, the system's real-time data acquisition and processing module is activated. This module synchronously performs the following tasks to construct multi-source dynamic constraints for driving the physical simulation:

[0076] First, anatomical feature points are identified and continuously tracked. In this embodiment, the intraoperative imaging equipment is a C-arm digital subtraction angiography (DSA) system, which continuously acquires a two-dimensional fluoroscopic image stream containing rich vascular information at a specific frame rate (e.g., 15 frames / second). Using the three-dimensional model constructed in Embodiment 2, the system projects pre-selected anatomically significant feature points (such as specific vascular bifurcation points) onto the current two-dimensional image plane to dynamically define the prediction search area. Subsequently, a deep learning tracking algorithm based on a Siamese Network runs within this area, performing similarity matching between real-time image blocks and reference templates of feature points to achieve precise localization and stable tracking of the feature points.

[0077] Secondly, the spatial pose of the surgical instrument is acquired simultaneously. An array of optical markers fixedly attached to the surgical instrument (e.g., an 18G puncture needle) is positioned in real time by a high-precision optical spatial positioning system (e.g., the NDI Polaris Vega system based on binocular infrared tracking technology) at a higher frequency (e.g., 60 Hz). By calculating the six-degree-of-freedom spatial pose of the marker array in the navigation coordinate system, this system can accurately and continuously obtain the three-dimensional spatial position of the instrument needle tip.

[0078] Finally, spatiotemporal synchronization and correlation of heterogeneous data are achieved. All feature point location data from the DSA image stream and instrument pose data from the optical tracking system are assigned synchronization timestamps by a unified high-precision Network Time Protocol (NTP) server. A data fusion unit performs rigorous temporal alignment and correlation processing on these heterogeneous data streams based on these timestamps, ultimately integrating them into a spatiotemporally strictly aligned "multi-source dynamic constraint set" as the real-time input of the physics simulation engine.

[0079] Upon receiving the "multi-source dynamic constraint set," the physics simulation engine resolves it into boundary conditions and internal constraints applied to the biomechanical model. Subsequently, a numerical solver based on the nonlinear finite element method (FEM), following predefined hyperelastic constitutive relations (e.g., the Ogden model), rapidly calculates a three-dimensional non-rigid deformation field characterizing the liver's current overall deformation state. This deformation field is immediately used as a spatial transformation function to reconstruct the model's reference geometry and simultaneously calculates the internal stress and strain tensors caused by the deformation. Finally, a dynamic physical navigation model integrating real-time geometry and internal physical state is generated and pushed to subsequent modules.

[0080] Furthermore, to provide instantaneous response to critical operator actions, the system incorporates an event-driven mechanism. This mechanism continuously monitors the spatial pose of surgical instruments, and when it identifies drastic movements that conform to a preset "operator-initiated intervention" model, it immediately triggers a high-priority recalculation. This recalculation uses the intervened instrument pose as a strong constraint, thereby rapidly updating the deformation field and navigation model, ensuring the system's real-time dynamic response capability.

[0081] Example 4

[0082] Reference Figure 3 This embodiment, based on the dynamic physical navigation model generated in the aforementioned embodiments, elaborates on the path planning and visualization steps of Embodiment 1 on this model to reveal the specific implementation of its cognitive assistance and intelligent decision-making layer. The key feature is that step 5, which involves path planning and visualization on the dynamic physical navigation model, is further specified as follows: for any candidate path in the dynamic physical navigation model, a multi-dimensional dynamic cost function is constructed. This construction process includes quantifying the geometric properties of the path, its safety relationship with key anatomical structures, and the biomechanical risk of tissue damage, and integrating these quantified risk dimensions into a comprehensive cost value according to a preset fusion rule.

[0083] The method employs a stochastic path optimization approach to conduct iterative path exploration within the space defined by a dynamic physical navigation model. Its key feature is that the exploration process is guided by comprehensive cost value, prioritizing expansion in low-risk areas. Furthermore, the method's exploration strategy is adaptive, dynamically adjusting its exploration behavior based on the local risk level of the area to balance global search efficiency with local obstacle avoidance accuracy.

[0084] Furthermore, the optimal surgical path obtained through iterative path exploration will generate perceptible safety guidance information for augmented reality visualization. The perceptible attributes of the safety guidance information will be dynamically modulated according to the comprehensive cost value corresponding to each point along the path to warn of changes in risk in the path. Finally, the safety guidance information will be spatially registered with the navigation coordinate system and fused and overlaid with the surgeon's perceived scene for display.

[0085] The integration steps based on the preset fusion rules are further specified as follows: the quantified geometric attributes, the safety relationship with key anatomical structures, and the biomechanical-based tissue damage risk are collectively defined as a risk dimension set;

[0086] The fusion rule is a dynamic weighting mechanism used to assign weight coefficients to each dimension in the risk dimension set. The process of determining these weight coefficients is based on the contextual analysis of the local area where the candidate path is located in the dynamic physical navigation model. This contextual analysis aims to assess the risk level of the anatomical environment around the candidate path, and adaptively adjusts the weight coefficients assigned to each dimension in the risk dimension set according to the assessed risk level, so that the relative importance of each risk dimension can change dynamically in different anatomical environments.

[0087] The augmented reality visualization process for the optimal surgical path further includes: constructing a dynamically updated three-dimensional risk field model from the dynamic physical navigation model, which represents the distribution of comprehensive risk value in the three-dimensional space around the optimal surgical path; presenting the three-dimensional risk field model in the form of a three-dimensional visualization object, the perceptible attributes of which correspond to the risk level in the space; and the display range and level of detail of the three-dimensional visualization object can be interactively adjusted by the surgeon; combining the guidance information of the optimal surgical path with the adjusted three-dimensional visualization object; and merging and overlaying the combined information with the surgeon's perceived scene.

[0088] When the path planning function is activated, the system constructs and solves a multidimensional dynamic cost function in real time on the dynamic physical navigation model to comprehensively quantify the risk of candidate paths from the puncture point on the body surface to the tumor target. This function integrates at least three core risk dimensions: geometric properties (the Euclidean length of the path), safety relationship (the minimum spatial distance between the path and the vascular system), and biomechanical damage risk (the integral of the tissue strain energy density accumulated along the path). The safety relationship is calculated by querying a pre-calculated distance field and mapping it to a risk value via an exponential function, thereby creating a strong repulsive field around the blood vessel. The biomechanical damage risk is directly related to the potential microscopic damage to the tissue caused by the puncture.

[0089] For any candidate path The multidimensional dynamic cost function It can be specifically defined as a line integral along the path:

[0090]

[0091] in, For path The point on, Let be a path element. This is the normalized path length cost, which is typically 1. For safety reasons, the cost, for example, can be defined as the distance to the nearest blood vessel wall. A higher-order inverse proportional function, such as ; The cost of biomechanical damage is equal to The strain energy density of the tissue at the point. , and These are the weighting coefficients for each item.

[0092] The security relationship cost function middle: To prevent small constants with a denominator of zero; Representing path points The Euclidean distance to the wall of the nearest critical vascular system is obtained in real time by the system by querying a pre-calculated distance field. It is an exponent greater than or equal to 1, used to adjust the strength and range of the repulsive field. At that time, the cost is inversely proportional to the distance; when (For example, When the distance decreases, the cost increases sharply and non-linearly, thus forming a virtual barrier around the blood vessel to achieve a more conservative and safer obstacle avoidance strategy. This parameter can be preset and adjusted according to the surgical risk level.

[0093] The risk values ​​of these three dimensions are intelligently fused through a dynamic adaptive weighting mechanism, rather than a simple fixed weighting. This mechanism performs real-time contextual analysis of the local anatomical environment in which the candidate path is located, assessing the vascular density and average strain energy in its neighborhood. When the analysis result indicates a "safe zone," the fusion rule automatically increases the weight of the geometric attribute dimension to pursue efficiency; conversely, when entering a "danger zone," the weight of safety relationships and biomechanical damage risk is significantly increased to ensure absolute safety and minimal tissue damage.

[0094] Specifically, this dynamic weighting mechanism is achieved by analyzing the anatomical environment within a local neighborhood around a waypoint. For example, the system calculates the density of blood vessels within this neighborhood and associates the weight of safety factors with it, so that the denser the blood vessels, the higher the weight. Simultaneously, the system also calculates the average tissue strain energy within this neighborhood and associates the weight of biomechanical damage risk with it, so that the higher the strain energy, the higher the weight. This approach ensures that when the path is about to enter a dangerous area, the weight of the corresponding risk factor can be automatically and predictively increased, thereby guiding the path planning algorithm to proactively avoid danger.

[0095] Subsequently, an improved stochastic path optimization algorithm, such as RRT* (Asymptotically Optimal Fast Exploratory Random Tree), guided by this dynamically changing comprehensive cost value, performs efficient iterative path exploration within the navigation model space. Specifically, the exploration and optimization process of the RRT* algorithm is as follows: The algorithm first randomly samples a target point within the 3D space defined by the navigation model. Then, in the already generated random tree structure, it searches for the existing node closest to the randomly sampled point. Next, the algorithm extends from the existing node towards the randomly sampled point by a preset step size, thereby generating a new node. For path segments connecting the existing node and the new node, the system performs collision detection and calculates its cost value based on the aforementioned cost function. If there is no collision, the new node is added to the random tree structure; furthermore, the algorithm checks other nodes within a certain neighborhood of the new node and attempts to perform a crucial "reconnection" operation. This operation aims to find whether there is a lower-cost path connected through this new node to optimize the neighboring connection relationship, thereby continuously improving the path quality of the entire tree structure. This iterative process continues until a surgical path is found that connects the surgical starting point and the target point, and whose overall cost is optimal.

[0096] The algorithm's exploration step size is also adaptive. It uses a large step size for rapid exploration in low-risk areas and automatically switches to a small step size for fine obstacle avoidance in high-risk areas, thereby quickly planning an optimal surgical path that achieves the best balance between global efficiency and local safety.

[0097] The visualization of the optimal surgical path utilizes augmented reality (AR) technology, constructing a multi-layered, interactive risk perception interface. The system first renders the optimal path as a three-dimensional "safety corridor" for basic directional guidance. Furthermore, the system constructs a three-dimensional risk field model, synchronously updated with the liver model, from the real-time calculated "tissue strain energy" field in the navigation model, and visualizes it as an interactive three-dimensional heatmap. The heatmap's color spectrum (e.g., from cool to warm tones) visually marks "vulnerable areas" caused by internal stress in three-dimensional space. The surgeon can use AR devices to dynamically adjust the heatmap's display threshold to highlight high-risk areas through gestures or voice interaction, or rotate the model to examine the relationship between the path and the risk field from any angle. Finally, this basic "safety corridor" and the interactive "risk heatmap" are precisely combined, and this composite, decision-information-rich virtual image is accurately overlaid onto the surgeon's actual surgical field of view through the AR display system with correct perspective. This visualization solution not only indicates "which path to take," but also provides a comprehensive three-dimensional risk situational awareness in a predictive way, showing "how safe the area around the path is" and "how dangerous it is to deviate from the path."

[0098] In summary, this invention significantly improves the targeting accuracy of punctures by employing a dynamically driven four-dimensional biomechanical model to precisely compensate for the real-time non-rigid deformation of the liver caused by respiratory movements and instrument interactions during surgery. Furthermore, this invention achieves a qualitative leap in path planning by incorporating biomechanical damage risk into the core optimization dimension. This ensures that the planned path not only safely avoids blood vessels but also minimizes tissue trauma at the physical level, which is of significant value in reducing the risk of postoperative complications. Finally, this invention uses augmented reality technology to intuitively present the dynamic optimal path and three-dimensional risk field to the surgeon, significantly reducing the difficulty of surgical procedures and cognitive load, thereby effectively improving surgical efficiency and safety.

[0099] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A dynamic planning system for interventional surgical pathways in liver cancer based on multimodal image fusion, characterized in that, Includes the following modules: The multimodal image modeling module is configured to spatially register and fuse multimodal image data acquired before surgery, extract anatomical structure information and tissue biophysical property information of the liver and vascular system, and construct a static biomechanical model; based on dynamic image data covering the patient's respiratory cycle, it acquires continuous three-dimensional deformation information of the liver during the respiratory process, extracts liver motion features based on the continuous three-dimensional deformation information, and couples the liver motion features with the static biomechanical model to form a personalized four-dimensional biomechanical model; The intraoperative information acquisition module is configured to acquire two-dimensional intraoperative images of the patient in real time during the operation, identify and continuously track anatomical feature points from them, and acquire the real-time three-dimensional spatial position of the surgical instrument tip through the spatial positioning system. The deformation field calculation module is configured to use the two-dimensional real-time position of anatomical feature points and the three-dimensional real-time position of surgical instruments as multi-source dynamic constraints to jointly drive the personalized four-dimensional biomechanical model to perform physical simulation calculations, so as to calculate the three-dimensional non-rigid deformation field of the liver caused by respiratory movement and instrument interaction in real time. The dynamic navigation model generation module is configured to use the three-dimensional non-rigid deformation field of the liver to reconstruct the preoperative model in three dimensions, and generate a dynamic physical navigation model that accurately reflects the true shape, position and internal stress distribution of the current anatomical structure. The path planning module is configured to define a multidimensional dynamic cost function on the dynamic physical navigation model, which weights and fuses the path geometric length, the safety margin of key vessels, and the tissue strain energy accumulated along the path; and to use a stochastic path optimization method to iteratively optimize under the guidance of the multidimensional dynamic cost function, and to plan an optimal surgical path from the body surface to the tumor target in real time that can minimize the risk of tissue damage caused by puncture, and to perform augmented reality visualization of the optimal surgical path.

2. The dynamic planning system for interventional surgical pathways for liver cancer based on multimodal image fusion according to claim 1, characterized in that, The multimodal image modeling module is further configured to: spatially register and fuse the multimodal image data acquired by the patient before surgery, thereby generating a fused image dataset in a unified coordinate system; then, using an automated image segmentation method, simultaneously identify and three-dimensionally reconstruct multiple targets such as liver parenchyma, tumor lesions, and internal multi-level vascular system in the fused image dataset, thereby generating a high-precision geometric model representing the anatomical morphology. Within the liver volume defined by the high-precision geometric model, a finite element volume mesh with adaptive spatial resolution is generated. The element density of the finite element volume mesh is locally adjusted according to the complexity of the anatomical structure. At the same time, biomechanical property parameters representing the spatial variation of liver tissue heterogeneity are extracted and calculated from the registered image data. The biomechanical property parameters are assigned to the finite element volume mesh, thereby constructing a static biomechanical model that integrates precise geometric morphology and heterogeneous physical properties. To endow the static biomechanical model with dynamic characteristics, a four-dimensional image sequence covering the patient's complete respiratory cycle is used to capture and quantify the continuous three-dimensional deformation of the liver during respiration. By statistically modeling this continuous deformation, its main motion patterns are extracted and a dynamic model that can characterize individualized respiratory motion patterns is constructed. This dynamic model is coupled with the static biomechanical model to finally form the personalized four-dimensional biomechanical model.

3. The dynamic planning system for interventional surgical pathways for liver cancer based on multimodal image fusion as described in claim 1, characterized in that, The intraoperative information acquisition module is further configured to: after acquiring the patient's two-dimensional intraoperative image stream in real time, dynamically define the corresponding prediction search area for the anatomical markers by mapping the predefined anatomical markers in the personalized four-dimensional biomechanical model to the two-dimensional image space; and then, within the prediction search area, match and identify the anatomical markers based on their pre-stored visual appearance features to complete the localization; for continuous image frames, a temporal motion model is established to predict the subsequent positions of the feature points, so as to use the predicted subsequent positions to correct and maintain the tracking process. The real-time spatial pose of surgical instruments in the navigation coordinate system is continuously acquired by the spatial positioning system. The image data corresponding to the two-dimensional intraoperative image is time-synchronized and correlated with the real-time spatial pose. The synchronized anatomical feature point position information and the surgical instrument spatial pose information are used to form a multi-source dynamic constraint for driving physical simulation.

4. The dynamic planning system for interventional surgical pathways for liver cancer based on multimodal image fusion as described in claim 1, characterized in that, The deformation field solution module is further configured to: analyze the multi-source dynamic constraints, and based on this, set boundary conditions and internal constraints on the personalized four-dimensional biomechanical model to limit its current state; The physical response of the personalized four-dimensional biomechanical model under the combined action of the boundary conditions and internal constraints is calculated using a numerical solver. The calculation follows a preset biological tissue constitutive relation. The calculation process aims to solve for the physical equilibrium state that simultaneously satisfies the boundary conditions, internal constraints, and biological tissue constitutive relation, and thereby determine the three-dimensional displacement vector of the discrete unit nodes in the model, thus forming the three-dimensional non-rigid deformation field.

5. The dynamic planning system for interventional surgical pathways for liver cancer based on multimodal image fusion as described in claim 1, characterized in that, The dynamic navigation model generation module is further configured to: use the three-dimensional non-rigid deformation field of the liver as a spatial transformation function and apply it to the reference geometry of the personalized four-dimensional biomechanical model. This application process updates the spatial coordinates of all discrete unit nodes by applying their corresponding three-dimensional displacement vectors in the deformation field to the discrete unit nodes in the model, thereby reconstructing the transient geometric configuration that characterizes the current macroscopic morphology and position of the liver. Based on the transient geometric configuration, the internal physical state changes caused by tissue deformation are calculated by further utilizing the three-dimensional non-rigid deformation field of the liver and the pre-set constitutive relation in the personalized four-dimensional biomechanical model. This calculation is performed for each discrete element in the model to solve the stress tensor and strain tensor of the model under the current deformation state. The spatial coordinate information of the transient geometric configuration is fused with the stress tensor and strain tensor calculated for all discrete elements in the model. This fusion process integrates the spatial coordinate information and the stress tensor and strain tensor into a unified data entity, thereby forming the dynamic physical navigation model.

6. The dynamic planning system for interventional surgical pathways for liver cancer based on multimodal image fusion as described in claim 1, characterized in that, The path planning module is further configured to: construct a multidimensional dynamic cost function for any candidate path in the dynamic physical navigation model. The construction process includes quantifying the geometric properties of the path, the safety relationship with key anatomical structures, and the biomechanical risk of tissue damage, and integrating these quantified risk dimensions into a comprehensive cost value according to a preset fusion rule. Based on the comprehensive cost value, iterative path exploration is performed in the space corresponding to the dynamic physical navigation model, and the path exploration range is adjusted according to the local risk level of the candidate path area to obtain the target surgical path that meets the preset path screening conditions. Furthermore, the target surgical path generates safety guidance information for augmented reality visualization. The perceptible attributes of the safety guidance information are dynamically modulated according to the comprehensive cost value corresponding to each position along the target surgical path to warn of changes in risk in the path. Finally, the safety guidance information is spatially registered with the navigation coordinate system and fused and overlaid with the surgeon's perceived scene for display.

7. The dynamic planning system for interventional surgical pathways for liver cancer based on multimodal image fusion as described in claim 6, characterized in that, The path planning module is configured to determine the weight coefficients corresponding to each risk dimension according to a dynamic weighting mechanism: the quantified geometric attributes, the safety relationship with key anatomical structures, and the biomechanical-based tissue damage risk are collectively defined as a set of risk dimensions. The fusion rule is a dynamic weighting mechanism used to assign weight coefficients to each dimension in the risk dimension set. The process of determining the weight coefficient is based on the context analysis of the local area where the candidate path is located in the dynamic physical navigation model. This context analysis aims to assess the risk level of the anatomical environment around the candidate path, and adaptively adjust the weight coefficients assigned to each dimension in the risk dimension set according to the assessed risk level, so that the relative importance of each risk dimension can change dynamically in different anatomical environments.

8. The dynamic planning system for interventional surgical pathways for liver cancer based on multimodal image fusion as described in claim 4, characterized in that, The deformation field calculation module is also configured to: continuously monitor and analyze the real-time spatial pose of the surgical instruments obtained by the spatial positioning system, and trigger physical simulation recalculation when the change characteristics of the real-time spatial pose meet the preset active intervention event conditions; During the recalculation process, the spatial pose of the surgical instruments at the end of the active intervention event is used as the boundary condition, and the physical simulation is recalculated based on the boundary condition and the personalized four-dimensional biomechanical model to update the three-dimensional non-rigid deformation field of the liver. The updated three-dimensional non-rigid deformation field of the liver is then used to update the dynamic physical navigation model and the target surgical path planned on the model.

9. The dynamic planning system for interventional surgical pathways for liver cancer based on multimodal image fusion as described in claim 6, characterized in that, The path planning module is further configured to: construct a dynamically updated three-dimensional risk field model from the dynamic physical navigation model, representing the distribution of the comprehensive risk value in the three-dimensional space around the optimal surgical path; present the three-dimensional risk field model in the form of a three-dimensional visualization object, wherein the perceptible attributes of the visualization object correspond to the risk level in the space; and the display range and level of detail of the three-dimensional visualization object can be interactively adjusted by the surgeon; combine the guidance information of the target surgical path with the adjusted three-dimensional visualization object; and fuse and overlay the combined information with the surgeon's perceived scene for display.

10. The dynamic planning system for interventional surgical pathways for liver cancer based on multimodal image fusion as described in claim 2, characterized in that, The multimodal image modeling module is also configured to: continuously acquire the image positions of anatomical feature points predicted by the personalized four-dimensional biomechanical model and the corresponding anatomical feature point image positions observed by real-time intraoperative images during the operation, and compare the two sets of position data to generate a deviation signal. When the deviation signal meets the preset triggering conditions, the system initiates a reverse optimization of the model parameters. This optimization process aims to adjust the spatially varying biomechanical property parameters in the static biomechanical model to reduce the inconsistency between the model's predicted location and the actual observed location. The output of this optimization process is used to update the biomechanical property parameters in the personalized four-dimensional biomechanical model.

Images